1201 lines
45 KiB
TypeScript
Raw Normal View History

import fs from "node:fs/promises";
import os from "node:os";
import type { AgentMessage } from "@mariozechner/pi-agent-core";
import {
createAgentSession,
DefaultResourceLoader,
estimateTokens,
SessionManager,
} from "@mariozechner/pi-coding-agent";
import { resolveSignalReactionLevel } from "../../../extensions/signal/src/reaction-level.js";
import { resolveTelegramInlineButtonsScope } from "../../../extensions/telegram/src/inline-buttons.js";
import { resolveTelegramReactionLevel } from "../../../extensions/telegram/src/reaction-level.js";
2026-02-18 01:29:02 +00:00
import { resolveHeartbeatPrompt } from "../../auto-reply/heartbeat.js";
import type { ReasoningLevel, ThinkLevel } from "../../auto-reply/thinking.js";
2026-02-18 01:29:02 +00:00
import { resolveChannelCapabilities } from "../../config/channel-capabilities.js";
import type { OpenClawConfig } from "../../config/config.js";
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
import {
ensureContextEnginesInitialized,
resolveContextEngine,
} from "../../context-engine/index.js";
import { createInternalHookEvent, triggerInternalHook } from "../../hooks/internal-hooks.js";
2026-01-14 01:08:15 +00:00
import { getMachineDisplayName } from "../../infra/machine-name.js";
import { generateSecureToken } from "../../infra/secure-random.js";
import { getMemorySearchManager } from "../../memory/index.js";
Plugin API: compaction/reset hooks, bootstrap file globs, memory plugin status (#13287) * feat: add before_compaction and before_reset plugin hooks with session context - Pass session messages to before_compaction hook - Add before_reset plugin hook for /new and /reset commands - Add sessionId to plugin hook agent context * feat: extraBootstrapFiles config with glob pattern support Add extraBootstrapFiles to agent defaults config, allowing glob patterns (e.g. "projects/*/TOOLS.md") to auto-load project-level bootstrap files into agent context every turn. Missing files silently skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(status): show custom memory plugins as enabled, not unavailable The status command probes memory availability using the built-in memory-core manager. Custom memory plugins (e.g. via plugin slot) can't be probed this way, so they incorrectly showed "unavailable". Now they show "enabled (plugin X)" without the misleading label. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: use async fs.glob and capture pre-compaction messages - Replace globSync (node:fs) with fs.glob (node:fs/promises) to match codebase conventions for async file operations - Capture session.messages BEFORE replaceMessages(limited) so before_compaction hook receives the full conversation history, not the already-truncated list * fix: resolve lint errors from CI (oxlint strict mode) - Add void to fire-and-forget IIFE (no-floating-promises) - Use String() for unknown catch params in template literals - Add curly braces to single-statement if (curly rule) * fix: resolve remaining CI lint errors in workspace.ts - Remove `| string` from WorkspaceBootstrapFileName union (made all typeof members redundant per no-redundant-type-constituents) - Use type assertion for extra bootstrap file names - Drop redundant await on fs.glob() AsyncIterable (await-thenable) * fix: address Greptile review — path traversal guard + fs/promises import - workspace.ts: use path.resolve() + traversal check in loadExtraBootstrapFiles() - commands-core.ts: import fs from node:fs/promises, drop fs.promises prefix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve symlinks before workspace boundary check Greptile correctly identified that symlinks inside the workspace could point to files outside it, bypassing the path prefix check. Now uses fs.realpath() to resolve symlinks before verifying the real path stays within the workspace boundary. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Greptile review — hook reliability and type safety 1. before_compaction: add compactingCount field so plugins know both the full pre-compaction message count and the truncated count being fed to the compaction LLM. Clarify semantics in comment. 2. loadExtraBootstrapFiles: use path.basename() for the name field so "projects/quaid/TOOLS.md" maps to the known "TOOLS.md" type instead of an invalid WorkspaceBootstrapFileName cast. 3. before_reset: fire the hook even when no session file exists. Previously, short sessions without a persisted file would silently skip the hook. Now fires with empty messages array so plugins always know a reset occurred. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate bootstrap filenames and add compaction hook timeout - Only load extra bootstrap files whose basename matches a recognized workspace filename (AGENTS.md, TOOLS.md, etc.), preventing arbitrary files from being injected into agent context. - Wrap before_compaction hook in a 30-second Promise.race timeout so misbehaving plugins cannot stall the compaction pipeline. - Clarify hook comments: before_compaction is intentionally awaited (plugins need messages before they're discarded) but bounded. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: make before_compaction non-blocking, add sessionFile to after_compaction - before_compaction is now true fire-and-forget — no await, no timeout. Plugins that need full conversation data should persist it themselves and return quickly, or use after_compaction for async processing. - after_compaction now includes sessionFile path so plugins can read the full JSONL transcript asynchronously. All pre-compaction messages are preserved on disk, eliminating the need to block compaction. - Removes Promise.race timeout pattern that didn't actually cancel slow hooks (just raced past them while they continued running). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add sessionFile to before_compaction for parallel processing The session JSONL already has all messages on disk before compaction starts. By providing sessionFile in before_compaction, plugins can read and extract data in parallel with the compaction LLM call rather than waiting for after_compaction. This is the optimal path for memory plugins that need the full conversation history. sessionFile is also kept on after_compaction for plugins that only need to act after compaction completes (analytics, cleanup, etc.). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: move bootstrap extras into bundled hook --------- Co-authored-by: Solomon Steadman <solstead@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Clawdbot <clawdbot@alfie.local> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:45:45 +07:00
import { getGlobalHookRunner } from "../../plugins/hook-runner-global.js";
import { prepareProviderRuntimeAuth } from "../../plugins/provider-runtime.js";
import { type enqueueCommand, enqueueCommandInLane } from "../../process/command-queue.js";
Agents: add nested subagent orchestration controls and reduce subagent token waste (#14447) * Agents: add subagent orchestration controls * Agents: add subagent orchestration controls (WIP uncommitted changes) * feat(subagents): add depth-based spawn gating for sub-sub-agents * feat(subagents): tool policy, registry, and announce chain for nested agents * feat(subagents): system prompt, docs, changelog for nested sub-agents * fix(subagents): prevent model fallback override, show model during active runs, and block context overflow fallback Bug 1: When a session has an explicit model override (e.g., gpt/openai-codex), the fallback candidate logic in resolveFallbackCandidates silently appended the global primary model (opus) as a backstop. On reinjection/steer with a transient error, the session could fall back to opus which has a smaller context window and crash. Fix: when storedModelOverride is set, pass fallbacksOverride ?? [] instead of undefined, preventing the implicit primary backstop. Bug 2: Active subagents showed 'model n/a' in /subagents list because resolveModelDisplay only read entry.model/modelProvider (populated after run completes). Fix: fall back to modelOverride/providerOverride fields which are populated at spawn time via sessions.patch. Bug 3: Context overflow errors (prompt too long, context_length_exceeded) could theoretically escape runEmbeddedPiAgent and be treated as failover candidates in runWithModelFallback, causing a switch to a model with a smaller context window. Fix: in runWithModelFallback, detect context overflow errors via isLikelyContextOverflowError and rethrow them immediately instead of trying the next model candidate. * fix(subagents): track spawn depth in session store and fix announce routing for nested agents * Fix compaction status tracking and dedupe overflow compaction triggers * fix(subagents): enforce depth block via session store and implement cascade kill * fix: inject group chat context into system prompt * fix(subagents): always write model to session store at spawn time * Preserve spawnDepth when agent handler rewrites session entry * fix(subagents): suppress announce on steer-restart * fix(subagents): fallback spawned session model to runtime default * fix(subagents): enforce spawn depth when caller key resolves by sessionId * feat(subagents): implement active-first ordering for numeric targets and enhance task display - Added a test to verify that subagents with numeric targets follow an active-first list ordering. - Updated `resolveSubagentTarget` to sort subagent runs based on active status and recent activity. - Enhanced task display in command responses to prevent truncation of long task descriptions. - Introduced new utility functions for compacting task text and managing subagent run states. * fix(subagents): show model for active runs via run record fallback When the spawned model matches the agent's default model, the session store's override fields are intentionally cleared (isDefault: true). The model/modelProvider fields are only populated after the run completes. This left active subagents showing 'model n/a'. Fix: store the resolved model on SubagentRunRecord at registration time, and use it as a fallback in both display paths (subagents tool and /subagents command) when the session store entry has no model info. Changes: - SubagentRunRecord: add optional model field - registerSubagentRun: accept and persist model param - sessions-spawn-tool: pass resolvedModel to registerSubagentRun - subagents-tool: pass run record model as fallback to resolveModelDisplay - commands-subagents: pass run record model as fallback to resolveModelDisplay * feat(chat): implement session key resolution and reset on sidebar navigation - Added functions to resolve the main session key and reset chat state when switching sessions from the sidebar. - Updated the `renderTab` function to handle session key changes when navigating to the chat tab. - Introduced a test to verify that the session resets to "main" when opening chat from the sidebar navigation. * fix: subagent timeout=0 passthrough and fallback prompt duplication Bug 1: runTimeoutSeconds=0 now means 'no timeout' instead of applying 600s default - sessions-spawn-tool: default to undefined (not 0) when neither timeout param is provided; use != null check so explicit 0 passes through to gateway - agent.ts: accept 0 as valid timeout (resolveAgentTimeoutMs already handles 0 → MAX_SAFE_TIMEOUT_MS) Bug 2: model fallback no longer re-injects the original prompt as a duplicate - agent.ts: track fallback attempt index; on retries use a short continuation message instead of the full original prompt since the session file already contains it from the first attempt - Also skip re-sending images on fallback retries (already in session) * feat(subagents): truncate long task descriptions in subagents command output - Introduced a new utility function to format task previews, limiting their length to improve readability. - Updated the command handler to use the new formatting function, ensuring task descriptions are truncated appropriately. - Adjusted related tests to verify that long task descriptions are now truncated in the output. * refactor(subagents): update subagent registry path resolution and improve command output formatting - Replaced direct import of STATE_DIR with a utility function to resolve the state directory dynamically. - Enhanced the formatting of command output for active and recent subagents, adding separators for better readability. - Updated related tests to reflect changes in command output structure. * fix(subagent): default sessions_spawn to no timeout when runTimeoutSeconds omitted The previous fix (75a791106) correctly handled the case where runTimeoutSeconds was explicitly set to 0 ("no timeout"). However, when models omit the parameter entirely (which is common since the schema marks it as optional), runTimeoutSeconds resolved to undefined. undefined flowed through the chain as: sessions_spawn → timeout: undefined (since undefined != null is false) → gateway agent handler → agentCommand opts.timeout: undefined → resolveAgentTimeoutMs({ overrideSeconds: undefined }) → DEFAULT_AGENT_TIMEOUT_SECONDS (600s = 10 minutes) This caused subagents to be killed at exactly 10 minutes even though the user's intent (via TOOLS.md) was for subagents to run without a timeout. Fix: default runTimeoutSeconds to 0 (no timeout) when neither runTimeoutSeconds nor timeoutSeconds is provided by the caller. Subagent spawns are long-running by design and should not inherit the 600s agent-command default timeout. * fix(subagent): accept timeout=0 in agent-via-gateway path (second 600s default) * fix: thread timeout override through getReplyFromConfig dispatch path getReplyFromConfig called resolveAgentTimeoutMs({ cfg }) with no override, always falling back to the config default (600s). Add timeoutOverrideSeconds to GetReplyOptions and pass it through as overrideSeconds so callers of the dispatch chain can specify a custom timeout (0 = no timeout). This complements the existing timeout threading in agentCommand and the cron isolated-agent runner, which already pass overrideSeconds correctly. * feat(model-fallback): normalize OpenAI Codex model references and enhance fallback handling - Added normalization for OpenAI Codex model references, specifically converting "gpt-5.3-codex" to "openai-codex" before execution. - Updated the `resolveFallbackCandidates` function to utilize the new normalization logic. - Enhanced tests to verify the correct behavior of model normalization and fallback mechanisms. - Introduced a new test case to ensure that the normalization process works as expected for various input formats. * feat(tests): add unit tests for steer failure behavior in openclaw-tools - Introduced a new test file to validate the behavior of subagents when steer replacement dispatch fails. - Implemented tests to ensure that the announce behavior is restored correctly and that the suppression reason is cleared as expected. - Enhanced the subagent registry with a new function to clear steer restart suppression. - Updated related components to support the new test scenarios. * fix(subagents): replace stop command with kill in slash commands and documentation - Updated the `/subagents` command to replace `stop` with `kill` for consistency in controlling sub-agent runs. - Modified related documentation to reflect the change in command usage. - Removed legacy timeoutSeconds references from the sessions-spawn-tool schema and tests to streamline timeout handling. - Enhanced tests to ensure correct behavior of the updated commands and their interactions. * feat(tests): add unit tests for readLatestAssistantReply function - Introduced a new test file for the `readLatestAssistantReply` function to validate its behavior with various message scenarios. - Implemented tests to ensure the function correctly retrieves the latest assistant message and handles cases where the latest message has no text. - Mocked the gateway call to simulate different message histories for comprehensive testing. * feat(tests): enhance subagent kill-all cascade tests and announce formatting - Added a new test to verify that the `kill-all` command cascades through ended parents to active descendants in subagents. - Updated the subagent announce formatting tests to reflect changes in message structure, including the replacement of "Findings:" with "Result:" and the addition of new expectations for message content. - Improved the handling of long findings and stats in the announce formatting logic to ensure concise output. - Refactored related functions to enhance clarity and maintainability in the subagent registry and tools. * refactor(subagent): update announce formatting and remove unused constants - Modified the subagent announce formatting to replace "Findings:" with "Result:" and adjusted related expectations in tests. - Removed constants for maximum announce findings characters and summary words, simplifying the announcement logic. - Updated the handling of findings to retain full content instead of truncating, ensuring more informative outputs. - Cleaned up unused imports in the commands-subagents file to enhance code clarity. * feat(tests): enhance billing error handling in user-facing text - Added tests to ensure that normal text mentioning billing plans is not rewritten, preserving user context. - Updated the `isBillingErrorMessage` and `sanitizeUserFacingText` functions to improve handling of billing-related messages. - Introduced new test cases for various scenarios involving billing messages to ensure accurate processing and output. - Enhanced the subagent announce flow to correctly manage active descendant runs, preventing premature announcements. * feat(subagent): enhance workflow guidance and auto-announcement clarity - Added a new guideline in the subagent system prompt to emphasize trust in push-based completion, discouraging busy polling for status updates. - Updated documentation to clarify that sub-agents will automatically announce their results, improving user understanding of the workflow. - Enhanced tests to verify the new guidance on avoiding polling loops and to ensure the accuracy of the updated prompts. * fix(cron): avoid announcing interim subagent spawn acks * chore: clean post-rebase imports * fix(cron): fall back to child replies when parent stays interim * fix(subagents): make active-run guidance advisory * fix(subagents): update announce flow to handle active descendants and enhance test coverage - Modified the announce flow to defer announcements when active descendant runs are present, ensuring accurate status reporting. - Updated tests to verify the new behavior, including scenarios where no fallback requester is available and ensuring proper handling of finished subagents. - Enhanced the announce formatting to include an `expectFinal` flag for better clarity in the announcement process. * fix(subagents): enhance announce flow and formatting for user updates - Updated the announce flow to provide clearer instructions for user updates based on active subagent runs and requester context. - Refactored the announcement logic to improve clarity and ensure internal context remains private. - Enhanced tests to verify the new message expectations and formatting, including updated prompts for user-facing updates. - Introduced a new function to build reply instructions based on session context, improving the overall announcement process. * fix: resolve prep blockers and changelog placement (#14447) (thanks @tyler6204) * fix: restore cron delivery-plan import after rebase (#14447) (thanks @tyler6204) * fix: resolve test failures from rebase conflicts (#14447) (thanks @tyler6204) * fix: apply formatting after rebase (#14447) (thanks @tyler6204)
2026-02-14 22:03:45 -08:00
import { isCronSessionKey, isSubagentSessionKey } from "../../routing/session-key.js";
import { emitSessionTranscriptUpdate } from "../../sessions/transcript-events.js";
import { buildTtsSystemPromptHint } from "../../tts/tts.js";
import { resolveUserPath } from "../../utils.js";
2026-01-14 01:08:15 +00:00
import { normalizeMessageChannel } from "../../utils/message-channel.js";
import { isReasoningTagProvider } from "../../utils/provider-utils.js";
2026-01-30 03:15:10 +01:00
import { resolveOpenClawAgentDir } from "../agent-paths.js";
import { resolveSessionAgentId, resolveSessionAgentIds } from "../agent-scope.js";
import type { ExecElevatedDefaults } from "../bash-tools.js";
import { makeBootstrapWarn, resolveBootstrapContextForRun } from "../bootstrap-files.js";
import { listChannelSupportedActions, resolveChannelMessageToolHints } from "../channel-tools.js";
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
import { resolveContextWindowInfo } from "../context-window-guard.js";
import { ensureCustomApiRegistered } from "../custom-api-registry.js";
import { formatUserTime, resolveUserTimeFormat, resolveUserTimezone } from "../date-time.js";
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
import { DEFAULT_CONTEXT_TOKENS, DEFAULT_MODEL, DEFAULT_PROVIDER } from "../defaults.js";
import { resolveOpenClawDocsPath } from "../docs-path.js";
import { resolveMemorySearchConfig } from "../memory-search.js";
import {
applyLocalNoAuthHeaderOverride,
getApiKeyForModel,
resolveModelAuthMode,
} from "../model-auth.js";
import { supportsModelTools } from "../model-tool-support.js";
2026-01-30 03:15:10 +01:00
import { ensureOpenClawModelsJson } from "../models-config.js";
import { createConfiguredOllamaStreamFn } from "../ollama-stream.js";
import { resolveOwnerDisplaySetting } from "../owner-display.js";
2026-01-14 01:08:15 +00:00
import {
ensureSessionHeader,
validateAnthropicTurns,
validateGeminiTurns,
} from "../pi-embedded-helpers.js";
import { createPreparedEmbeddedPiSettingsManager } from "../pi-project-settings.js";
2026-01-30 03:15:10 +01:00
import { createOpenClawCodingTools } from "../pi-tools.js";
import { ensureRuntimePluginsLoaded } from "../runtime-plugins.js";
2026-01-14 01:08:15 +00:00
import { resolveSandboxContext } from "../sandbox.js";
import { repairSessionFileIfNeeded } from "../session-file-repair.js";
2026-01-14 01:08:15 +00:00
import { guardSessionManager } from "../session-tool-result-guard-wrapper.js";
import { sanitizeToolUseResultPairing } from "../session-transcript-repair.js";
import {
acquireSessionWriteLock,
resolveSessionLockMaxHoldFromTimeout,
} from "../session-write-lock.js";
import { detectRuntimeShell } from "../shell-utils.js";
2026-01-14 01:08:15 +00:00
import {
applySkillEnvOverrides,
applySkillEnvOverridesFromSnapshot,
resolveSkillsPromptForRun,
type SkillSnapshot,
} from "../skills.js";
import { resolveTranscriptPolicy } from "../transcript-policy.js";
import {
compactWithSafetyTimeout,
feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds (#46889) * feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds The hardcoded 5-minute (300s) compaction timeout causes large sessions to enter a death spiral where compaction repeatedly fails and the session grows indefinitely. This adds agents.defaults.compaction.timeoutSeconds to allow operators to override the compaction safety timeout. Default raised to 900s (15min) which is sufficient for sessions up to ~400k tokens. The resolved timeout is also used for the session write lock duration so locks don't expire before compaction completes. Fixes #38233 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add resolveCompactionTimeoutMs tests Cover config resolution edge cases: undefined config, missing compaction section, valid seconds, fractional values, zero, negative, NaN, and Infinity. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add timeoutSeconds to compaction Zod schema The compaction object schema uses .strict(), so setting the new timeoutSeconds config option would fail validation at startup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: enforce integer constraint on compaction timeoutSeconds schema Prevents sub-second values like 0.5 which would floor to 0ms and cause immediate compaction timeout. Matches pattern of other integer timeout fields in the schema. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: clamp compaction timeout to Node timer-safe maximum Values above ~2.1B ms overflow Node's setTimeout to 1ms, causing immediate timeout. Clamp to MAX_SAFE_TIMEOUT_MS matching the pattern in agents/timeout.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add FIELD_LABELS entry for compaction timeoutSeconds Maintains label/help parity invariant enforced by schema.help.quality.test.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: align compaction timeouts with abort handling * fix: land compaction timeout handling (#46889) (thanks @asyncjason) --------- Co-authored-by: Jason Separovic <jason@wilma.dog> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-14 23:34:48 -07:00
resolveCompactionTimeoutMs,
} from "./compaction-safety-timeout.js";
import { buildEmbeddedExtensionFactories } from "./extensions.js";
import {
logToolSchemasForGoogle,
sanitizeSessionHistory,
sanitizeToolsForGoogle,
} from "./google.js";
import { getDmHistoryLimitFromSessionKey, limitHistoryTurns } from "./history.js";
2026-01-14 01:08:15 +00:00
import { resolveGlobalLane, resolveSessionLane } from "./lanes.js";
import { log } from "./logger.js";
fix(openrouter): silently dropped images for new OpenRouter models — runtime capability detection (#45824) * fix: fetch OpenRouter model capabilities at runtime for unknown models When an OpenRouter model is not in the built-in static snapshot from pi-ai, the fallback hardcodes input: ["text"], silently dropping images. Query the OpenRouter API at runtime to detect actual capabilities (image support, reasoning, context window) for models not in the built-in list. Results are cached in memory for 1 hour. On API failure/timeout, falls back to text-only (no regression). * feat(openrouter): add disk cache for OpenRouter model capabilities Persist the OpenRouter model catalog to ~/.openclaw/cache/openrouter-models.json so it survives process restarts. Cache lookup order: 1. In-memory Map (instant) 2. On-disk JSON file (avoids network on restart) 3. OpenRouter API fetch (populates both layers) Also triggers a background refresh when a model is not found in the cache, in case it was newly added to OpenRouter. * refactor(openrouter): remove pre-warm, use pure lazy-load with disk cache - Remove eager ensureOpenRouterModelCache() from run.ts - Remove TTL — model capabilities are stable, no periodic re-fetching - Cache lookup: in-memory → disk → API fetch (only when needed) - API is only called when no cache exists or a model is not found - Disk cache persists across gateway restarts * fix(openrouter): address review feedback - Fix timer leak: move clearTimeout to finally block - Fix modality check: only check input side of "->" separator to avoid matching image-generation models (text->image) - Use resolveStateDir() instead of hardcoded homedir()/.openclaw - Separate cache dir and filename constants - Add utf-8 encoding to writeFileSync for consistency - Add data validation when reading disk cache * ci: retrigger checks * fix: preload unknown OpenRouter model capabilities before resolve * fix: accept top-level OpenRouter max token metadata * fix: update changelog for OpenRouter runtime capability lookup (#45824) (thanks @DJjjjhao) * fix: avoid redundant OpenRouter refetches and preserve suppression guards --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-15 14:18:39 +08:00
import { buildModelAliasLines, resolveModelAsync } from "./model.js";
2026-01-14 01:08:15 +00:00
import { buildEmbeddedSandboxInfo } from "./sandbox-info.js";
import { prewarmSessionFile, trackSessionManagerAccess } from "./session-manager-cache.js";
import { resolveEmbeddedRunSkillEntries } from "./skills-runtime.js";
import {
applySystemPromptOverrideToSession,
buildEmbeddedSystemPrompt,
createSystemPromptOverride,
} from "./system-prompt.js";
import { collectAllowedToolNames } from "./tool-name-allowlist.js";
2026-01-14 01:08:15 +00:00
import { splitSdkTools } from "./tool-split.js";
import type { EmbeddedPiCompactResult } from "./types.js";
import { describeUnknownError, mapThinkingLevel } from "./utils.js";
fix(agents): wait for agent idle before flushing pending tool results (#13746) * fix(agents): wait for agent idle before flushing pending tool results When pi-agent-core's auto-retry mechanism handles overloaded/rate-limit errors, it resolves waitForRetry() on assistant message receipt — before tool execution completes in the retried agent loop. This causes the attempt's finally block to call flushPendingToolResults() while tools are still executing, inserting synthetic 'missing tool result' errors and causing silent agent failures. The fix adds a waitForIdle() call before the flush to ensure the agent's retry loop (including tool execution) has fully completed. Evidence from real session: tool call and synthetic error were only 53ms apart — the tool never had a chance to execute before being flushed. Root cause is in pi-agent-core's _resolveRetry() firing on message_end instead of agent_end, but this workaround in OpenClaw prevents the symptom without requiring an upstream fix. Fixes #8643 Fixes #13351 Refs #6682, #12595 * test: add tests for tool result flush race condition Validates that: - Real tool results are not replaced by synthetic errors when they arrive in time - Flush correctly inserts synthetic errors for genuinely orphaned tool calls - Flush is a no-op after real tool results have already been received Refs #8643, #13748 * fix(agents): add waitForIdle to all flushPendingToolResults call sites The original fix only covered the main run finally block, but there are two additional call sites that can trigger flushPendingToolResults while tools are still executing: 1. The catch block in attempt.ts (session setup error handler) 2. The finally block in compact.ts (compaction teardown) Both now await agent.waitForIdle() with a 30s timeout before flushing, matching the pattern already applied to the main finally block. Production testing on VPS with debug logging confirmed these additional paths can fire during sub-agent runs, producing spurious synthetic 'missing tool result' errors. * fix(agents): centralize idle-wait flush and clear timeout handle --------- Co-authored-by: Renue Development <dev@renuebyscience.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:35:43 +11:00
import { flushPendingToolResultsAfterIdle } from "./wait-for-idle-before-flush.js";
2026-01-14 01:08:15 +00:00
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
export type CompactEmbeddedPiSessionParams = {
2026-01-14 01:08:15 +00:00
sessionId: string;
runId?: string;
2026-01-14 01:08:15 +00:00
sessionKey?: string;
messageChannel?: string;
messageProvider?: string;
agentAccountId?: string;
authProfileId?: string;
/** Group id for channel-level tool policy resolution. */
groupId?: string | null;
/** Group channel label (e.g. #general) for channel-level tool policy resolution. */
groupChannel?: string | null;
/** Group space label (e.g. guild/team id) for channel-level tool policy resolution. */
groupSpace?: string | null;
/** Parent session key for subagent policy inheritance. */
spawnedBy?: string | null;
/** Whether the sender is an owner (required for owner-only tools). */
senderIsOwner?: boolean;
2026-01-14 01:08:15 +00:00
sessionFile: string;
/** Optional caller-observed live prompt tokens used for compaction diagnostics. */
currentTokenCount?: number;
2026-01-14 01:08:15 +00:00
workspaceDir: string;
agentDir?: string;
2026-01-30 03:15:10 +01:00
config?: OpenClawConfig;
2026-01-14 01:08:15 +00:00
skillsSnapshot?: SkillSnapshot;
provider?: string;
model?: string;
thinkLevel?: ThinkLevel;
reasoningLevel?: ReasoningLevel;
bashElevated?: ExecElevatedDefaults;
customInstructions?: string;
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
tokenBudget?: number;
force?: boolean;
trigger?: "overflow" | "manual";
diagId?: string;
attempt?: number;
maxAttempts?: number;
2026-01-14 01:08:15 +00:00
lane?: string;
enqueue?: typeof enqueueCommand;
extraSystemPrompt?: string;
ownerNumbers?: string[];
feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds (#46889) * feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds The hardcoded 5-minute (300s) compaction timeout causes large sessions to enter a death spiral where compaction repeatedly fails and the session grows indefinitely. This adds agents.defaults.compaction.timeoutSeconds to allow operators to override the compaction safety timeout. Default raised to 900s (15min) which is sufficient for sessions up to ~400k tokens. The resolved timeout is also used for the session write lock duration so locks don't expire before compaction completes. Fixes #38233 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add resolveCompactionTimeoutMs tests Cover config resolution edge cases: undefined config, missing compaction section, valid seconds, fractional values, zero, negative, NaN, and Infinity. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add timeoutSeconds to compaction Zod schema The compaction object schema uses .strict(), so setting the new timeoutSeconds config option would fail validation at startup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: enforce integer constraint on compaction timeoutSeconds schema Prevents sub-second values like 0.5 which would floor to 0ms and cause immediate compaction timeout. Matches pattern of other integer timeout fields in the schema. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: clamp compaction timeout to Node timer-safe maximum Values above ~2.1B ms overflow Node's setTimeout to 1ms, causing immediate timeout. Clamp to MAX_SAFE_TIMEOUT_MS matching the pattern in agents/timeout.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add FIELD_LABELS entry for compaction timeoutSeconds Maintains label/help parity invariant enforced by schema.help.quality.test.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: align compaction timeouts with abort handling * fix: land compaction timeout handling (#46889) (thanks @asyncjason) --------- Co-authored-by: Jason Separovic <jason@wilma.dog> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-14 23:34:48 -07:00
abortSignal?: AbortSignal;
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
};
2026-01-14 01:08:15 +00:00
type CompactionMessageMetrics = {
messages: number;
historyTextChars: number;
toolResultChars: number;
estTokens?: number;
contributors: Array<{ role: string; chars: number; tool?: string }>;
};
function hasRealConversationContent(msg: AgentMessage): boolean {
return msg.role === "user" || msg.role === "assistant" || msg.role === "toolResult";
}
function createCompactionDiagId(): string {
return `cmp-${Date.now().toString(36)}-${generateSecureToken(4)}`;
}
function normalizeObservedTokenCount(value: unknown): number | undefined {
return typeof value === "number" && Number.isFinite(value) && value > 0
? Math.floor(value)
: undefined;
}
function getMessageTextChars(msg: AgentMessage): number {
const content = (msg as { content?: unknown }).content;
if (typeof content === "string") {
return content.length;
}
if (!Array.isArray(content)) {
return 0;
}
let total = 0;
for (const block of content) {
if (!block || typeof block !== "object") {
continue;
}
const text = (block as { text?: unknown }).text;
if (typeof text === "string") {
total += text.length;
}
}
return total;
}
function resolveMessageToolLabel(msg: AgentMessage): string | undefined {
const candidate =
(msg as { toolName?: unknown }).toolName ??
(msg as { name?: unknown }).name ??
(msg as { tool?: unknown }).tool;
return typeof candidate === "string" && candidate.trim().length > 0 ? candidate : undefined;
}
function summarizeCompactionMessages(messages: AgentMessage[]): CompactionMessageMetrics {
let historyTextChars = 0;
let toolResultChars = 0;
const contributors: Array<{ role: string; chars: number; tool?: string }> = [];
let estTokens = 0;
let tokenEstimationFailed = false;
for (const msg of messages) {
const role = typeof msg.role === "string" ? msg.role : "unknown";
const chars = getMessageTextChars(msg);
historyTextChars += chars;
if (role === "toolResult") {
toolResultChars += chars;
}
contributors.push({ role, chars, tool: resolveMessageToolLabel(msg) });
if (!tokenEstimationFailed) {
try {
estTokens += estimateTokens(msg);
} catch {
tokenEstimationFailed = true;
}
}
}
return {
messages: messages.length,
historyTextChars,
toolResultChars,
estTokens: tokenEstimationFailed ? undefined : estTokens,
contributors: contributors.toSorted((a, b) => b.chars - a.chars).slice(0, 3),
};
}
function classifyCompactionReason(reason?: string): string {
const text = (reason ?? "").trim().toLowerCase();
if (!text) {
return "unknown";
}
if (text.includes("nothing to compact")) {
return "no_compactable_entries";
}
if (text.includes("below threshold")) {
return "below_threshold";
}
if (text.includes("already compacted")) {
return "already_compacted_recently";
}
if (text.includes("still exceeds target")) {
return "live_context_still_exceeds_target";
}
if (text.includes("guard")) {
return "guard_blocked";
}
if (text.includes("summary")) {
return "summary_failed";
}
if (text.includes("timed out") || text.includes("timeout")) {
return "timeout";
}
if (
text.includes("400") ||
text.includes("401") ||
text.includes("403") ||
text.includes("429")
) {
return "provider_error_4xx";
}
if (
text.includes("500") ||
text.includes("502") ||
text.includes("503") ||
text.includes("504")
) {
return "provider_error_5xx";
}
return "unknown";
}
function resolvePostCompactionIndexSyncMode(config?: OpenClawConfig): "off" | "async" | "await" {
const mode = config?.agents?.defaults?.compaction?.postIndexSync;
if (mode === "off" || mode === "async" || mode === "await") {
return mode;
}
return "async";
}
async function runPostCompactionSessionMemorySync(params: {
config?: OpenClawConfig;
sessionKey?: string;
sessionFile: string;
}): Promise<void> {
if (!params.config) {
return;
}
try {
const sessionFile = params.sessionFile.trim();
if (!sessionFile) {
return;
}
const agentId = resolveSessionAgentId({
sessionKey: params.sessionKey,
config: params.config,
});
const resolvedMemory = resolveMemorySearchConfig(params.config, agentId);
if (!resolvedMemory || !resolvedMemory.sources.includes("sessions")) {
return;
}
if (!resolvedMemory.sync.sessions.postCompactionForce) {
return;
}
const { manager } = await getMemorySearchManager({
cfg: params.config,
agentId,
});
if (!manager?.sync) {
return;
}
const syncTask = manager.sync({
reason: "post-compaction",
sessionFiles: [sessionFile],
});
await syncTask;
} catch (err) {
log.warn(`memory sync skipped (post-compaction): ${String(err)}`);
}
}
function syncPostCompactionSessionMemory(params: {
config?: OpenClawConfig;
sessionKey?: string;
sessionFile: string;
mode: "off" | "async" | "await";
}): Promise<void> {
if (params.mode === "off" || !params.config) {
return Promise.resolve();
}
const syncTask = runPostCompactionSessionMemorySync({
config: params.config,
sessionKey: params.sessionKey,
sessionFile: params.sessionFile,
});
if (params.mode === "await") {
return syncTask;
}
void syncTask;
return Promise.resolve();
}
async function runPostCompactionSideEffects(params: {
config?: OpenClawConfig;
sessionKey?: string;
sessionFile: string;
}): Promise<void> {
const sessionFile = params.sessionFile.trim();
if (!sessionFile) {
return;
}
emitSessionTranscriptUpdate(sessionFile);
await syncPostCompactionSessionMemory({
config: params.config,
sessionKey: params.sessionKey,
sessionFile,
mode: resolvePostCompactionIndexSyncMode(params.config),
});
}
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
/**
* Core compaction logic without lane queueing.
* Use this when already inside a session/global lane to avoid deadlocks.
*/
export async function compactEmbeddedPiSessionDirect(
params: CompactEmbeddedPiSessionParams,
): Promise<EmbeddedPiCompactResult> {
const startedAt = Date.now();
const diagId = params.diagId?.trim() || createCompactionDiagId();
const trigger = params.trigger ?? "manual";
const attempt = params.attempt ?? 1;
const maxAttempts = params.maxAttempts ?? 1;
const runId = params.runId ?? params.sessionId;
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const resolvedWorkspace = resolveUserPath(params.workspaceDir);
ensureRuntimePluginsLoaded({
config: params.config,
workspaceDir: resolvedWorkspace,
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const prevCwd = process.cwd();
// Resolve compaction model: prefer config override, then fall back to caller-supplied model
const compactionModelOverride = params.config?.agents?.defaults?.compaction?.model?.trim();
let provider: string;
let modelId: string;
// When switching provider via override, drop the primary auth profile to avoid
// sending the wrong credentials (e.g. OpenAI profile token to OpenRouter).
let authProfileId: string | undefined = params.authProfileId;
if (compactionModelOverride) {
const slashIdx = compactionModelOverride.indexOf("/");
if (slashIdx > 0) {
provider = compactionModelOverride.slice(0, slashIdx).trim();
modelId = compactionModelOverride.slice(slashIdx + 1).trim() || DEFAULT_MODEL;
// Provider changed — drop primary auth profile so getApiKeyForModel
// falls back to provider-based key resolution for the override model.
if (provider !== (params.provider ?? "").trim()) {
authProfileId = undefined;
}
} else {
provider = (params.provider ?? DEFAULT_PROVIDER).trim() || DEFAULT_PROVIDER;
modelId = compactionModelOverride;
}
} else {
provider = (params.provider ?? DEFAULT_PROVIDER).trim() || DEFAULT_PROVIDER;
modelId = (params.model ?? DEFAULT_MODEL).trim() || DEFAULT_MODEL;
}
const fail = (reason: string): EmbeddedPiCompactResult => {
log.warn(
`[compaction-diag] end runId=${runId} sessionKey=${params.sessionKey ?? params.sessionId} ` +
`diagId=${diagId} trigger=${trigger} provider=${provider}/${modelId} ` +
`attempt=${attempt} maxAttempts=${maxAttempts} outcome=failed reason=${classifyCompactionReason(reason)} ` +
`durationMs=${Date.now() - startedAt}`,
);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
return {
ok: false,
compacted: false,
reason,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
};
};
const agentDir = params.agentDir ?? resolveOpenClawAgentDir();
await ensureOpenClawModelsJson(params.config, agentDir);
fix(openrouter): silently dropped images for new OpenRouter models — runtime capability detection (#45824) * fix: fetch OpenRouter model capabilities at runtime for unknown models When an OpenRouter model is not in the built-in static snapshot from pi-ai, the fallback hardcodes input: ["text"], silently dropping images. Query the OpenRouter API at runtime to detect actual capabilities (image support, reasoning, context window) for models not in the built-in list. Results are cached in memory for 1 hour. On API failure/timeout, falls back to text-only (no regression). * feat(openrouter): add disk cache for OpenRouter model capabilities Persist the OpenRouter model catalog to ~/.openclaw/cache/openrouter-models.json so it survives process restarts. Cache lookup order: 1. In-memory Map (instant) 2. On-disk JSON file (avoids network on restart) 3. OpenRouter API fetch (populates both layers) Also triggers a background refresh when a model is not found in the cache, in case it was newly added to OpenRouter. * refactor(openrouter): remove pre-warm, use pure lazy-load with disk cache - Remove eager ensureOpenRouterModelCache() from run.ts - Remove TTL — model capabilities are stable, no periodic re-fetching - Cache lookup: in-memory → disk → API fetch (only when needed) - API is only called when no cache exists or a model is not found - Disk cache persists across gateway restarts * fix(openrouter): address review feedback - Fix timer leak: move clearTimeout to finally block - Fix modality check: only check input side of "->" separator to avoid matching image-generation models (text->image) - Use resolveStateDir() instead of hardcoded homedir()/.openclaw - Separate cache dir and filename constants - Add utf-8 encoding to writeFileSync for consistency - Add data validation when reading disk cache * ci: retrigger checks * fix: preload unknown OpenRouter model capabilities before resolve * fix: accept top-level OpenRouter max token metadata * fix: update changelog for OpenRouter runtime capability lookup (#45824) (thanks @DJjjjhao) * fix: avoid redundant OpenRouter refetches and preserve suppression guards --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-15 14:18:39 +08:00
const { model, error, authStorage, modelRegistry } = await resolveModelAsync(
provider,
modelId,
agentDir,
params.config,
);
if (!model) {
const reason = error ?? `Unknown model: ${provider}/${modelId}`;
return fail(reason);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
}
let runtimeModel = model;
let apiKeyInfo: Awaited<ReturnType<typeof getApiKeyForModel>> | null = null;
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
try {
apiKeyInfo = await getApiKeyForModel({
model: runtimeModel,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
cfg: params.config,
profileId: authProfileId,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
agentDir,
});
if (!apiKeyInfo.apiKey) {
if (apiKeyInfo.mode !== "aws-sdk") {
throw new Error(
`No API key resolved for provider "${runtimeModel.provider}" (auth mode: ${apiKeyInfo.mode}).`,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
);
2026-01-14 01:08:15 +00:00
}
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
} else {
const preparedAuth = await prepareProviderRuntimeAuth({
provider: runtimeModel.provider,
config: params.config,
workspaceDir: resolvedWorkspace,
env: process.env,
context: {
config: params.config,
agentDir,
workspaceDir: resolvedWorkspace,
env: process.env,
provider: runtimeModel.provider,
modelId,
model: runtimeModel,
apiKey: apiKeyInfo.apiKey,
authMode: apiKeyInfo.mode,
profileId: apiKeyInfo.profileId,
},
});
if (preparedAuth?.baseUrl) {
runtimeModel = { ...runtimeModel, baseUrl: preparedAuth.baseUrl };
}
const runtimeApiKey = preparedAuth?.apiKey ?? apiKeyInfo.apiKey;
if (!runtimeApiKey) {
throw new Error(`Provider "${runtimeModel.provider}" runtime auth returned no apiKey.`);
}
authStorage.setRuntimeApiKey(runtimeModel.provider, runtimeApiKey);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
}
} catch (err) {
const reason = describeUnknownError(err);
return fail(reason);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
}
await fs.mkdir(resolvedWorkspace, { recursive: true });
const sandboxSessionKey = params.sessionKey?.trim() || params.sessionId;
const sandbox = await resolveSandboxContext({
config: params.config,
sessionKey: sandboxSessionKey,
workspaceDir: resolvedWorkspace,
});
const effectiveWorkspace = sandbox?.enabled
? sandbox.workspaceAccess === "rw"
? resolvedWorkspace
: sandbox.workspaceDir
: resolvedWorkspace;
await fs.mkdir(effectiveWorkspace, { recursive: true });
await ensureSessionHeader({
sessionFile: params.sessionFile,
sessionId: params.sessionId,
cwd: effectiveWorkspace,
});
let restoreSkillEnv: (() => void) | undefined;
process.chdir(effectiveWorkspace);
try {
const { shouldLoadSkillEntries, skillEntries } = resolveEmbeddedRunSkillEntries({
workspaceDir: effectiveWorkspace,
config: params.config,
skillsSnapshot: params.skillsSnapshot,
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
restoreSkillEnv = params.skillsSnapshot
? applySkillEnvOverridesFromSnapshot({
snapshot: params.skillsSnapshot,
config: params.config,
})
: applySkillEnvOverrides({
skills: skillEntries ?? [],
config: params.config,
2026-01-14 01:08:15 +00:00
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const skillsPrompt = resolveSkillsPromptForRun({
skillsSnapshot: params.skillsSnapshot,
entries: shouldLoadSkillEntries ? skillEntries : undefined,
config: params.config,
workspaceDir: effectiveWorkspace,
});
2026-01-14 01:08:15 +00:00
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const sessionLabel = params.sessionKey ?? params.sessionId;
const resolvedMessageProvider = params.messageChannel ?? params.messageProvider;
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const { contextFiles } = await resolveBootstrapContextForRun({
workspaceDir: effectiveWorkspace,
config: params.config,
sessionKey: params.sessionKey,
sessionId: params.sessionId,
warn: makeBootstrapWarn({ sessionLabel, warn: (message) => log.warn(message) }),
});
// Apply contextTokens cap to model so pi-coding-agent's auto-compaction
// threshold uses the effective limit, not the native context window.
const ctxInfo = resolveContextWindowInfo({
cfg: params.config,
provider,
modelId,
modelContextWindow: runtimeModel.contextWindow,
defaultTokens: DEFAULT_CONTEXT_TOKENS,
});
const effectiveModel = applyLocalNoAuthHeaderOverride(
ctxInfo.tokens < (runtimeModel.contextWindow ?? Infinity)
? { ...runtimeModel, contextWindow: ctxInfo.tokens }
: runtimeModel,
apiKeyInfo,
);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const runAbortController = new AbortController();
2026-01-30 03:15:10 +01:00
const toolsRaw = createOpenClawCodingTools({
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
exec: {
elevated: params.bashElevated,
},
sandbox,
messageProvider: resolvedMessageProvider,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
agentAccountId: params.agentAccountId,
sessionKey: sandboxSessionKey,
sessionId: params.sessionId,
runId: params.runId,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
groupId: params.groupId,
groupChannel: params.groupChannel,
groupSpace: params.groupSpace,
spawnedBy: params.spawnedBy,
senderIsOwner: params.senderIsOwner,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
agentDir,
workspaceDir: effectiveWorkspace,
config: params.config,
abortSignal: runAbortController.signal,
modelProvider: model.provider,
modelId,
modelContextWindowTokens: ctxInfo.tokens,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
modelAuthMode: resolveModelAuthMode(model.provider, params.config),
});
const tools = sanitizeToolsForGoogle({
tools: supportsModelTools(runtimeModel) ? toolsRaw : [],
provider,
});
const allowedToolNames = collectAllowedToolNames({ tools });
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
logToolSchemasForGoogle({ tools, provider });
const machineName = await getMachineDisplayName();
const runtimeChannel = normalizeMessageChannel(params.messageChannel ?? params.messageProvider);
let runtimeCapabilities = runtimeChannel
? (resolveChannelCapabilities({
cfg: params.config,
channel: runtimeChannel,
accountId: params.agentAccountId,
}) ?? [])
: undefined;
if (runtimeChannel === "telegram" && params.config) {
const inlineButtonsScope = resolveTelegramInlineButtonsScope({
cfg: params.config,
accountId: params.agentAccountId ?? undefined,
});
if (inlineButtonsScope !== "off") {
if (!runtimeCapabilities) {
runtimeCapabilities = [];
}
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
if (
!runtimeCapabilities.some((cap) => String(cap).trim().toLowerCase() === "inlinebuttons")
) {
runtimeCapabilities.push("inlineButtons");
2026-01-14 01:08:15 +00:00
}
}
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
}
2026-01-25 03:20:09 +00:00
const reactionGuidance =
runtimeChannel && params.config
? (() => {
if (runtimeChannel === "telegram") {
const resolved = resolveTelegramReactionLevel({
cfg: params.config,
accountId: params.agentAccountId ?? undefined,
});
const level = resolved.agentReactionGuidance;
return level ? { level, channel: "Telegram" } : undefined;
}
if (runtimeChannel === "signal") {
const resolved = resolveSignalReactionLevel({
cfg: params.config,
accountId: params.agentAccountId ?? undefined,
});
const level = resolved.agentReactionGuidance;
return level ? { level, channel: "Signal" } : undefined;
}
return undefined;
})()
: undefined;
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
// Resolve channel-specific message actions for system prompt
const channelActions = runtimeChannel
? listChannelSupportedActions({
cfg: params.config,
channel: runtimeChannel,
})
: undefined;
const messageToolHints = runtimeChannel
? resolveChannelMessageToolHints({
cfg: params.config,
channel: runtimeChannel,
accountId: params.agentAccountId,
})
: undefined;
2026-01-14 01:08:15 +00:00
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const runtimeInfo = {
host: machineName,
os: `${os.type()} ${os.release()}`,
arch: os.arch(),
node: process.version,
model: `${provider}/${modelId}`,
shell: detectRuntimeShell(),
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
channel: runtimeChannel,
capabilities: runtimeCapabilities,
channelActions,
};
const sandboxInfo = buildEmbeddedSandboxInfo(sandbox, params.bashElevated);
const reasoningTagHint = isReasoningTagProvider(provider);
const userTimezone = resolveUserTimezone(params.config?.agents?.defaults?.userTimezone);
const userTimeFormat = resolveUserTimeFormat(params.config?.agents?.defaults?.timeFormat);
const userTime = formatUserTime(new Date(), userTimezone, userTimeFormat);
const { defaultAgentId, sessionAgentId } = resolveSessionAgentIds({
sessionKey: params.sessionKey,
config: params.config,
});
const isDefaultAgent = sessionAgentId === defaultAgentId;
Agents: add nested subagent orchestration controls and reduce subagent token waste (#14447) * Agents: add subagent orchestration controls * Agents: add subagent orchestration controls (WIP uncommitted changes) * feat(subagents): add depth-based spawn gating for sub-sub-agents * feat(subagents): tool policy, registry, and announce chain for nested agents * feat(subagents): system prompt, docs, changelog for nested sub-agents * fix(subagents): prevent model fallback override, show model during active runs, and block context overflow fallback Bug 1: When a session has an explicit model override (e.g., gpt/openai-codex), the fallback candidate logic in resolveFallbackCandidates silently appended the global primary model (opus) as a backstop. On reinjection/steer with a transient error, the session could fall back to opus which has a smaller context window and crash. Fix: when storedModelOverride is set, pass fallbacksOverride ?? [] instead of undefined, preventing the implicit primary backstop. Bug 2: Active subagents showed 'model n/a' in /subagents list because resolveModelDisplay only read entry.model/modelProvider (populated after run completes). Fix: fall back to modelOverride/providerOverride fields which are populated at spawn time via sessions.patch. Bug 3: Context overflow errors (prompt too long, context_length_exceeded) could theoretically escape runEmbeddedPiAgent and be treated as failover candidates in runWithModelFallback, causing a switch to a model with a smaller context window. Fix: in runWithModelFallback, detect context overflow errors via isLikelyContextOverflowError and rethrow them immediately instead of trying the next model candidate. * fix(subagents): track spawn depth in session store and fix announce routing for nested agents * Fix compaction status tracking and dedupe overflow compaction triggers * fix(subagents): enforce depth block via session store and implement cascade kill * fix: inject group chat context into system prompt * fix(subagents): always write model to session store at spawn time * Preserve spawnDepth when agent handler rewrites session entry * fix(subagents): suppress announce on steer-restart * fix(subagents): fallback spawned session model to runtime default * fix(subagents): enforce spawn depth when caller key resolves by sessionId * feat(subagents): implement active-first ordering for numeric targets and enhance task display - Added a test to verify that subagents with numeric targets follow an active-first list ordering. - Updated `resolveSubagentTarget` to sort subagent runs based on active status and recent activity. - Enhanced task display in command responses to prevent truncation of long task descriptions. - Introduced new utility functions for compacting task text and managing subagent run states. * fix(subagents): show model for active runs via run record fallback When the spawned model matches the agent's default model, the session store's override fields are intentionally cleared (isDefault: true). The model/modelProvider fields are only populated after the run completes. This left active subagents showing 'model n/a'. Fix: store the resolved model on SubagentRunRecord at registration time, and use it as a fallback in both display paths (subagents tool and /subagents command) when the session store entry has no model info. Changes: - SubagentRunRecord: add optional model field - registerSubagentRun: accept and persist model param - sessions-spawn-tool: pass resolvedModel to registerSubagentRun - subagents-tool: pass run record model as fallback to resolveModelDisplay - commands-subagents: pass run record model as fallback to resolveModelDisplay * feat(chat): implement session key resolution and reset on sidebar navigation - Added functions to resolve the main session key and reset chat state when switching sessions from the sidebar. - Updated the `renderTab` function to handle session key changes when navigating to the chat tab. - Introduced a test to verify that the session resets to "main" when opening chat from the sidebar navigation. * fix: subagent timeout=0 passthrough and fallback prompt duplication Bug 1: runTimeoutSeconds=0 now means 'no timeout' instead of applying 600s default - sessions-spawn-tool: default to undefined (not 0) when neither timeout param is provided; use != null check so explicit 0 passes through to gateway - agent.ts: accept 0 as valid timeout (resolveAgentTimeoutMs already handles 0 → MAX_SAFE_TIMEOUT_MS) Bug 2: model fallback no longer re-injects the original prompt as a duplicate - agent.ts: track fallback attempt index; on retries use a short continuation message instead of the full original prompt since the session file already contains it from the first attempt - Also skip re-sending images on fallback retries (already in session) * feat(subagents): truncate long task descriptions in subagents command output - Introduced a new utility function to format task previews, limiting their length to improve readability. - Updated the command handler to use the new formatting function, ensuring task descriptions are truncated appropriately. - Adjusted related tests to verify that long task descriptions are now truncated in the output. * refactor(subagents): update subagent registry path resolution and improve command output formatting - Replaced direct import of STATE_DIR with a utility function to resolve the state directory dynamically. - Enhanced the formatting of command output for active and recent subagents, adding separators for better readability. - Updated related tests to reflect changes in command output structure. * fix(subagent): default sessions_spawn to no timeout when runTimeoutSeconds omitted The previous fix (75a791106) correctly handled the case where runTimeoutSeconds was explicitly set to 0 ("no timeout"). However, when models omit the parameter entirely (which is common since the schema marks it as optional), runTimeoutSeconds resolved to undefined. undefined flowed through the chain as: sessions_spawn → timeout: undefined (since undefined != null is false) → gateway agent handler → agentCommand opts.timeout: undefined → resolveAgentTimeoutMs({ overrideSeconds: undefined }) → DEFAULT_AGENT_TIMEOUT_SECONDS (600s = 10 minutes) This caused subagents to be killed at exactly 10 minutes even though the user's intent (via TOOLS.md) was for subagents to run without a timeout. Fix: default runTimeoutSeconds to 0 (no timeout) when neither runTimeoutSeconds nor timeoutSeconds is provided by the caller. Subagent spawns are long-running by design and should not inherit the 600s agent-command default timeout. * fix(subagent): accept timeout=0 in agent-via-gateway path (second 600s default) * fix: thread timeout override through getReplyFromConfig dispatch path getReplyFromConfig called resolveAgentTimeoutMs({ cfg }) with no override, always falling back to the config default (600s). Add timeoutOverrideSeconds to GetReplyOptions and pass it through as overrideSeconds so callers of the dispatch chain can specify a custom timeout (0 = no timeout). This complements the existing timeout threading in agentCommand and the cron isolated-agent runner, which already pass overrideSeconds correctly. * feat(model-fallback): normalize OpenAI Codex model references and enhance fallback handling - Added normalization for OpenAI Codex model references, specifically converting "gpt-5.3-codex" to "openai-codex" before execution. - Updated the `resolveFallbackCandidates` function to utilize the new normalization logic. - Enhanced tests to verify the correct behavior of model normalization and fallback mechanisms. - Introduced a new test case to ensure that the normalization process works as expected for various input formats. * feat(tests): add unit tests for steer failure behavior in openclaw-tools - Introduced a new test file to validate the behavior of subagents when steer replacement dispatch fails. - Implemented tests to ensure that the announce behavior is restored correctly and that the suppression reason is cleared as expected. - Enhanced the subagent registry with a new function to clear steer restart suppression. - Updated related components to support the new test scenarios. * fix(subagents): replace stop command with kill in slash commands and documentation - Updated the `/subagents` command to replace `stop` with `kill` for consistency in controlling sub-agent runs. - Modified related documentation to reflect the change in command usage. - Removed legacy timeoutSeconds references from the sessions-spawn-tool schema and tests to streamline timeout handling. - Enhanced tests to ensure correct behavior of the updated commands and their interactions. * feat(tests): add unit tests for readLatestAssistantReply function - Introduced a new test file for the `readLatestAssistantReply` function to validate its behavior with various message scenarios. - Implemented tests to ensure the function correctly retrieves the latest assistant message and handles cases where the latest message has no text. - Mocked the gateway call to simulate different message histories for comprehensive testing. * feat(tests): enhance subagent kill-all cascade tests and announce formatting - Added a new test to verify that the `kill-all` command cascades through ended parents to active descendants in subagents. - Updated the subagent announce formatting tests to reflect changes in message structure, including the replacement of "Findings:" with "Result:" and the addition of new expectations for message content. - Improved the handling of long findings and stats in the announce formatting logic to ensure concise output. - Refactored related functions to enhance clarity and maintainability in the subagent registry and tools. * refactor(subagent): update announce formatting and remove unused constants - Modified the subagent announce formatting to replace "Findings:" with "Result:" and adjusted related expectations in tests. - Removed constants for maximum announce findings characters and summary words, simplifying the announcement logic. - Updated the handling of findings to retain full content instead of truncating, ensuring more informative outputs. - Cleaned up unused imports in the commands-subagents file to enhance code clarity. * feat(tests): enhance billing error handling in user-facing text - Added tests to ensure that normal text mentioning billing plans is not rewritten, preserving user context. - Updated the `isBillingErrorMessage` and `sanitizeUserFacingText` functions to improve handling of billing-related messages. - Introduced new test cases for various scenarios involving billing messages to ensure accurate processing and output. - Enhanced the subagent announce flow to correctly manage active descendant runs, preventing premature announcements. * feat(subagent): enhance workflow guidance and auto-announcement clarity - Added a new guideline in the subagent system prompt to emphasize trust in push-based completion, discouraging busy polling for status updates. - Updated documentation to clarify that sub-agents will automatically announce their results, improving user understanding of the workflow. - Enhanced tests to verify the new guidance on avoiding polling loops and to ensure the accuracy of the updated prompts. * fix(cron): avoid announcing interim subagent spawn acks * chore: clean post-rebase imports * fix(cron): fall back to child replies when parent stays interim * fix(subagents): make active-run guidance advisory * fix(subagents): update announce flow to handle active descendants and enhance test coverage - Modified the announce flow to defer announcements when active descendant runs are present, ensuring accurate status reporting. - Updated tests to verify the new behavior, including scenarios where no fallback requester is available and ensuring proper handling of finished subagents. - Enhanced the announce formatting to include an `expectFinal` flag for better clarity in the announcement process. * fix(subagents): enhance announce flow and formatting for user updates - Updated the announce flow to provide clearer instructions for user updates based on active subagent runs and requester context. - Refactored the announcement logic to improve clarity and ensure internal context remains private. - Enhanced tests to verify the new message expectations and formatting, including updated prompts for user-facing updates. - Introduced a new function to build reply instructions based on session context, improving the overall announcement process. * fix: resolve prep blockers and changelog placement (#14447) (thanks @tyler6204) * fix: restore cron delivery-plan import after rebase (#14447) (thanks @tyler6204) * fix: resolve test failures from rebase conflicts (#14447) (thanks @tyler6204) * fix: apply formatting after rebase (#14447) (thanks @tyler6204)
2026-02-14 22:03:45 -08:00
const promptMode =
isSubagentSessionKey(params.sessionKey) || isCronSessionKey(params.sessionKey)
? "minimal"
: "full";
2026-01-30 03:15:10 +01:00
const docsPath = await resolveOpenClawDocsPath({
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
workspaceDir: effectiveWorkspace,
argv1: process.argv[1],
cwd: process.cwd(),
moduleUrl: import.meta.url,
});
const ttsHint = params.config ? buildTtsSystemPromptHint(params.config) : undefined;
const ownerDisplay = resolveOwnerDisplaySetting(params.config);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const appendPrompt = buildEmbeddedSystemPrompt({
workspaceDir: effectiveWorkspace,
defaultThinkLevel: params.thinkLevel,
reasoningLevel: params.reasoningLevel ?? "off",
extraSystemPrompt: params.extraSystemPrompt,
ownerNumbers: params.ownerNumbers,
ownerDisplay: ownerDisplay.ownerDisplay,
ownerDisplaySecret: ownerDisplay.ownerDisplaySecret,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
reasoningTagHint,
heartbeatPrompt: isDefaultAgent
? resolveHeartbeatPrompt(params.config?.agents?.defaults?.heartbeat?.prompt)
: undefined,
skillsPrompt,
docsPath: docsPath ?? undefined,
ttsHint,
promptMode,
feat: ACP thread-bound agents (#23580) * docs: add ACP thread-bound agents plan doc * docs: expand ACP implementation specification * feat(acp): route ACP sessions through core dispatch and lifecycle cleanup * feat(acp): add /acp commands and Discord spawn gate * ACP: add acpx runtime plugin backend * fix(subagents): defer transient lifecycle errors before announce * Agents: harden ACP sessions_spawn and tighten spawn guidance * Agents: require explicit ACP target for runtime spawns * docs: expand ACP control-plane implementation plan * ACP: harden metadata seeding and spawn guidance * ACP: centralize runtime control-plane manager and fail-closed dispatch * ACP: harden runtime manager and unify spawn helpers * Commands: route ACP sessions through ACP runtime in agent command * ACP: require persisted metadata for runtime spawns * Sessions: preserve ACP metadata when updating entries * Plugins: harden ACP backend registry across loaders * ACPX: make availability probe compatible with adapters * E2E: add manual Discord ACP plain-language smoke script * ACPX: preserve streamed spacing across Discord delivery * Docs: add ACP Discord streaming strategy * ACP: harden Discord stream buffering for thread replies * ACP: reuse shared block reply pipeline for projector * ACP: unify streaming config and adopt coalesceIdleMs * Docs: add temporary ACP production hardening plan * Docs: trim temporary ACP hardening plan goals * Docs: gate ACP thread controls by backend capabilities * ACP: add capability-gated runtime controls and /acp operator commands * Docs: remove temporary ACP hardening plan * ACP: fix spawn target validation and close cache cleanup * ACP: harden runtime dispatch and recovery paths * ACP: split ACP command/runtime internals and centralize policy * ACP: harden runtime lifecycle, validation, and observability * ACP: surface runtime and backend session IDs in thread bindings * docs: add temp plan for binding-service migration * ACP: migrate thread binding flows to SessionBindingService * ACP: address review feedback and preserve prompt wording * ACPX plugin: pin runtime dependency and prefer bundled CLI * Discord: complete binding-service migration cleanup and restore ACP plan * Docs: add standalone ACP agents guide * ACP: route harness intents to thread-bound ACP sessions * ACP: fix spawn thread routing and queue-owner stall * ACP: harden startup reconciliation and command bypass handling * ACP: fix dispatch bypass type narrowing * ACP: align runtime metadata to agentSessionId * ACP: normalize session identifier handling and labels * ACP: mark thread banner session ids provisional until first reply * ACP: stabilize session identity mapping and startup reconciliation * ACP: add resolved session-id notices and cwd in thread intros * Discord: prefix thread meta notices consistently * Discord: unify ACP/thread meta notices with gear prefix * Discord: split thread persona naming from meta formatting * Extensions: bump acpx plugin dependency to 0.1.9 * Agents: gate ACP prompt guidance behind acp.enabled * Docs: remove temp experiment plan docs * Docs: scope streaming plan to holy grail refactor * Docs: refactor ACP agents guide for human-first flow * Docs/Skill: add ACP feature-flag guidance and direct acpx telephone-game flow * Docs/Skill: add OpenCode and Pi to ACP harness lists * Docs/Skill: align ACP harness list with current acpx registry * Dev/Test: move ACP plain-language smoke script and mark as keep * Docs/Skill: reorder ACP harness lists with Pi first * ACP: split control-plane manager into core/types/utils modules * Docs: refresh ACP thread-bound agents plan * ACP: extract dispatch lane and split manager domains * ACP: centralize binding context and remove reverse deps * Infra: unify system message formatting * ACP: centralize error boundaries and session id rendering * ACP: enforce init concurrency cap and strict meta clear * Tests: fix ACP dispatch binding mock typing * Tests: fix Discord thread-binding mock drift and ACP request id * ACP: gate slash bypass and persist cleared overrides * ACPX: await pre-abort cancel before runTurn return * Extension: pin acpx runtime dependency to 0.1.11 * Docs: add pinned acpx install strategy for ACP extension * Extensions/acpx: enforce strict local pinned startup * Extensions/acpx: tighten acp-router install guidance * ACPX: retry runtime test temp-dir cleanup * Extensions/acpx: require proactive ACPX repair for thread spawns * Extensions/acpx: require restart offer after acpx reinstall * extensions/acpx: remove workspace protocol devDependency * extensions/acpx: bump pinned acpx to 0.1.13 * extensions/acpx: sync lockfile after dependency bump * ACPX: make runtime spawn Windows-safe * fix: align doctor-config-flow repair tests with default-account migration (#23580) (thanks @osolmaz)
2026-02-26 11:00:09 +01:00
acpEnabled: params.config?.acp?.enabled !== false,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
runtimeInfo,
2026-01-25 03:20:09 +00:00
reactionGuidance,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
messageToolHints,
sandboxInfo,
tools,
modelAliasLines: buildModelAliasLines(params.config),
userTimezone,
userTime,
userTimeFormat,
contextFiles,
memoryCitationsMode: params.config?.memory?.citations,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
});
const systemPromptOverride = createSystemPromptOverride(appendPrompt);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds (#46889) * feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds The hardcoded 5-minute (300s) compaction timeout causes large sessions to enter a death spiral where compaction repeatedly fails and the session grows indefinitely. This adds agents.defaults.compaction.timeoutSeconds to allow operators to override the compaction safety timeout. Default raised to 900s (15min) which is sufficient for sessions up to ~400k tokens. The resolved timeout is also used for the session write lock duration so locks don't expire before compaction completes. Fixes #38233 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add resolveCompactionTimeoutMs tests Cover config resolution edge cases: undefined config, missing compaction section, valid seconds, fractional values, zero, negative, NaN, and Infinity. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add timeoutSeconds to compaction Zod schema The compaction object schema uses .strict(), so setting the new timeoutSeconds config option would fail validation at startup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: enforce integer constraint on compaction timeoutSeconds schema Prevents sub-second values like 0.5 which would floor to 0ms and cause immediate compaction timeout. Matches pattern of other integer timeout fields in the schema. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: clamp compaction timeout to Node timer-safe maximum Values above ~2.1B ms overflow Node's setTimeout to 1ms, causing immediate timeout. Clamp to MAX_SAFE_TIMEOUT_MS matching the pattern in agents/timeout.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add FIELD_LABELS entry for compaction timeoutSeconds Maintains label/help parity invariant enforced by schema.help.quality.test.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: align compaction timeouts with abort handling * fix: land compaction timeout handling (#46889) (thanks @asyncjason) --------- Co-authored-by: Jason Separovic <jason@wilma.dog> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-14 23:34:48 -07:00
const compactionTimeoutMs = resolveCompactionTimeoutMs(params.config);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const sessionLock = await acquireSessionWriteLock({
sessionFile: params.sessionFile,
maxHoldMs: resolveSessionLockMaxHoldFromTimeout({
feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds (#46889) * feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds The hardcoded 5-minute (300s) compaction timeout causes large sessions to enter a death spiral where compaction repeatedly fails and the session grows indefinitely. This adds agents.defaults.compaction.timeoutSeconds to allow operators to override the compaction safety timeout. Default raised to 900s (15min) which is sufficient for sessions up to ~400k tokens. The resolved timeout is also used for the session write lock duration so locks don't expire before compaction completes. Fixes #38233 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add resolveCompactionTimeoutMs tests Cover config resolution edge cases: undefined config, missing compaction section, valid seconds, fractional values, zero, negative, NaN, and Infinity. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add timeoutSeconds to compaction Zod schema The compaction object schema uses .strict(), so setting the new timeoutSeconds config option would fail validation at startup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: enforce integer constraint on compaction timeoutSeconds schema Prevents sub-second values like 0.5 which would floor to 0ms and cause immediate compaction timeout. Matches pattern of other integer timeout fields in the schema. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: clamp compaction timeout to Node timer-safe maximum Values above ~2.1B ms overflow Node's setTimeout to 1ms, causing immediate timeout. Clamp to MAX_SAFE_TIMEOUT_MS matching the pattern in agents/timeout.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add FIELD_LABELS entry for compaction timeoutSeconds Maintains label/help parity invariant enforced by schema.help.quality.test.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: align compaction timeouts with abort handling * fix: land compaction timeout handling (#46889) (thanks @asyncjason) --------- Co-authored-by: Jason Separovic <jason@wilma.dog> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-14 23:34:48 -07:00
timeoutMs: compactionTimeoutMs,
}),
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
});
try {
await repairSessionFileIfNeeded({
sessionFile: params.sessionFile,
warn: (message) => log.warn(message),
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
await prewarmSessionFile(params.sessionFile);
const transcriptPolicy = resolveTranscriptPolicy({
modelApi: model.api,
provider,
modelId,
});
const sessionManager = guardSessionManager(SessionManager.open(params.sessionFile), {
agentId: sessionAgentId,
sessionKey: params.sessionKey,
allowSyntheticToolResults: transcriptPolicy.allowSyntheticToolResults,
allowedToolNames,
2026-01-14 01:08:15 +00:00
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
trackSessionManagerAccess(params.sessionFile);
const settingsManager = createPreparedEmbeddedPiSettingsManager({
cwd: effectiveWorkspace,
agentDir,
cfg: params.config,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
});
// Sets compaction/pruning runtime state and returns extension factories
// that must be passed to the resource loader for the safeguard to be active.
const extensionFactories = buildEmbeddedExtensionFactories({
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
cfg: params.config,
sessionManager,
provider,
modelId,
model,
2026-01-14 01:08:15 +00:00
});
// Only create an explicit resource loader when there are extension factories
// to register; otherwise let createAgentSession use its built-in default.
let resourceLoader: DefaultResourceLoader | undefined;
if (extensionFactories.length > 0) {
resourceLoader = new DefaultResourceLoader({
cwd: resolvedWorkspace,
agentDir,
settingsManager,
extensionFactories,
});
await resourceLoader.reload();
}
2026-01-14 01:08:15 +00:00
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const { builtInTools, customTools } = splitSdkTools({
tools,
sandboxEnabled: !!sandbox?.enabled,
});
2026-01-14 01:08:15 +00:00
const { session } = await createAgentSession({
cwd: effectiveWorkspace,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
agentDir,
authStorage,
modelRegistry,
model: effectiveModel,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
thinkingLevel: mapThinkingLevel(params.thinkLevel),
tools: builtInTools,
customTools,
sessionManager,
settingsManager,
resourceLoader,
});
applySystemPromptOverrideToSession(session, systemPromptOverride());
if (model.api === "ollama") {
const providerBaseUrl =
typeof params.config?.models?.providers?.[model.provider]?.baseUrl === "string"
? params.config.models.providers[model.provider]?.baseUrl
: undefined;
ensureCustomApiRegistered(
model.api,
createConfiguredOllamaStreamFn({
model,
providerBaseUrl,
}),
);
}
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
try {
const prior = await sanitizeSessionHistory({
messages: session.messages,
modelApi: model.api,
2026-01-14 01:08:15 +00:00
modelId,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
provider,
allowedToolNames,
config: params.config,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
sessionManager,
sessionId: params.sessionId,
policy: transcriptPolicy,
2026-01-14 01:08:15 +00:00
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
const validatedGemini = transcriptPolicy.validateGeminiTurns
? validateGeminiTurns(prior)
: prior;
const validated = transcriptPolicy.validateAnthropicTurns
? validateAnthropicTurns(validatedGemini)
: validatedGemini;
// Apply validated transcript to the live session even when no history limit is configured,
// so compaction and hook metrics are based on the same message set.
session.agent.replaceMessages(validated);
// "Original" compaction metrics should describe the validated transcript that enters
// limiting/compaction, not the raw on-disk session snapshot.
const originalMessages = session.messages.slice();
const truncated = limitHistoryTurns(
session.messages,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
getDmHistoryLimitFromSessionKey(params.sessionKey, params.config),
2026-01-14 01:08:15 +00:00
);
// Re-run tool_use/tool_result pairing repair after truncation, since
// limitHistoryTurns can orphan tool_result blocks by removing the
// assistant message that contained the matching tool_use.
const limited = transcriptPolicy.repairToolUseResultPairing
? sanitizeToolUseResultPairing(truncated)
: truncated;
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
if (limited.length > 0) {
session.agent.replaceMessages(limited);
2026-01-16 20:16:35 +00:00
}
const missingSessionKey = !params.sessionKey || !params.sessionKey.trim();
const hookSessionKey = params.sessionKey?.trim() || params.sessionId;
Plugin API: compaction/reset hooks, bootstrap file globs, memory plugin status (#13287) * feat: add before_compaction and before_reset plugin hooks with session context - Pass session messages to before_compaction hook - Add before_reset plugin hook for /new and /reset commands - Add sessionId to plugin hook agent context * feat: extraBootstrapFiles config with glob pattern support Add extraBootstrapFiles to agent defaults config, allowing glob patterns (e.g. "projects/*/TOOLS.md") to auto-load project-level bootstrap files into agent context every turn. Missing files silently skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(status): show custom memory plugins as enabled, not unavailable The status command probes memory availability using the built-in memory-core manager. Custom memory plugins (e.g. via plugin slot) can't be probed this way, so they incorrectly showed "unavailable". Now they show "enabled (plugin X)" without the misleading label. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: use async fs.glob and capture pre-compaction messages - Replace globSync (node:fs) with fs.glob (node:fs/promises) to match codebase conventions for async file operations - Capture session.messages BEFORE replaceMessages(limited) so before_compaction hook receives the full conversation history, not the already-truncated list * fix: resolve lint errors from CI (oxlint strict mode) - Add void to fire-and-forget IIFE (no-floating-promises) - Use String() for unknown catch params in template literals - Add curly braces to single-statement if (curly rule) * fix: resolve remaining CI lint errors in workspace.ts - Remove `| string` from WorkspaceBootstrapFileName union (made all typeof members redundant per no-redundant-type-constituents) - Use type assertion for extra bootstrap file names - Drop redundant await on fs.glob() AsyncIterable (await-thenable) * fix: address Greptile review — path traversal guard + fs/promises import - workspace.ts: use path.resolve() + traversal check in loadExtraBootstrapFiles() - commands-core.ts: import fs from node:fs/promises, drop fs.promises prefix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve symlinks before workspace boundary check Greptile correctly identified that symlinks inside the workspace could point to files outside it, bypassing the path prefix check. Now uses fs.realpath() to resolve symlinks before verifying the real path stays within the workspace boundary. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Greptile review — hook reliability and type safety 1. before_compaction: add compactingCount field so plugins know both the full pre-compaction message count and the truncated count being fed to the compaction LLM. Clarify semantics in comment. 2. loadExtraBootstrapFiles: use path.basename() for the name field so "projects/quaid/TOOLS.md" maps to the known "TOOLS.md" type instead of an invalid WorkspaceBootstrapFileName cast. 3. before_reset: fire the hook even when no session file exists. Previously, short sessions without a persisted file would silently skip the hook. Now fires with empty messages array so plugins always know a reset occurred. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate bootstrap filenames and add compaction hook timeout - Only load extra bootstrap files whose basename matches a recognized workspace filename (AGENTS.md, TOOLS.md, etc.), preventing arbitrary files from being injected into agent context. - Wrap before_compaction hook in a 30-second Promise.race timeout so misbehaving plugins cannot stall the compaction pipeline. - Clarify hook comments: before_compaction is intentionally awaited (plugins need messages before they're discarded) but bounded. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: make before_compaction non-blocking, add sessionFile to after_compaction - before_compaction is now true fire-and-forget — no await, no timeout. Plugins that need full conversation data should persist it themselves and return quickly, or use after_compaction for async processing. - after_compaction now includes sessionFile path so plugins can read the full JSONL transcript asynchronously. All pre-compaction messages are preserved on disk, eliminating the need to block compaction. - Removes Promise.race timeout pattern that didn't actually cancel slow hooks (just raced past them while they continued running). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add sessionFile to before_compaction for parallel processing The session JSONL already has all messages on disk before compaction starts. By providing sessionFile in before_compaction, plugins can read and extract data in parallel with the compaction LLM call rather than waiting for after_compaction. This is the optimal path for memory plugins that need the full conversation history. sessionFile is also kept on after_compaction for plugins that only need to act after compaction completes (analytics, cleanup, etc.). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: move bootstrap extras into bundled hook --------- Co-authored-by: Solomon Steadman <solstead@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Clawdbot <clawdbot@alfie.local> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:45:45 +07:00
const hookRunner = getGlobalHookRunner();
const observedTokenCount = normalizeObservedTokenCount(params.currentTokenCount);
const messageCountOriginal = originalMessages.length;
let tokenCountOriginal: number | undefined;
try {
tokenCountOriginal = 0;
for (const message of originalMessages) {
tokenCountOriginal += estimateTokens(message);
}
} catch {
tokenCountOriginal = undefined;
}
const messageCountBefore = session.messages.length;
let tokenCountBefore = observedTokenCount;
if (tokenCountBefore === undefined) {
try {
tokenCountBefore = 0;
for (const message of session.messages) {
tokenCountBefore += estimateTokens(message);
}
} catch {
tokenCountBefore = undefined;
}
}
// TODO(#7175): Consider exposing full message snapshots or pre-compaction injection
// hooks; current events only report counts/metadata.
try {
const hookEvent = createInternalHookEvent("session", "compact:before", hookSessionKey, {
sessionId: params.sessionId,
missingSessionKey,
messageCount: messageCountBefore,
tokenCount: tokenCountBefore,
messageCountOriginal,
tokenCountOriginal,
});
await triggerInternalHook(hookEvent);
} catch (err) {
log.warn("session:compact:before hook failed", {
errorMessage: err instanceof Error ? err.message : String(err),
errorStack: err instanceof Error ? err.stack : undefined,
});
}
Plugin API: compaction/reset hooks, bootstrap file globs, memory plugin status (#13287) * feat: add before_compaction and before_reset plugin hooks with session context - Pass session messages to before_compaction hook - Add before_reset plugin hook for /new and /reset commands - Add sessionId to plugin hook agent context * feat: extraBootstrapFiles config with glob pattern support Add extraBootstrapFiles to agent defaults config, allowing glob patterns (e.g. "projects/*/TOOLS.md") to auto-load project-level bootstrap files into agent context every turn. Missing files silently skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(status): show custom memory plugins as enabled, not unavailable The status command probes memory availability using the built-in memory-core manager. Custom memory plugins (e.g. via plugin slot) can't be probed this way, so they incorrectly showed "unavailable". Now they show "enabled (plugin X)" without the misleading label. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: use async fs.glob and capture pre-compaction messages - Replace globSync (node:fs) with fs.glob (node:fs/promises) to match codebase conventions for async file operations - Capture session.messages BEFORE replaceMessages(limited) so before_compaction hook receives the full conversation history, not the already-truncated list * fix: resolve lint errors from CI (oxlint strict mode) - Add void to fire-and-forget IIFE (no-floating-promises) - Use String() for unknown catch params in template literals - Add curly braces to single-statement if (curly rule) * fix: resolve remaining CI lint errors in workspace.ts - Remove `| string` from WorkspaceBootstrapFileName union (made all typeof members redundant per no-redundant-type-constituents) - Use type assertion for extra bootstrap file names - Drop redundant await on fs.glob() AsyncIterable (await-thenable) * fix: address Greptile review — path traversal guard + fs/promises import - workspace.ts: use path.resolve() + traversal check in loadExtraBootstrapFiles() - commands-core.ts: import fs from node:fs/promises, drop fs.promises prefix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve symlinks before workspace boundary check Greptile correctly identified that symlinks inside the workspace could point to files outside it, bypassing the path prefix check. Now uses fs.realpath() to resolve symlinks before verifying the real path stays within the workspace boundary. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Greptile review — hook reliability and type safety 1. before_compaction: add compactingCount field so plugins know both the full pre-compaction message count and the truncated count being fed to the compaction LLM. Clarify semantics in comment. 2. loadExtraBootstrapFiles: use path.basename() for the name field so "projects/quaid/TOOLS.md" maps to the known "TOOLS.md" type instead of an invalid WorkspaceBootstrapFileName cast. 3. before_reset: fire the hook even when no session file exists. Previously, short sessions without a persisted file would silently skip the hook. Now fires with empty messages array so plugins always know a reset occurred. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate bootstrap filenames and add compaction hook timeout - Only load extra bootstrap files whose basename matches a recognized workspace filename (AGENTS.md, TOOLS.md, etc.), preventing arbitrary files from being injected into agent context. - Wrap before_compaction hook in a 30-second Promise.race timeout so misbehaving plugins cannot stall the compaction pipeline. - Clarify hook comments: before_compaction is intentionally awaited (plugins need messages before they're discarded) but bounded. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: make before_compaction non-blocking, add sessionFile to after_compaction - before_compaction is now true fire-and-forget — no await, no timeout. Plugins that need full conversation data should persist it themselves and return quickly, or use after_compaction for async processing. - after_compaction now includes sessionFile path so plugins can read the full JSONL transcript asynchronously. All pre-compaction messages are preserved on disk, eliminating the need to block compaction. - Removes Promise.race timeout pattern that didn't actually cancel slow hooks (just raced past them while they continued running). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add sessionFile to before_compaction for parallel processing The session JSONL already has all messages on disk before compaction starts. By providing sessionFile in before_compaction, plugins can read and extract data in parallel with the compaction LLM call rather than waiting for after_compaction. This is the optimal path for memory plugins that need the full conversation history. sessionFile is also kept on after_compaction for plugins that only need to act after compaction completes (analytics, cleanup, etc.). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: move bootstrap extras into bundled hook --------- Co-authored-by: Solomon Steadman <solstead@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Clawdbot <clawdbot@alfie.local> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:45:45 +07:00
if (hookRunner?.hasHooks("before_compaction")) {
try {
await hookRunner.runBeforeCompaction(
{
messageCount: messageCountBefore,
tokenCount: tokenCountBefore,
},
Plugin API: compaction/reset hooks, bootstrap file globs, memory plugin status (#13287) * feat: add before_compaction and before_reset plugin hooks with session context - Pass session messages to before_compaction hook - Add before_reset plugin hook for /new and /reset commands - Add sessionId to plugin hook agent context * feat: extraBootstrapFiles config with glob pattern support Add extraBootstrapFiles to agent defaults config, allowing glob patterns (e.g. "projects/*/TOOLS.md") to auto-load project-level bootstrap files into agent context every turn. Missing files silently skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(status): show custom memory plugins as enabled, not unavailable The status command probes memory availability using the built-in memory-core manager. Custom memory plugins (e.g. via plugin slot) can't be probed this way, so they incorrectly showed "unavailable". Now they show "enabled (plugin X)" without the misleading label. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: use async fs.glob and capture pre-compaction messages - Replace globSync (node:fs) with fs.glob (node:fs/promises) to match codebase conventions for async file operations - Capture session.messages BEFORE replaceMessages(limited) so before_compaction hook receives the full conversation history, not the already-truncated list * fix: resolve lint errors from CI (oxlint strict mode) - Add void to fire-and-forget IIFE (no-floating-promises) - Use String() for unknown catch params in template literals - Add curly braces to single-statement if (curly rule) * fix: resolve remaining CI lint errors in workspace.ts - Remove `| string` from WorkspaceBootstrapFileName union (made all typeof members redundant per no-redundant-type-constituents) - Use type assertion for extra bootstrap file names - Drop redundant await on fs.glob() AsyncIterable (await-thenable) * fix: address Greptile review — path traversal guard + fs/promises import - workspace.ts: use path.resolve() + traversal check in loadExtraBootstrapFiles() - commands-core.ts: import fs from node:fs/promises, drop fs.promises prefix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve symlinks before workspace boundary check Greptile correctly identified that symlinks inside the workspace could point to files outside it, bypassing the path prefix check. Now uses fs.realpath() to resolve symlinks before verifying the real path stays within the workspace boundary. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Greptile review — hook reliability and type safety 1. before_compaction: add compactingCount field so plugins know both the full pre-compaction message count and the truncated count being fed to the compaction LLM. Clarify semantics in comment. 2. loadExtraBootstrapFiles: use path.basename() for the name field so "projects/quaid/TOOLS.md" maps to the known "TOOLS.md" type instead of an invalid WorkspaceBootstrapFileName cast. 3. before_reset: fire the hook even when no session file exists. Previously, short sessions without a persisted file would silently skip the hook. Now fires with empty messages array so plugins always know a reset occurred. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate bootstrap filenames and add compaction hook timeout - Only load extra bootstrap files whose basename matches a recognized workspace filename (AGENTS.md, TOOLS.md, etc.), preventing arbitrary files from being injected into agent context. - Wrap before_compaction hook in a 30-second Promise.race timeout so misbehaving plugins cannot stall the compaction pipeline. - Clarify hook comments: before_compaction is intentionally awaited (plugins need messages before they're discarded) but bounded. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: make before_compaction non-blocking, add sessionFile to after_compaction - before_compaction is now true fire-and-forget — no await, no timeout. Plugins that need full conversation data should persist it themselves and return quickly, or use after_compaction for async processing. - after_compaction now includes sessionFile path so plugins can read the full JSONL transcript asynchronously. All pre-compaction messages are preserved on disk, eliminating the need to block compaction. - Removes Promise.race timeout pattern that didn't actually cancel slow hooks (just raced past them while they continued running). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add sessionFile to before_compaction for parallel processing The session JSONL already has all messages on disk before compaction starts. By providing sessionFile in before_compaction, plugins can read and extract data in parallel with the compaction LLM call rather than waiting for after_compaction. This is the optimal path for memory plugins that need the full conversation history. sessionFile is also kept on after_compaction for plugins that only need to act after compaction completes (analytics, cleanup, etc.). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: move bootstrap extras into bundled hook --------- Co-authored-by: Solomon Steadman <solstead@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Clawdbot <clawdbot@alfie.local> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:45:45 +07:00
{
sessionId: params.sessionId,
agentId: sessionAgentId,
sessionKey: hookSessionKey,
workspaceDir: effectiveWorkspace,
messageProvider: resolvedMessageProvider,
Plugin API: compaction/reset hooks, bootstrap file globs, memory plugin status (#13287) * feat: add before_compaction and before_reset plugin hooks with session context - Pass session messages to before_compaction hook - Add before_reset plugin hook for /new and /reset commands - Add sessionId to plugin hook agent context * feat: extraBootstrapFiles config with glob pattern support Add extraBootstrapFiles to agent defaults config, allowing glob patterns (e.g. "projects/*/TOOLS.md") to auto-load project-level bootstrap files into agent context every turn. Missing files silently skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(status): show custom memory plugins as enabled, not unavailable The status command probes memory availability using the built-in memory-core manager. Custom memory plugins (e.g. via plugin slot) can't be probed this way, so they incorrectly showed "unavailable". Now they show "enabled (plugin X)" without the misleading label. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: use async fs.glob and capture pre-compaction messages - Replace globSync (node:fs) with fs.glob (node:fs/promises) to match codebase conventions for async file operations - Capture session.messages BEFORE replaceMessages(limited) so before_compaction hook receives the full conversation history, not the already-truncated list * fix: resolve lint errors from CI (oxlint strict mode) - Add void to fire-and-forget IIFE (no-floating-promises) - Use String() for unknown catch params in template literals - Add curly braces to single-statement if (curly rule) * fix: resolve remaining CI lint errors in workspace.ts - Remove `| string` from WorkspaceBootstrapFileName union (made all typeof members redundant per no-redundant-type-constituents) - Use type assertion for extra bootstrap file names - Drop redundant await on fs.glob() AsyncIterable (await-thenable) * fix: address Greptile review — path traversal guard + fs/promises import - workspace.ts: use path.resolve() + traversal check in loadExtraBootstrapFiles() - commands-core.ts: import fs from node:fs/promises, drop fs.promises prefix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve symlinks before workspace boundary check Greptile correctly identified that symlinks inside the workspace could point to files outside it, bypassing the path prefix check. Now uses fs.realpath() to resolve symlinks before verifying the real path stays within the workspace boundary. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Greptile review — hook reliability and type safety 1. before_compaction: add compactingCount field so plugins know both the full pre-compaction message count and the truncated count being fed to the compaction LLM. Clarify semantics in comment. 2. loadExtraBootstrapFiles: use path.basename() for the name field so "projects/quaid/TOOLS.md" maps to the known "TOOLS.md" type instead of an invalid WorkspaceBootstrapFileName cast. 3. before_reset: fire the hook even when no session file exists. Previously, short sessions without a persisted file would silently skip the hook. Now fires with empty messages array so plugins always know a reset occurred. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate bootstrap filenames and add compaction hook timeout - Only load extra bootstrap files whose basename matches a recognized workspace filename (AGENTS.md, TOOLS.md, etc.), preventing arbitrary files from being injected into agent context. - Wrap before_compaction hook in a 30-second Promise.race timeout so misbehaving plugins cannot stall the compaction pipeline. - Clarify hook comments: before_compaction is intentionally awaited (plugins need messages before they're discarded) but bounded. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: make before_compaction non-blocking, add sessionFile to after_compaction - before_compaction is now true fire-and-forget — no await, no timeout. Plugins that need full conversation data should persist it themselves and return quickly, or use after_compaction for async processing. - after_compaction now includes sessionFile path so plugins can read the full JSONL transcript asynchronously. All pre-compaction messages are preserved on disk, eliminating the need to block compaction. - Removes Promise.race timeout pattern that didn't actually cancel slow hooks (just raced past them while they continued running). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add sessionFile to before_compaction for parallel processing The session JSONL already has all messages on disk before compaction starts. By providing sessionFile in before_compaction, plugins can read and extract data in parallel with the compaction LLM call rather than waiting for after_compaction. This is the optimal path for memory plugins that need the full conversation history. sessionFile is also kept on after_compaction for plugins that only need to act after compaction completes (analytics, cleanup, etc.). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: move bootstrap extras into bundled hook --------- Co-authored-by: Solomon Steadman <solstead@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Clawdbot <clawdbot@alfie.local> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:45:45 +07:00
},
);
} catch (err) {
log.warn("before_compaction hook failed", {
errorMessage: err instanceof Error ? err.message : String(err),
errorStack: err instanceof Error ? err.stack : undefined,
Plugin API: compaction/reset hooks, bootstrap file globs, memory plugin status (#13287) * feat: add before_compaction and before_reset plugin hooks with session context - Pass session messages to before_compaction hook - Add before_reset plugin hook for /new and /reset commands - Add sessionId to plugin hook agent context * feat: extraBootstrapFiles config with glob pattern support Add extraBootstrapFiles to agent defaults config, allowing glob patterns (e.g. "projects/*/TOOLS.md") to auto-load project-level bootstrap files into agent context every turn. Missing files silently skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(status): show custom memory plugins as enabled, not unavailable The status command probes memory availability using the built-in memory-core manager. Custom memory plugins (e.g. via plugin slot) can't be probed this way, so they incorrectly showed "unavailable". Now they show "enabled (plugin X)" without the misleading label. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: use async fs.glob and capture pre-compaction messages - Replace globSync (node:fs) with fs.glob (node:fs/promises) to match codebase conventions for async file operations - Capture session.messages BEFORE replaceMessages(limited) so before_compaction hook receives the full conversation history, not the already-truncated list * fix: resolve lint errors from CI (oxlint strict mode) - Add void to fire-and-forget IIFE (no-floating-promises) - Use String() for unknown catch params in template literals - Add curly braces to single-statement if (curly rule) * fix: resolve remaining CI lint errors in workspace.ts - Remove `| string` from WorkspaceBootstrapFileName union (made all typeof members redundant per no-redundant-type-constituents) - Use type assertion for extra bootstrap file names - Drop redundant await on fs.glob() AsyncIterable (await-thenable) * fix: address Greptile review — path traversal guard + fs/promises import - workspace.ts: use path.resolve() + traversal check in loadExtraBootstrapFiles() - commands-core.ts: import fs from node:fs/promises, drop fs.promises prefix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve symlinks before workspace boundary check Greptile correctly identified that symlinks inside the workspace could point to files outside it, bypassing the path prefix check. Now uses fs.realpath() to resolve symlinks before verifying the real path stays within the workspace boundary. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Greptile review — hook reliability and type safety 1. before_compaction: add compactingCount field so plugins know both the full pre-compaction message count and the truncated count being fed to the compaction LLM. Clarify semantics in comment. 2. loadExtraBootstrapFiles: use path.basename() for the name field so "projects/quaid/TOOLS.md" maps to the known "TOOLS.md" type instead of an invalid WorkspaceBootstrapFileName cast. 3. before_reset: fire the hook even when no session file exists. Previously, short sessions without a persisted file would silently skip the hook. Now fires with empty messages array so plugins always know a reset occurred. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate bootstrap filenames and add compaction hook timeout - Only load extra bootstrap files whose basename matches a recognized workspace filename (AGENTS.md, TOOLS.md, etc.), preventing arbitrary files from being injected into agent context. - Wrap before_compaction hook in a 30-second Promise.race timeout so misbehaving plugins cannot stall the compaction pipeline. - Clarify hook comments: before_compaction is intentionally awaited (plugins need messages before they're discarded) but bounded. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: make before_compaction non-blocking, add sessionFile to after_compaction - before_compaction is now true fire-and-forget — no await, no timeout. Plugins that need full conversation data should persist it themselves and return quickly, or use after_compaction for async processing. - after_compaction now includes sessionFile path so plugins can read the full JSONL transcript asynchronously. All pre-compaction messages are preserved on disk, eliminating the need to block compaction. - Removes Promise.race timeout pattern that didn't actually cancel slow hooks (just raced past them while they continued running). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add sessionFile to before_compaction for parallel processing The session JSONL already has all messages on disk before compaction starts. By providing sessionFile in before_compaction, plugins can read and extract data in parallel with the compaction LLM call rather than waiting for after_compaction. This is the optimal path for memory plugins that need the full conversation history. sessionFile is also kept on after_compaction for plugins that only need to act after compaction completes (analytics, cleanup, etc.). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: move bootstrap extras into bundled hook --------- Co-authored-by: Solomon Steadman <solstead@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Clawdbot <clawdbot@alfie.local> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:45:45 +07:00
});
}
Plugin API: compaction/reset hooks, bootstrap file globs, memory plugin status (#13287) * feat: add before_compaction and before_reset plugin hooks with session context - Pass session messages to before_compaction hook - Add before_reset plugin hook for /new and /reset commands - Add sessionId to plugin hook agent context * feat: extraBootstrapFiles config with glob pattern support Add extraBootstrapFiles to agent defaults config, allowing glob patterns (e.g. "projects/*/TOOLS.md") to auto-load project-level bootstrap files into agent context every turn. Missing files silently skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(status): show custom memory plugins as enabled, not unavailable The status command probes memory availability using the built-in memory-core manager. Custom memory plugins (e.g. via plugin slot) can't be probed this way, so they incorrectly showed "unavailable". Now they show "enabled (plugin X)" without the misleading label. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: use async fs.glob and capture pre-compaction messages - Replace globSync (node:fs) with fs.glob (node:fs/promises) to match codebase conventions for async file operations - Capture session.messages BEFORE replaceMessages(limited) so before_compaction hook receives the full conversation history, not the already-truncated list * fix: resolve lint errors from CI (oxlint strict mode) - Add void to fire-and-forget IIFE (no-floating-promises) - Use String() for unknown catch params in template literals - Add curly braces to single-statement if (curly rule) * fix: resolve remaining CI lint errors in workspace.ts - Remove `| string` from WorkspaceBootstrapFileName union (made all typeof members redundant per no-redundant-type-constituents) - Use type assertion for extra bootstrap file names - Drop redundant await on fs.glob() AsyncIterable (await-thenable) * fix: address Greptile review — path traversal guard + fs/promises import - workspace.ts: use path.resolve() + traversal check in loadExtraBootstrapFiles() - commands-core.ts: import fs from node:fs/promises, drop fs.promises prefix Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve symlinks before workspace boundary check Greptile correctly identified that symlinks inside the workspace could point to files outside it, bypassing the path prefix check. Now uses fs.realpath() to resolve symlinks before verifying the real path stays within the workspace boundary. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Greptile review — hook reliability and type safety 1. before_compaction: add compactingCount field so plugins know both the full pre-compaction message count and the truncated count being fed to the compaction LLM. Clarify semantics in comment. 2. loadExtraBootstrapFiles: use path.basename() for the name field so "projects/quaid/TOOLS.md" maps to the known "TOOLS.md" type instead of an invalid WorkspaceBootstrapFileName cast. 3. before_reset: fire the hook even when no session file exists. Previously, short sessions without a persisted file would silently skip the hook. Now fires with empty messages array so plugins always know a reset occurred. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: validate bootstrap filenames and add compaction hook timeout - Only load extra bootstrap files whose basename matches a recognized workspace filename (AGENTS.md, TOOLS.md, etc.), preventing arbitrary files from being injected into agent context. - Wrap before_compaction hook in a 30-second Promise.race timeout so misbehaving plugins cannot stall the compaction pipeline. - Clarify hook comments: before_compaction is intentionally awaited (plugins need messages before they're discarded) but bounded. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: make before_compaction non-blocking, add sessionFile to after_compaction - before_compaction is now true fire-and-forget — no await, no timeout. Plugins that need full conversation data should persist it themselves and return quickly, or use after_compaction for async processing. - after_compaction now includes sessionFile path so plugins can read the full JSONL transcript asynchronously. All pre-compaction messages are preserved on disk, eliminating the need to block compaction. - Removes Promise.race timeout pattern that didn't actually cancel slow hooks (just raced past them while they continued running). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add sessionFile to before_compaction for parallel processing The session JSONL already has all messages on disk before compaction starts. By providing sessionFile in before_compaction, plugins can read and extract data in parallel with the compaction LLM call rather than waiting for after_compaction. This is the optimal path for memory plugins that need the full conversation history. sessionFile is also kept on after_compaction for plugins that only need to act after compaction completes (analytics, cleanup, etc.). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: move bootstrap extras into bundled hook --------- Co-authored-by: Solomon Steadman <solstead@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Clawdbot <clawdbot@alfie.local> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:45:45 +07:00
}
const diagEnabled = log.isEnabled("debug");
const preMetrics = diagEnabled ? summarizeCompactionMessages(session.messages) : undefined;
if (diagEnabled && preMetrics) {
log.debug(
`[compaction-diag] start runId=${runId} sessionKey=${params.sessionKey ?? params.sessionId} ` +
`diagId=${diagId} trigger=${trigger} provider=${provider}/${modelId} ` +
`attempt=${attempt} maxAttempts=${maxAttempts} ` +
`pre.messages=${preMetrics.messages} pre.historyTextChars=${preMetrics.historyTextChars} ` +
`pre.toolResultChars=${preMetrics.toolResultChars} pre.estTokens=${preMetrics.estTokens ?? "unknown"}`,
);
log.debug(
`[compaction-diag] contributors diagId=${diagId} top=${JSON.stringify(preMetrics.contributors)}`,
);
}
if (!session.messages.some(hasRealConversationContent)) {
log.info(
`[compaction] skipping — no real conversation messages (sessionKey=${params.sessionKey ?? params.sessionId})`,
);
return {
ok: true,
compacted: false,
reason: "no real conversation messages",
};
}
const compactStartedAt = Date.now();
// Measure compactedCount from the original pre-limiting transcript so compaction
// lifecycle metrics represent total reduction through the compaction pipeline.
const messageCountCompactionInput = messageCountOriginal;
// Estimate full session tokens BEFORE compaction (including system prompt,
// bootstrap context, workspace files, and all history). This is needed for
// a correct sanity check — result.tokensBefore only covers the summarizable
// history subset, not the full session.
let fullSessionTokensBefore = 0;
try {
fullSessionTokensBefore = limited.reduce((sum, msg) => sum + estimateTokens(msg), 0);
} catch {
// If token estimation throws on a malformed message, fall back to 0 so
// the sanity check below becomes a no-op instead of crashing compaction.
}
feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds (#46889) * feat: make compaction timeout configurable via agents.defaults.compaction.timeoutSeconds The hardcoded 5-minute (300s) compaction timeout causes large sessions to enter a death spiral where compaction repeatedly fails and the session grows indefinitely. This adds agents.defaults.compaction.timeoutSeconds to allow operators to override the compaction safety timeout. Default raised to 900s (15min) which is sufficient for sessions up to ~400k tokens. The resolved timeout is also used for the session write lock duration so locks don't expire before compaction completes. Fixes #38233 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add resolveCompactionTimeoutMs tests Cover config resolution edge cases: undefined config, missing compaction section, valid seconds, fractional values, zero, negative, NaN, and Infinity. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add timeoutSeconds to compaction Zod schema The compaction object schema uses .strict(), so setting the new timeoutSeconds config option would fail validation at startup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: enforce integer constraint on compaction timeoutSeconds schema Prevents sub-second values like 0.5 which would floor to 0ms and cause immediate compaction timeout. Matches pattern of other integer timeout fields in the schema. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: clamp compaction timeout to Node timer-safe maximum Values above ~2.1B ms overflow Node's setTimeout to 1ms, causing immediate timeout. Clamp to MAX_SAFE_TIMEOUT_MS matching the pattern in agents/timeout.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: add FIELD_LABELS entry for compaction timeoutSeconds Maintains label/help parity invariant enforced by schema.help.quality.test.ts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: align compaction timeouts with abort handling * fix: land compaction timeout handling (#46889) (thanks @asyncjason) --------- Co-authored-by: Jason Separovic <jason@wilma.dog> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-14 23:34:48 -07:00
const result = await compactWithSafetyTimeout(
() => session.compact(params.customInstructions),
compactionTimeoutMs,
{
abortSignal: params.abortSignal,
onCancel: () => {
session.abortCompaction();
},
},
);
await runPostCompactionSideEffects({
config: params.config,
sessionKey: params.sessionKey,
sessionFile: params.sessionFile,
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
// Estimate tokens after compaction by summing token estimates for remaining messages
let tokensAfter: number | undefined;
2026-01-14 01:08:15 +00:00
try {
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
tokensAfter = 0;
for (const message of session.messages) {
tokensAfter += estimateTokens(message);
2026-01-14 01:08:15 +00:00
}
// Sanity check: compare against the best full-session pre-compaction baseline.
// Prefer the provider-observed live count when available; otherwise use the
// heuristic full-session estimate with a 10% margin for counter jitter.
const sanityCheckBaseline = observedTokenCount ?? fullSessionTokensBefore;
if (
sanityCheckBaseline > 0 &&
tokensAfter >
(observedTokenCount !== undefined ? sanityCheckBaseline : sanityCheckBaseline * 1.1)
) {
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
tokensAfter = undefined; // Don't trust the estimate
}
} catch {
// If estimation fails, leave tokensAfter undefined
tokensAfter = undefined;
2026-01-14 01:08:15 +00:00
}
const messageCountAfter = session.messages.length;
const compactedCount = Math.max(0, messageCountCompactionInput - messageCountAfter);
const postMetrics = diagEnabled ? summarizeCompactionMessages(session.messages) : undefined;
if (diagEnabled && preMetrics && postMetrics) {
log.debug(
`[compaction-diag] end runId=${runId} sessionKey=${params.sessionKey ?? params.sessionId} ` +
`diagId=${diagId} trigger=${trigger} provider=${provider}/${modelId} ` +
`attempt=${attempt} maxAttempts=${maxAttempts} outcome=compacted reason=none ` +
`durationMs=${Date.now() - compactStartedAt} retrying=false ` +
`post.messages=${postMetrics.messages} post.historyTextChars=${postMetrics.historyTextChars} ` +
`post.toolResultChars=${postMetrics.toolResultChars} post.estTokens=${postMetrics.estTokens ?? "unknown"} ` +
`delta.messages=${postMetrics.messages - preMetrics.messages} ` +
`delta.historyTextChars=${postMetrics.historyTextChars - preMetrics.historyTextChars} ` +
`delta.toolResultChars=${postMetrics.toolResultChars - preMetrics.toolResultChars} ` +
`delta.estTokens=${typeof preMetrics.estTokens === "number" && typeof postMetrics.estTokens === "number" ? postMetrics.estTokens - preMetrics.estTokens : "unknown"}`,
);
}
// TODO(#9611): Consider exposing compaction summaries or post-compaction injection;
// current events only report summary metadata.
try {
const hookEvent = createInternalHookEvent("session", "compact:after", hookSessionKey, {
sessionId: params.sessionId,
missingSessionKey,
messageCount: messageCountAfter,
tokenCount: tokensAfter,
compactedCount,
summaryLength: typeof result.summary === "string" ? result.summary.length : undefined,
tokensBefore: result.tokensBefore,
tokensAfter,
firstKeptEntryId: result.firstKeptEntryId,
});
await triggerInternalHook(hookEvent);
} catch (err) {
log.warn("session:compact:after hook failed", {
errorMessage: err instanceof Error ? err.message : String(err),
errorStack: err instanceof Error ? err.stack : undefined,
});
}
if (hookRunner?.hasHooks("after_compaction")) {
try {
await hookRunner.runAfterCompaction(
{
messageCount: messageCountAfter,
tokenCount: tokensAfter,
compactedCount,
},
{
sessionId: params.sessionId,
agentId: sessionAgentId,
sessionKey: hookSessionKey,
workspaceDir: effectiveWorkspace,
messageProvider: resolvedMessageProvider,
},
);
} catch (err) {
log.warn("after_compaction hook failed", {
errorMessage: err instanceof Error ? err.message : String(err),
errorStack: err instanceof Error ? err.stack : undefined,
});
}
}
2026-01-14 01:08:15 +00:00
return {
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
ok: true,
compacted: true,
result: {
summary: result.summary,
firstKeptEntryId: result.firstKeptEntryId,
tokensBefore: observedTokenCount ?? result.tokensBefore,
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
tokensAfter,
details: result.details,
},
2026-01-14 01:08:15 +00:00
};
} finally {
fix(agents): wait for agent idle before flushing pending tool results (#13746) * fix(agents): wait for agent idle before flushing pending tool results When pi-agent-core's auto-retry mechanism handles overloaded/rate-limit errors, it resolves waitForRetry() on assistant message receipt — before tool execution completes in the retried agent loop. This causes the attempt's finally block to call flushPendingToolResults() while tools are still executing, inserting synthetic 'missing tool result' errors and causing silent agent failures. The fix adds a waitForIdle() call before the flush to ensure the agent's retry loop (including tool execution) has fully completed. Evidence from real session: tool call and synthetic error were only 53ms apart — the tool never had a chance to execute before being flushed. Root cause is in pi-agent-core's _resolveRetry() firing on message_end instead of agent_end, but this workaround in OpenClaw prevents the symptom without requiring an upstream fix. Fixes #8643 Fixes #13351 Refs #6682, #12595 * test: add tests for tool result flush race condition Validates that: - Real tool results are not replaced by synthetic errors when they arrive in time - Flush correctly inserts synthetic errors for genuinely orphaned tool calls - Flush is a no-op after real tool results have already been received Refs #8643, #13748 * fix(agents): add waitForIdle to all flushPendingToolResults call sites The original fix only covered the main run finally block, but there are two additional call sites that can trigger flushPendingToolResults while tools are still executing: 1. The catch block in attempt.ts (session setup error handler) 2. The finally block in compact.ts (compaction teardown) Both now await agent.waitForIdle() with a 30s timeout before flushing, matching the pattern already applied to the main finally block. Production testing on VPS with debug logging confirmed these additional paths can fire during sub-agent runs, producing spurious synthetic 'missing tool result' errors. * fix(agents): centralize idle-wait flush and clear timeout handle --------- Co-authored-by: Renue Development <dev@renuebyscience.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:35:43 +11:00
await flushPendingToolResultsAfterIdle({
agent: session?.agent,
sessionManager,
clearPendingOnTimeout: true,
fix(agents): wait for agent idle before flushing pending tool results (#13746) * fix(agents): wait for agent idle before flushing pending tool results When pi-agent-core's auto-retry mechanism handles overloaded/rate-limit errors, it resolves waitForRetry() on assistant message receipt — before tool execution completes in the retried agent loop. This causes the attempt's finally block to call flushPendingToolResults() while tools are still executing, inserting synthetic 'missing tool result' errors and causing silent agent failures. The fix adds a waitForIdle() call before the flush to ensure the agent's retry loop (including tool execution) has fully completed. Evidence from real session: tool call and synthetic error were only 53ms apart — the tool never had a chance to execute before being flushed. Root cause is in pi-agent-core's _resolveRetry() firing on message_end instead of agent_end, but this workaround in OpenClaw prevents the symptom without requiring an upstream fix. Fixes #8643 Fixes #13351 Refs #6682, #12595 * test: add tests for tool result flush race condition Validates that: - Real tool results are not replaced by synthetic errors when they arrive in time - Flush correctly inserts synthetic errors for genuinely orphaned tool calls - Flush is a no-op after real tool results have already been received Refs #8643, #13748 * fix(agents): add waitForIdle to all flushPendingToolResults call sites The original fix only covered the main run finally block, but there are two additional call sites that can trigger flushPendingToolResults while tools are still executing: 1. The catch block in attempt.ts (session setup error handler) 2. The finally block in compact.ts (compaction teardown) Both now await agent.waitForIdle() with a 30s timeout before flushing, matching the pattern already applied to the main finally block. Production testing on VPS with debug logging confirmed these additional paths can fire during sub-agent runs, producing spurious synthetic 'missing tool result' errors. * fix(agents): centralize idle-wait flush and clear timeout handle --------- Co-authored-by: Renue Development <dev@renuebyscience.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-02-14 06:35:43 +11:00
});
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
session.dispose();
2026-01-14 01:08:15 +00:00
}
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
} finally {
await sessionLock.release();
}
} catch (err) {
const reason = describeUnknownError(err);
return fail(reason);
fix: auto-compact on context overflow promptError before returning error (#1627) * fix: detect Anthropic 'Request size exceeds model context window' as context overflow Anthropic now returns 'Request size exceeds model context window' instead of the previously detected 'prompt is too long' format. This new error message was not recognized by isContextOverflowError(), causing auto-compaction to NOT trigger. Users would see the raw error twice without any recovery attempt. Changes: - Add 'exceeds model context window' and 'request size exceeds' to isContextOverflowError() detection patterns - Add tests that fail without the fix, verifying both the raw error string and the JSON-wrapped format from Anthropic's API - Add test for formatAssistantErrorText to ensure the friendly 'Context overflow' message is shown instead of the raw error Note: The upstream pi-ai package (@mariozechner/pi-ai) also needs a fix in its OVERFLOW_PATTERNS regex: /exceeds the context window/i should be changed to /exceeds.*context window/i to match both 'the' and 'model' variants for triggering auto-compaction retry. * fix(tests): remove unused imports and helper from test files Remove WorkspaceBootstrapFile references and _makeFile helper that were incorrectly copied from another test file. These caused type errors and were unrelated to the context overflow detection tests. * fix: trigger auto-compaction on context overflow promptError When the LLM rejects a request with a context overflow error that surfaces as a promptError (thrown exception rather than streamed error), the existing auto-compaction in pi-coding-agent never triggers. This happens because the error bypasses the agent's message_end → agent_end → _checkCompaction path. This fix adds a fallback compaction attempt directly in the run loop: - Detects context overflow in promptError (excluding compaction_failure) - Calls compactEmbeddedPiSessionDirect (bypassing lane queues since already in-lane) - Retries the prompt after successful compaction - Limits to one compaction attempt per run to prevent infinite loops Fixes: context overflow errors shown to user without auto-compaction attempt * style: format compact.ts and run.ts with oxfmt * fix: tighten context overflow match (#1627) (thanks @rodrigouroz) --------- Co-authored-by: Claude <claude@anthropic.com> Co-authored-by: Peter Steinberger <steipete@gmail.com>
2026-01-24 19:09:24 -03:00
} finally {
restoreSkillEnv?.();
process.chdir(prevCwd);
}
}
/**
* Compacts a session with lane queueing (session lane + global lane).
* Use this from outside a lane context. If already inside a lane, use
* `compactEmbeddedPiSessionDirect` to avoid deadlocks.
*/
export async function compactEmbeddedPiSession(
params: CompactEmbeddedPiSessionParams,
): Promise<EmbeddedPiCompactResult> {
const sessionLane = resolveSessionLane(params.sessionKey?.trim() || params.sessionId);
const globalLane = resolveGlobalLane(params.lane);
const enqueueGlobal =
params.enqueue ?? ((task, opts) => enqueueCommandInLane(globalLane, task, opts));
return enqueueCommandInLane(sessionLane, () =>
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
enqueueGlobal(async () => {
ensureRuntimePluginsLoaded({
config: params.config,
workspaceDir: params.workspaceDir,
});
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
ensureContextEnginesInitialized();
const contextEngine = await resolveContextEngine(params.config);
try {
// Resolve token budget from model context window so the context engine
// knows the compaction target. The runner's afterTurn path passes this
// automatically, but the /compact command path needs to compute it here.
const ceProvider = (params.provider ?? DEFAULT_PROVIDER).trim() || DEFAULT_PROVIDER;
const ceModelId = (params.model ?? DEFAULT_MODEL).trim() || DEFAULT_MODEL;
const agentDir = params.agentDir ?? resolveOpenClawAgentDir();
fix(openrouter): silently dropped images for new OpenRouter models — runtime capability detection (#45824) * fix: fetch OpenRouter model capabilities at runtime for unknown models When an OpenRouter model is not in the built-in static snapshot from pi-ai, the fallback hardcodes input: ["text"], silently dropping images. Query the OpenRouter API at runtime to detect actual capabilities (image support, reasoning, context window) for models not in the built-in list. Results are cached in memory for 1 hour. On API failure/timeout, falls back to text-only (no regression). * feat(openrouter): add disk cache for OpenRouter model capabilities Persist the OpenRouter model catalog to ~/.openclaw/cache/openrouter-models.json so it survives process restarts. Cache lookup order: 1. In-memory Map (instant) 2. On-disk JSON file (avoids network on restart) 3. OpenRouter API fetch (populates both layers) Also triggers a background refresh when a model is not found in the cache, in case it was newly added to OpenRouter. * refactor(openrouter): remove pre-warm, use pure lazy-load with disk cache - Remove eager ensureOpenRouterModelCache() from run.ts - Remove TTL — model capabilities are stable, no periodic re-fetching - Cache lookup: in-memory → disk → API fetch (only when needed) - API is only called when no cache exists or a model is not found - Disk cache persists across gateway restarts * fix(openrouter): address review feedback - Fix timer leak: move clearTimeout to finally block - Fix modality check: only check input side of "->" separator to avoid matching image-generation models (text->image) - Use resolveStateDir() instead of hardcoded homedir()/.openclaw - Separate cache dir and filename constants - Add utf-8 encoding to writeFileSync for consistency - Add data validation when reading disk cache * ci: retrigger checks * fix: preload unknown OpenRouter model capabilities before resolve * fix: accept top-level OpenRouter max token metadata * fix: update changelog for OpenRouter runtime capability lookup (#45824) (thanks @DJjjjhao) * fix: avoid redundant OpenRouter refetches and preserve suppression guards --------- Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-03-15 14:18:39 +08:00
const { model: ceModel } = await resolveModelAsync(
ceProvider,
ceModelId,
agentDir,
params.config,
);
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
const ceCtxInfo = resolveContextWindowInfo({
cfg: params.config,
provider: ceProvider,
modelId: ceModelId,
modelContextWindow: ceModel?.contextWindow,
defaultTokens: DEFAULT_CONTEXT_TOKENS,
});
// When the context engine owns compaction, its compact() implementation
// bypasses compactEmbeddedPiSessionDirect (which fires the hooks internally).
// Fire before_compaction / after_compaction hooks here so plugin subscribers
// are notified regardless of which engine is active.
const engineOwnsCompaction = contextEngine.info.ownsCompaction === true;
const hookRunner = engineOwnsCompaction ? getGlobalHookRunner() : null;
const hookSessionKey = params.sessionKey?.trim() || params.sessionId;
const { sessionAgentId } = resolveSessionAgentIds({
sessionKey: params.sessionKey,
config: params.config,
});
const resolvedMessageProvider = params.messageChannel ?? params.messageProvider;
const hookCtx = {
sessionId: params.sessionId,
agentId: sessionAgentId,
sessionKey: hookSessionKey,
workspaceDir: resolveUserPath(params.workspaceDir),
messageProvider: resolvedMessageProvider,
};
// Engine-owned compaction doesn't load the transcript at this level, so
// message counts are unavailable. We pass sessionFile so hook subscribers
// can read the transcript themselves if they need exact counts.
if (hookRunner?.hasHooks("before_compaction")) {
try {
await hookRunner.runBeforeCompaction(
{
messageCount: -1,
sessionFile: params.sessionFile,
},
hookCtx,
);
} catch (err) {
log.warn("before_compaction hook failed", {
errorMessage: err instanceof Error ? err.message : String(err),
});
}
}
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
const result = await contextEngine.compact({
sessionId: params.sessionId,
sessionKey: params.sessionKey,
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
sessionFile: params.sessionFile,
tokenBudget: ceCtxInfo.tokens,
currentTokenCount: params.currentTokenCount,
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
customInstructions: params.customInstructions,
force: params.trigger === "manual",
runtimeContext: params as Record<string, unknown>,
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
});
if (engineOwnsCompaction && result.ok && result.compacted) {
await runPostCompactionSideEffects({
config: params.config,
sessionKey: params.sessionKey,
sessionFile: params.sessionFile,
});
}
if (result.ok && result.compacted && hookRunner?.hasHooks("after_compaction")) {
try {
await hookRunner.runAfterCompaction(
{
messageCount: -1,
compactedCount: -1,
tokenCount: result.result?.tokensAfter,
sessionFile: params.sessionFile,
},
hookCtx,
);
} catch (err) {
log.warn("after_compaction hook failed", {
errorMessage: err instanceof Error ? err.message : String(err),
});
}
}
feature(context): extend plugin system to support custom context management (#22201) * feat(context-engine): add ContextEngine interface and registry Introduce the pluggable ContextEngine abstraction that allows external plugins to register custom context management strategies. - ContextEngine interface with lifecycle methods: bootstrap, ingest, ingestBatch, afterTurn, assemble, compact, prepareSubagentSpawn, onSubagentEnded, dispose - Module-level singleton registry with registerContextEngine() and resolveContextEngine() (config-driven slot selection) - LegacyContextEngine: pass-through implementation wrapping existing compaction behavior for 100% backward compatibility - ensureContextEnginesInitialized() guard for safe one-time registration - 19 tests covering contract, registry, resolution, and legacy parity * feat(plugins): add context-engine slot and registerContextEngine API Wire the ContextEngine abstraction into the plugin system so external plugins can register context engines via the standard plugin API. - Add 'context-engine' to PluginKind union type - Add 'contextEngine' slot to PluginSlotsConfig (default: 'legacy') - Wire registerContextEngine() through OpenClawPluginApi - Export ContextEngine types from plugin-sdk for external consumers - Restore proper slot-based resolution in registry * feat(context-engine): wire ContextEngine into agent run lifecycle Integrate the ContextEngine abstraction into the core agent run path: - Resolve context engine once per run (reused across retries) - Bootstrap: hydrate canonical store from session file on first run - Assemble: route context assembly through pluggable engine - Auto-compaction guard: disable built-in auto-compaction when the engine declares ownsCompaction (prevents double-compaction) - AfterTurn: post-turn lifecycle hook for ingest + background compaction decisions - Overflow compaction: route through contextEngine.compact() - Dispose: clean up engine resources in finally block - Notify context engine on subagent lifecycle events Legacy engine: all lifecycle methods are pass-through/no-op, preserving 100% backward compatibility for users without a context engine plugin. * feat(plugins): add scoped subagent methods and gateway request scope Expose runtime.subagent.{run, waitForRun, getSession, deleteSession} so external plugins can spawn sub-agent sessions without raw gateway dispatch access. Uses AsyncLocalStorage request-scope bridge to dispatch internally via handleGatewayRequest with a synthetic operator client. Methods are only available during gateway request handling. - Symbol.for-backed global singleton for cross-module-reload safety - Fallback gateway context for non-WS dispatch paths (Telegram/WhatsApp) - Set gateway request scope for all handlers, not just plugin handlers - 3 staleness tests for fallback context hardening * feat(context-engine): route /compact and sessions.get through context engine Wire the /compact command and sessions.get handler through the pluggable ContextEngine interface. - Thread tokenBudget and force parameters to context engine compact - Route /compact through contextEngine.compact() when registered - Wire sessions.get as runtime alias for plugin subagent dispatch - Add .pebbles/ to .gitignore * style: format with oxfmt 0.33.0 Fix duplicate import (ControlUiRootState in server.impl.ts) and import ordering across all changed files. * fix: update extension test mocks for context-engine types Add missing subagent property to bluebubbles PluginRuntime mock. Add missing registerContextEngine to lobster OpenClawPluginApi mock. * fix(subagents): keep deferred delete cleanup retryable * style: format run attempt for CI * fix(rebase): remove duplicate embedded-run imports * test: add missing gateway context mock export * fix: pass resolved auth profile into afterTurn compaction Ensure the embedded runner forwards resolved auth profile context into legacy context-engine compaction params on the normal afterTurn path, matching overflow compaction behavior. This allows downstream LCM summarization to use the intended provider auth/profile consistently. Also fix strict TS typing in external-link token dedupe and align an attempt unit test reasoningLevel value with the current ReasoningLevel enum. Regeneration-Prompt: | We were debugging context-engine compaction where downstream summary calls were missing the right auth/profile context in normal afterTurn flow, while overflow compaction already propagated it. Preserve current behavior and keep changes additive: thread the resolved authProfileId through run -> attempt -> legacy compaction param builder without broad refactors. Add tests that prove the auth profile is included in afterTurn legacy params and that overflow compaction still passes it through run attempts. Keep existing APIs stable, and only adjust small type issues needed for strict compilation. * fix: remove duplicate imports from rebase * feat: add context-engine system prompt additions * fix(rebase): dedupe attempt import declarations * test: fix fetch mock typing in ollama autodiscovery * fix(test): add registerContextEngine to diffs extension mock APIs * test(windows): use path.delimiter in ios-team-id fixture PATH * test(cron): add model formatting and precedence edge case tests Covers: - Provider/model string splitting (whitespace, nested paths, empty segments) - Provider normalization (casing, aliases like bedrock→amazon-bedrock) - Anthropic model alias normalization (opus-4.5→claude-opus-4-5) - Precedence: job payload > session override > config default - Sequential runs with different providers (CI flake regression pattern) - forceNew session preserving stored model overrides - Whitespace/empty model string edge cases - Config model as string vs object format * test(cron): fix model formatting test config types * test(phone-control): add registerContextEngine to mock API * fix: re-export ChannelKind from config-reload-plan * fix: add subagent mock to plugin-runtime-mock test util * docs: add changelog fragment for context engine PR #22201
2026-03-06 05:31:59 -08:00
return {
ok: result.ok,
compacted: result.compacted,
reason: result.reason,
result: result.result
? {
summary: result.result.summary ?? "",
firstKeptEntryId: result.result.firstKeptEntryId ?? "",
tokensBefore: result.result.tokensBefore,
tokensAfter: result.result.tokensAfter,
details: result.result.details,
}
: undefined,
};
} finally {
await contextEngine.dispose?.();
}
}),
2026-01-14 01:08:15 +00:00
);
}