744 lines
28 KiB
TypeScript
Raw Normal View History

import type { AgentMessage } from "@mariozechner/pi-agent-core";
import {
createAgentSession,
estimateTokens,
SessionManager,
SettingsManager,
} from "@mariozechner/pi-coding-agent";
import fs from "node:fs/promises";
import os from "node:os";
2026-01-14 01:08:15 +00:00
import type { ReasoningLevel, ThinkLevel } from "../../auto-reply/thinking.js";
2026-01-30 03:15:10 +01:00
import type { OpenClawConfig } from "../../config/config.js";
import type { ExecElevatedDefaults } from "../bash-tools.js";
import type { EmbeddedPiCompactResult } from "./types.js";
import { resolveHeartbeatPrompt } from "../../auto-reply/heartbeat.js";
import { resolveChannelCapabilities } from "../../config/channel-capabilities.js";
2026-01-14 01:08:15 +00:00
import { getMachineDisplayName } from "../../infra/machine-name.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 { type enqueueCommand, enqueueCommandInLane } from "../../process/command-queue.js";
import { isSubagentSessionKey } from "../../routing/session-key.js";
import { resolveSignalReactionLevel } from "../../signal/reaction-level.js";
2026-01-16 20:16:35 +00:00
import { resolveTelegramInlineButtonsScope } from "../../telegram/inline-buttons.js";
2026-01-25 03:20:09 +00:00
import { resolveTelegramReactionLevel } from "../../telegram/reaction-level.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";
2026-01-14 01:08:15 +00:00
import { resolveSessionAgentIds } from "../agent-scope.js";
import { makeBootstrapWarn, resolveBootstrapContextForRun } from "../bootstrap-files.js";
import { listChannelSupportedActions, resolveChannelMessageToolHints } from "../channel-tools.js";
import { formatUserTime, resolveUserTimeFormat, resolveUserTimezone } from "../date-time.js";
2026-01-14 01:08:15 +00:00
import { DEFAULT_MODEL, DEFAULT_PROVIDER } from "../defaults.js";
import { resolveOpenClawDocsPath } from "../docs-path.js";
2026-01-14 01:08:15 +00:00
import { getApiKeyForModel, resolveModelAuthMode } from "../model-auth.js";
2026-01-30 03:15:10 +01:00
import { ensureOpenClawModelsJson } from "../models-config.js";
2026-01-14 01:08:15 +00:00
import {
ensureSessionHeader,
validateAnthropicTurns,
validateGeminiTurns,
} from "../pi-embedded-helpers.js";
import {
ensurePiCompactionReserveTokens,
resolveCompactionReserveTokensFloor,
} from "../pi-settings.js";
2026-01-30 03:15:10 +01:00
import { createOpenClawCodingTools } from "../pi-tools.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";
2026-01-14 01:08:15 +00:00
import { acquireSessionWriteLock } from "../session-write-lock.js";
import { detectRuntimeShell } from "../shell-utils.js";
2026-01-14 01:08:15 +00:00
import {
applySkillEnvOverrides,
applySkillEnvOverridesFromSnapshot,
loadWorkspaceSkillEntries,
resolveSkillsPromptForRun,
type SkillSnapshot,
} from "../skills.js";
import { resolveTranscriptPolicy } from "../transcript-policy.js";
import { compactWithSafetyTimeout } from "./compaction-safety-timeout.js";
2026-01-14 01:08:15 +00:00
import { buildEmbeddedExtensionPaths } 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";
import { buildModelAliasLines, resolveModel } from "./model.js";
import { buildEmbeddedSandboxInfo } from "./sandbox-info.js";
import { prewarmSessionFile, trackSessionManagerAccess } from "./session-manager-cache.js";
import {
applySystemPromptOverrideToSession,
buildEmbeddedSystemPrompt,
createSystemPromptOverride,
} from "./system-prompt.js";
2026-01-14 01:08:15 +00:00
import { splitSdkTools } from "./tool-split.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;
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;
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[];
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 createCompactionDiagId(): string {
return `cmp-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 8)}`;
}
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("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";
}
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);
const prevCwd = process.cwd();
const provider = (params.provider ?? DEFAULT_PROVIDER).trim() || DEFAULT_PROVIDER;
const modelId = (params.model ?? DEFAULT_MODEL).trim() || DEFAULT_MODEL;
2026-01-30 03:15:10 +01:00
const agentDir = params.agentDir ?? resolveOpenClawAgentDir();
await ensureOpenClawModelsJson(params.config, agentDir);
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 { model, error, authStorage, modelRegistry } = resolveModel(
provider,
modelId,
agentDir,
params.config,
);
if (!model) {
const reason = error ?? `Unknown model: ${provider}/${modelId}`;
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
};
}
try {
const apiKeyInfo = await getApiKeyForModel({
model,
cfg: params.config,
profileId: params.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 "${model.provider}" (auth mode: ${apiKeyInfo.mode}).`,
);
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 if (model.provider === "github-copilot") {
const { resolveCopilotApiToken } = await import("../../providers/github-copilot-token.js");
const copilotToken = await resolveCopilotApiToken({
githubToken: apiKeyInfo.apiKey,
});
authStorage.setRuntimeApiKey(model.provider, copilotToken.token);
} else {
authStorage.setRuntimeApiKey(model.provider, apiKeyInfo.apiKey);
}
} catch (err) {
const reason = describeUnknownError(err);
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
};
}
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 = !params.skillsSnapshot || !params.skillsSnapshot.resolvedSkills;
const skillEntries = shouldLoadSkillEntries
? loadWorkspaceSkillEntries(effectiveWorkspace)
: [];
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 { contextFiles } = await resolveBootstrapContextForRun({
workspaceDir: effectiveWorkspace,
config: params.config,
sessionKey: params.sessionKey,
sessionId: params.sessionId,
warn: makeBootstrapWarn({ sessionLabel, warn: (message) => log.warn(message) }),
});
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: params.messageChannel ?? params.messageProvider,
agentAccountId: params.agentAccountId,
sessionKey: params.sessionKey ?? params.sessionId,
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,
modelAuthMode: resolveModelAuthMode(model.provider, params.config),
});
const tools = sanitizeToolsForGoogle({ tools: toolsRaw, provider });
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;
const promptMode = isSubagentSessionKey(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 appendPrompt = buildEmbeddedSystemPrompt({
workspaceDir: effectiveWorkspace,
defaultThinkLevel: params.thinkLevel,
reasoningLevel: params.reasoningLevel ?? "off",
extraSystemPrompt: params.extraSystemPrompt,
ownerNumbers: params.ownerNumbers,
reasoningTagHint,
heartbeatPrompt: isDefaultAgent
? resolveHeartbeatPrompt(params.config?.agents?.defaults?.heartbeat?.prompt)
: undefined,
skillsPrompt,
docsPath: docsPath ?? undefined,
ttsHint,
promptMode,
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
const sessionLock = await acquireSessionWriteLock({
sessionFile: params.sessionFile,
});
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,
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 = SettingsManager.create(effectiveWorkspace, agentDir);
ensurePiCompactionReserveTokens({
settingsManager,
minReserveTokens: resolveCompactionReserveTokensFloor(params.config),
});
// Call for side effects (sets compaction/pruning runtime state)
buildEmbeddedExtensionPaths({
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
});
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({
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
cwd: resolvedWorkspace,
agentDir,
authStorage,
modelRegistry,
model,
thinkingLevel: mapThinkingLevel(params.thinkLevel),
tools: builtInTools,
customTools,
sessionManager,
settingsManager,
});
applySystemPromptOverrideToSession(session, systemPromptOverride());
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,
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;
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
// Capture full message history BEFORE limiting — plugins need the complete conversation
const preCompactionMessages = [...session.messages];
const truncated = limitHistoryTurns(
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
validated,
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
}
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
// Run before_compaction hooks (fire-and-forget).
// The session JSONL already contains all messages on disk, so plugins
// can read sessionFile asynchronously and process in parallel with
// the compaction LLM call — no need to block or wait for after_compaction.
const hookRunner = getGlobalHookRunner();
const hookCtx = {
agentId: params.sessionKey?.split(":")[0] ?? "main",
sessionKey: params.sessionKey,
sessionId: params.sessionId,
workspaceDir: params.workspaceDir,
messageProvider: params.messageChannel ?? params.messageProvider,
};
if (hookRunner?.hasHooks("before_compaction")) {
hookRunner
.runBeforeCompaction(
{
messageCount: preCompactionMessages.length,
compactingCount: limited.length,
messages: preCompactionMessages,
sessionFile: params.sessionFile,
},
hookCtx,
)
.catch((hookErr: unknown) => {
log.warn(`before_compaction hook failed: ${String(hookErr)}`);
});
}
const diagEnabled = log.isEnabled("debug");
const preMetrics = diagEnabled ? summarizeCompactionMessages(session.messages) : undefined;
if (diagEnabled && preMetrics) {
log.debug(
`[compaction-diag] start runId=${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)}`,
);
}
const compactStartedAt = Date.now();
const result = await compactWithSafetyTimeout(() =>
session.compact(params.customInstructions),
);
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
}
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
// Sanity check: tokensAfter should be less than tokensBefore
if (tokensAfter > result.tokensBefore) {
tokensAfter = undefined; // Don't trust the estimate
}
} catch {
// If estimation fails, leave tokensAfter undefined
tokensAfter = undefined;
2026-01-14 01:08:15 +00:00
}
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
// Run after_compaction hooks (fire-and-forget).
// Also includes sessionFile for plugins that only need to act after
// compaction completes (e.g. analytics, cleanup).
if (hookRunner?.hasHooks("after_compaction")) {
hookRunner
.runAfterCompaction(
{
messageCount: session.messages.length,
tokenCount: tokensAfter,
compactedCount: limited.length - session.messages.length,
sessionFile: params.sessionFile,
},
hookCtx,
)
.catch((hookErr) => {
log.warn(`after_compaction hook failed: ${hookErr}`);
});
}
const postMetrics = diagEnabled ? summarizeCompactionMessages(session.messages) : undefined;
if (diagEnabled && preMetrics && postMetrics) {
log.debug(
`[compaction-diag] end runId=${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"}`,
);
}
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: result.tokensBefore,
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,
});
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);
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
};
} 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, () =>
enqueueGlobal(async () => compactEmbeddedPiSessionDirect(params)),
2026-01-14 01:08:15 +00:00
);
}