openclaw/src/auto-reply/reply/agent-runner.ts

526 lines
17 KiB
TypeScript
Raw Normal View History

2026-01-04 05:47:21 +01:00
import crypto from "node:crypto";
import fs from "node:fs";
import type { TypingMode } from "../../config/types.js";
import type { OriginatingChannelType, TemplateContext } from "../templating.js";
import type { GetReplyOptions, ReplyPayload } from "../types.js";
import type { TypingController } from "./typing.js";
2026-01-04 05:47:21 +01:00
import { lookupContextTokens } from "../../agents/context.js";
import { DEFAULT_CONTEXT_TOKENS } from "../../agents/defaults.js";
2026-01-09 02:21:17 +00:00
import { resolveModelAuthMode } from "../../agents/model-auth.js";
2026-01-10 23:31:25 +00:00
import { isCliProvider } from "../../agents/model-selection.js";
import { queueEmbeddedPiMessage } from "../../agents/pi-embedded.js";
import { hasNonzeroUsage } from "../../agents/usage.js";
2026-01-04 05:47:21 +01:00
import {
resolveAgentIdFromSessionKey,
resolveSessionFilePath,
resolveSessionTranscriptPath,
2026-01-04 05:47:21 +01:00
type SessionEntry,
updateSessionStore,
updateSessionStoreEntry,
2026-01-04 05:47:21 +01:00
} from "../../config/sessions.js";
import { emitDiagnosticEvent, isDiagnosticsEnabled } from "../../infra/diagnostic-events.js";
2026-01-04 05:47:21 +01:00
import { defaultRuntime } from "../../runtime.js";
import { estimateUsageCost, resolveModelCostConfig } from "../../utils/usage-format.js";
import { resolveResponseUsageMode, type VerboseLevel } from "../thinking.js";
import { runAgentTurnWithFallback } from "./agent-runner-execution.js";
import {
2026-01-17 05:33:27 +00:00
createShouldEmitToolOutput,
createShouldEmitToolResult,
finalizeWithFollowup,
isAudioPayload,
signalTypingIfNeeded,
} from "./agent-runner-helpers.js";
import { runMemoryFlushIfNeeded } from "./agent-runner-memory.js";
import { buildReplyPayloads } from "./agent-runner-payloads.js";
import { appendUsageLine, formatResponseUsageLine } from "./agent-runner-utils.js";
import { createAudioAsVoiceBuffer, createBlockReplyPipeline } from "./block-reply-pipeline.js";
import { resolveBlockStreamingCoalescing } from "./block-streaming.js";
2026-01-04 05:47:21 +01:00
import { createFollowupRunner } from "./followup-runner.js";
import { enqueueFollowupRun, type FollowupRun, type QueueSettings } from "./queue.js";
import { createReplyToModeFilterForChannel, resolveReplyToMode } from "./reply-threading.js";
import { incrementCompactionCount } from "./session-updates.js";
import { persistSessionUsageUpdate } from "./session-usage.js";
import { createTypingSignaler } from "./typing-mode.js";
2026-01-04 05:47:21 +01:00
const BLOCK_REPLY_SEND_TIMEOUT_MS = 15_000;
2026-01-06 22:43:29 +01:00
2026-01-04 05:47:21 +01:00
export async function runReplyAgent(params: {
commandBody: string;
followupRun: FollowupRun;
queueKey: string;
resolvedQueue: QueueSettings;
shouldSteer: boolean;
shouldFollowup: boolean;
isActive: boolean;
isStreaming: boolean;
opts?: GetReplyOptions;
typing: TypingController;
sessionEntry?: SessionEntry;
sessionStore?: Record<string, SessionEntry>;
sessionKey?: string;
storePath?: string;
defaultModel: string;
agentCfgContextTokens?: number;
resolvedVerboseLevel: VerboseLevel;
isNewSession: boolean;
blockStreamingEnabled: boolean;
blockReplyChunking?: {
minChars: number;
maxChars: number;
breakPreference: "paragraph" | "newline" | "sentence";
fix: flush block streaming on paragraph boundaries for chunkMode=newline (#7014) * feat: Implement paragraph boundary flushing in block streaming - Added `flushOnParagraph` option to `BlockReplyChunking` for immediate flushing on paragraph breaks. - Updated `EmbeddedBlockChunker` to handle paragraph boundaries during chunking. - Enhanced `createBlockReplyCoalescer` to support flushing on enqueue. - Added tests to verify behavior of flushing with and without `flushOnEnqueue` set. - Updated relevant types and interfaces to include `flushOnParagraph` and `flushOnEnqueue` options. * fix: Improve streaming behavior and enhance block chunking logic - Resolved issue with stuck typing indicator after streamed BlueBubbles replies. - Refactored `EmbeddedBlockChunker` to streamline fence-split handling and ensure maxChars fallback for newline chunking. - Added tests to validate new chunking behavior, including handling of paragraph breaks and fence scenarios. - Updated changelog to reflect these changes. * test: Add test for clamping long paragraphs in EmbeddedBlockChunker - Introduced a new test case to verify that long paragraphs are correctly clamped to maxChars when flushOnParagraph is enabled. - Updated logic in EmbeddedBlockChunker to handle cases where the next paragraph break exceeds maxChars, ensuring proper chunking behavior. * refactor: streamline logging and improve error handling in message processing - Removed verbose logging statements from the `processMessage` function to reduce clutter. - Enhanced error handling by using `runtime.error` for typing restart failures. - Updated the `applySystemPromptOverrideToSession` function to accept a string directly instead of a function, simplifying the prompt application process. - Adjusted the `runEmbeddedAttempt` function to directly use the system prompt override without invoking it as a function.
2026-02-02 01:22:41 -08:00
flushOnParagraph?: boolean;
2026-01-04 05:47:21 +01:00
};
resolvedBlockStreamingBreak: "text_end" | "message_end";
sessionCtx: TemplateContext;
shouldInjectGroupIntro: boolean;
2026-01-07 21:58:54 +00:00
typingMode: TypingMode;
2026-01-04 05:47:21 +01:00
}): Promise<ReplyPayload | ReplyPayload[] | undefined> {
const {
commandBody,
followupRun,
queueKey,
resolvedQueue,
shouldSteer,
shouldFollowup,
isActive,
isStreaming,
opts,
typing,
sessionEntry,
sessionStore,
sessionKey,
storePath,
defaultModel,
agentCfgContextTokens,
resolvedVerboseLevel,
isNewSession,
blockStreamingEnabled,
blockReplyChunking,
resolvedBlockStreamingBreak,
sessionCtx,
shouldInjectGroupIntro,
2026-01-07 21:58:54 +00:00
typingMode,
2026-01-04 05:47:21 +01:00
} = params;
let activeSessionEntry = sessionEntry;
2026-01-12 00:32:43 +00:00
const activeSessionStore = sessionStore;
let activeIsNewSession = isNewSession;
const isHeartbeat = opts?.isHeartbeat === true;
const typingSignals = createTypingSignaler({
typing,
mode: typingMode,
isHeartbeat,
});
const shouldEmitToolResult = createShouldEmitToolResult({
sessionKey,
storePath,
resolvedVerboseLevel,
});
2026-01-17 05:33:27 +00:00
const shouldEmitToolOutput = createShouldEmitToolOutput({
sessionKey,
storePath,
resolvedVerboseLevel,
});
2026-01-04 05:47:21 +01:00
const pendingToolTasks = new Set<Promise<void>>();
const blockReplyTimeoutMs = opts?.blockReplyTimeoutMs ?? BLOCK_REPLY_SEND_TIMEOUT_MS;
const replyToChannel =
sessionCtx.OriginatingChannel ??
((sessionCtx.Surface ?? sessionCtx.Provider)?.toLowerCase() as
| OriginatingChannelType
| undefined);
const replyToMode = resolveReplyToMode(
followupRun.run.config,
replyToChannel,
Move provider to a plugin-architecture (#661) * refactor: introduce provider plugin registry * refactor: move provider CLI to plugins * docs: add provider plugin implementation notes * refactor: shift provider runtime logic into plugins * refactor: add plugin defaults and summaries * docs: update provider plugin notes * feat(commands): add /commands slash list * Auto-reply: tidy help message * Auto-reply: fix status command lint * Tests: align google shared expectations * Auto-reply: tidy help message * Auto-reply: fix status command lint * refactor: move provider routing into plugins * test: align agent routing expectations * docs: update provider plugin notes * refactor: route replies via provider plugins * docs: note route-reply plugin hooks * refactor: extend provider plugin contract * refactor: derive provider status from plugins * refactor: unify gateway provider control * refactor: use plugin metadata in auto-reply * fix: parenthesize cron target selection * refactor: derive gateway methods from plugins * refactor: generalize provider logout * refactor: route provider logout through plugins * refactor: move WhatsApp web login methods into plugin * refactor: generalize provider log prefixes * refactor: centralize default chat provider * refactor: derive provider lists from registry * refactor: move provider reload noops into plugins * refactor: resolve web login provider via alias * refactor: derive CLI provider options from plugins * refactor: derive prompt provider list from plugins * style: apply biome lint fixes * fix: resolve provider routing edge cases * docs: update provider plugin refactor notes * fix(gateway): harden agent provider routing * refactor: move provider routing into plugins * refactor: move provider CLI to plugins * refactor: derive provider lists from registry * fix: restore slash command parsing * refactor: align provider ids for schema * refactor: unify outbound target resolution * fix: keep outbound labels stable * feat: add msteams to cron surfaces * fix: clean up lint build issues * refactor: localize chat provider alias normalization * refactor: drive gateway provider lists from plugins * docs: update provider plugin notes * style: format message-provider * fix: avoid provider registry init cycles * style: sort message-provider imports * fix: relax provider alias map typing * refactor: move provider routing into plugins * refactor: add plugin pairing/config adapters * refactor: route pairing and provider removal via plugins * refactor: align auto-reply provider typing * test: stabilize telegram media mocks * docs: update provider plugin refactor notes * refactor: pluginize outbound targets * refactor: pluginize provider selection * refactor: generalize text chunk limits * docs: update provider plugin notes * refactor: generalize group session/config * fix: normalize provider id for room detection * fix: avoid provider init in system prompt * style: formatting cleanup * refactor: normalize agent delivery targets * test: update outbound delivery labels * chore: fix lint regressions * refactor: extend provider plugin adapters * refactor: move elevated/block streaming defaults to plugins * refactor: defer outbound send deps to plugins * docs: note plugin-driven streaming/elevated defaults * refactor: centralize webchat provider constant * refactor: add provider setup adapters * refactor: delegate provider add config to plugins * docs: document plugin-driven provider add * refactor: add plugin state/binding metadata * refactor: build agent provider status from plugins * docs: note plugin-driven agent bindings * refactor: centralize internal provider constant usage * fix: normalize WhatsApp targets for groups and E.164 (#631) (thanks @imfing) * refactor: centralize default chat provider * refactor: centralize WhatsApp target normalization * refactor: move provider routing into plugins * refactor: normalize agent delivery targets * chore: fix lint regressions * fix: normalize WhatsApp targets for groups and E.164 (#631) (thanks @imfing) * feat: expand provider plugin adapters * refactor: route auto-reply via provider plugins * fix: align WhatsApp target normalization * fix: normalize WhatsApp targets for groups and E.164 (#631) (thanks @imfing) * refactor: centralize WhatsApp target normalization * feat: add /config chat config updates * docs: add /config get alias * feat(commands): add /commands slash list * refactor: centralize default chat provider * style: apply biome lint fixes * chore: fix lint regressions * fix: clean up whatsapp allowlist typing * style: format config command helpers * refactor: pluginize tool threading context * refactor: normalize session announce targets * docs: note new plugin threading and announce hooks * refactor: pluginize message actions * docs: update provider plugin actions notes * fix: align provider action adapters * refactor: centralize webchat checks * style: format message provider helpers * refactor: move provider onboarding into adapters * docs: note onboarding provider adapters * feat: add msteams onboarding adapter * style: organize onboarding imports * fix: normalize msteams allowFrom types * feat: add plugin text chunk limits * refactor: use plugin chunk limit fallbacks * feat: add provider mention stripping hooks * style: organize provider plugin type imports * refactor: generalize health snapshots * refactor: update macOS health snapshot handling * docs: refresh health snapshot notes * style: format health snapshot updates * refactor: drive security warnings via plugins * docs: note provider security adapter * style: format provider security adapters * refactor: centralize provider account defaults * refactor: type gateway client identity constants * chore: regen gateway protocol swift * fix: degrade health on failed provider probe * refactor: centralize pairing approve hint * docs: add plugin CLI command references * refactor: route auth and tool sends through plugins * docs: expand provider plugin hooks * refactor: document provider docking touchpoints * refactor: normalize internal provider defaults * refactor: streamline outbound delivery wiring * refactor: make provider onboarding plugin-owned * refactor: support provider-owned agent tools * refactor: move telegram draft chunking into telegram module * refactor: infer provider tool sends via extractToolSend * fix: repair plugin onboarding imports * refactor: de-dup outbound target normalization * style: tidy plugin and agent imports * refactor: data-drive provider selection line * fix: satisfy lint after provider plugin rebase * test: deflake gateway-cli coverage * style: format gateway-cli coverage test * refactor(provider-plugins): simplify provider ids * test(pairing-cli): avoid provider-specific ternary * style(macos): swiftformat HealthStore * refactor(sandbox): derive provider tool denylist * fix(sandbox): avoid plugin init in defaults * refactor(provider-plugins): centralize provider aliases * style(test): satisfy biome * refactor(protocol): v3 providers.status maps * refactor(ui): adapt to protocol v3 * refactor(macos): adapt to protocol v3 * test: update providers.status v3 fixtures * refactor(gateway): map provider runtime snapshot * test(gateway): update reload runtime snapshot * refactor(whatsapp): normalize heartbeat provider id * docs(refactor): update provider plugin notes * style: satisfy biome after rebase * fix: describe sandboxed elevated in prompt * feat(gateway): add agent image attachments + live probe * refactor: derive CLI provider options from plugins * fix(gateway): harden agent provider routing * fix(gateway): harden agent provider routing * refactor: align provider ids for schema * fix(protocol): keep agent provider string * fix(gateway): harden agent provider routing * fix(protocol): keep agent provider string * refactor: normalize agent delivery targets * refactor: support provider-owned agent tools * refactor(config): provider-keyed elevated allowFrom * style: satisfy biome * fix(gateway): appease provider narrowing * style: satisfy biome * refactor(reply): move group intro hints into plugin * fix(reply): avoid plugin registry init cycle * refactor(providers): add lightweight provider dock * refactor(gateway): use typed client id in connect * refactor(providers): document docks and avoid init cycles * refactor(providers): make media limit helper generic * fix(providers): break plugin registry import cycles * style: satisfy biome * refactor(status-all): build providers table from plugins * refactor(gateway): delegate web login to provider plugin * refactor(provider): drop web alias * refactor(provider): lazy-load monitors * style: satisfy lint/format * style: format status-all providers table * style: swiftformat gateway discovery model * test: make reload plan plugin-driven * fix: avoid token stringification in status-all * refactor: make provider IDs explicit in status * feat: warn on signal/imessage provider runtime errors * test: cover gateway provider runtime warnings in status * fix: add runtime kind to provider status issues * test: cover health degradation on probe failure * fix: keep routeReply lightweight * style: organize routeReply imports * refactor(web): extract auth-store helpers * refactor(whatsapp): lazy login imports * refactor(outbound): route replies via plugin outbound * docs: update provider plugin notes * style: format provider status issues * fix: make sandbox scope warning wrap-safe * refactor: load outbound adapters from provider plugins * docs: update provider plugin outbound notes * style(macos): fix swiftformat lint * docs: changelog for provider plugins * fix(macos): satisfy swiftformat * fix(macos): open settings via menu action * style: format after rebase * fix(macos): open Settings via menu action --------- Co-authored-by: LK <luke@kyohere.com> Co-authored-by: Luke K (pr-0f3t) <2609441+lc0rp@users.noreply.github.com> Co-authored-by: Xin <xin@imfing.com>
2026-01-11 11:45:25 +00:00
sessionCtx.AccountId,
sessionCtx.ChatType,
);
const applyReplyToMode = createReplyToModeFilterForChannel(replyToMode, replyToChannel);
2026-01-09 02:21:17 +00:00
const cfg = followupRun.run.config;
const blockReplyCoalescing =
blockStreamingEnabled && opts?.onBlockReply
? resolveBlockStreamingCoalescing(
cfg,
sessionCtx.Provider,
sessionCtx.AccountId,
blockReplyChunking,
)
: undefined;
const blockReplyPipeline =
blockStreamingEnabled && opts?.onBlockReply
? createBlockReplyPipeline({
onBlockReply: opts.onBlockReply,
timeoutMs: blockReplyTimeoutMs,
coalescing: blockReplyCoalescing,
buffer: createAudioAsVoiceBuffer({ isAudioPayload }),
})
: null;
2026-01-04 05:47:21 +01:00
if (shouldSteer && isStreaming) {
const steered = queueEmbeddedPiMessage(followupRun.run.sessionId, followupRun.prompt);
2026-01-04 05:47:21 +01:00
if (steered && !shouldFollowup) {
if (activeSessionEntry && activeSessionStore && sessionKey) {
2026-01-18 22:26:01 +00:00
const updatedAt = Date.now();
activeSessionEntry.updatedAt = updatedAt;
activeSessionStore[sessionKey] = activeSessionEntry;
2026-01-04 05:47:21 +01:00
if (storePath) {
2026-01-18 22:26:01 +00:00
await updateSessionStoreEntry({
storePath,
sessionKey,
update: async () => ({ updatedAt }),
});
2026-01-04 05:47:21 +01:00
}
}
typing.cleanup();
return undefined;
}
}
if (isActive && (shouldFollowup || resolvedQueue.mode === "steer")) {
enqueueFollowupRun(queueKey, followupRun, resolvedQueue);
if (activeSessionEntry && activeSessionStore && sessionKey) {
2026-01-18 22:26:01 +00:00
const updatedAt = Date.now();
activeSessionEntry.updatedAt = updatedAt;
activeSessionStore[sessionKey] = activeSessionEntry;
2026-01-04 05:47:21 +01:00
if (storePath) {
2026-01-18 22:26:01 +00:00
await updateSessionStoreEntry({
storePath,
sessionKey,
update: async () => ({ updatedAt }),
});
2026-01-04 05:47:21 +01:00
}
}
typing.cleanup();
return undefined;
}
2026-01-21 16:54:54 +00:00
await typingSignals.signalRunStart();
activeSessionEntry = await runMemoryFlushIfNeeded({
cfg,
followupRun,
sessionCtx,
opts,
defaultModel,
agentCfgContextTokens,
resolvedVerboseLevel,
sessionEntry: activeSessionEntry,
sessionStore: activeSessionStore,
sessionKey,
storePath,
isHeartbeat,
});
2026-01-12 05:28:17 +00:00
2026-01-04 05:47:21 +01:00
const runFollowupTurn = createFollowupRunner({
opts,
typing,
2026-01-07 21:58:54 +00:00
typingMode,
sessionEntry: activeSessionEntry,
sessionStore: activeSessionStore,
2026-01-04 05:47:21 +01:00
sessionKey,
storePath,
defaultModel,
agentCfgContextTokens,
});
2026-01-09 02:21:17 +00:00
let responseUsageLine: string | undefined;
2026-01-16 09:33:36 +00:00
type SessionResetOptions = {
failureLabel: string;
buildLogMessage: (nextSessionId: string) => string;
cleanupTranscripts?: boolean;
};
2026-01-16 09:33:36 +00:00
const resetSession = async ({
failureLabel,
buildLogMessage,
cleanupTranscripts,
}: SessionResetOptions): Promise<boolean> => {
if (!sessionKey || !activeSessionStore || !storePath) {
return false;
}
const prevEntry = activeSessionStore[sessionKey] ?? activeSessionEntry;
if (!prevEntry) {
return false;
}
2026-01-16 09:33:36 +00:00
const prevSessionId = cleanupTranscripts ? prevEntry.sessionId : undefined;
const nextSessionId = crypto.randomUUID();
const nextEntry: SessionEntry = {
2026-01-16 09:33:36 +00:00
...prevEntry,
sessionId: nextSessionId,
updatedAt: Date.now(),
systemSent: false,
abortedLastRun: false,
};
const agentId = resolveAgentIdFromSessionKey(sessionKey);
const nextSessionFile = resolveSessionTranscriptPath(
nextSessionId,
agentId,
sessionCtx.MessageThreadId,
);
nextEntry.sessionFile = nextSessionFile;
activeSessionStore[sessionKey] = nextEntry;
try {
await updateSessionStore(storePath, (store) => {
store[sessionKey] = nextEntry;
});
} catch (err) {
defaultRuntime.error(
2026-01-16 09:33:36 +00:00
`Failed to persist session reset after ${failureLabel} (${sessionKey}): ${String(err)}`,
);
}
followupRun.run.sessionId = nextSessionId;
followupRun.run.sessionFile = nextSessionFile;
activeSessionEntry = nextEntry;
activeIsNewSession = true;
2026-01-16 09:33:36 +00:00
defaultRuntime.error(buildLogMessage(nextSessionId));
if (cleanupTranscripts && prevSessionId) {
const transcriptCandidates = new Set<string>();
const resolved = resolveSessionFilePath(prevSessionId, prevEntry, { agentId });
if (resolved) {
transcriptCandidates.add(resolved);
}
transcriptCandidates.add(resolveSessionTranscriptPath(prevSessionId, agentId));
for (const candidate of transcriptCandidates) {
try {
fs.unlinkSync(candidate);
} catch {
// Best-effort cleanup.
}
}
}
return true;
};
2026-01-16 09:33:36 +00:00
const resetSessionAfterCompactionFailure = async (reason: string): Promise<boolean> =>
resetSession({
failureLabel: "compaction failure",
buildLogMessage: (nextSessionId) =>
`Auto-compaction failed (${reason}). Restarting session ${sessionKey} -> ${nextSessionId} and retrying.`,
});
const resetSessionAfterRoleOrderingConflict = async (reason: string): Promise<boolean> =>
resetSession({
failureLabel: "role ordering conflict",
buildLogMessage: (nextSessionId) =>
`Role ordering conflict (${reason}). Restarting session ${sessionKey} -> ${nextSessionId}.`,
cleanupTranscripts: true,
});
2026-01-04 05:47:21 +01:00
try {
const runStartedAt = Date.now();
const runOutcome = await runAgentTurnWithFallback({
commandBody,
followupRun,
sessionCtx,
opts,
typingSignals,
blockReplyPipeline,
blockStreamingEnabled,
blockReplyChunking,
resolvedBlockStreamingBreak,
applyReplyToMode,
shouldEmitToolResult,
2026-01-17 05:33:27 +00:00
shouldEmitToolOutput,
pendingToolTasks,
resetSessionAfterCompactionFailure,
resetSessionAfterRoleOrderingConflict,
isHeartbeat,
sessionKey,
getActiveSessionEntry: () => activeSessionEntry,
activeSessionStore,
storePath,
resolvedVerboseLevel,
});
if (runOutcome.kind === "final") {
return finalizeWithFollowup(runOutcome.payload, queueKey, runFollowupTurn);
2026-01-04 05:47:21 +01:00
}
const { runResult, fallbackProvider, fallbackModel, directlySentBlockKeys } = runOutcome;
let { didLogHeartbeatStrip, autoCompactionCompleted } = runOutcome;
2026-01-04 05:47:21 +01:00
if (
shouldInjectGroupIntro &&
activeSessionEntry &&
activeSessionStore &&
2026-01-04 05:47:21 +01:00
sessionKey &&
activeSessionEntry.groupActivationNeedsSystemIntro
2026-01-04 05:47:21 +01:00
) {
2026-01-18 22:26:01 +00:00
const updatedAt = Date.now();
activeSessionEntry.groupActivationNeedsSystemIntro = false;
2026-01-18 22:26:01 +00:00
activeSessionEntry.updatedAt = updatedAt;
activeSessionStore[sessionKey] = activeSessionEntry;
2026-01-04 05:47:21 +01:00
if (storePath) {
2026-01-18 22:26:01 +00:00
await updateSessionStoreEntry({
storePath,
sessionKey,
update: async () => ({
groupActivationNeedsSystemIntro: false,
updatedAt,
}),
});
2026-01-04 05:47:21 +01:00
}
}
const payloadArray = runResult.payloads ?? [];
if (blockReplyPipeline) {
await blockReplyPipeline.flush({ force: true });
blockReplyPipeline.stop();
2026-01-04 05:47:21 +01:00
}
if (pendingToolTasks.size > 0) {
await Promise.allSettled(pendingToolTasks);
}
const usage = runResult.meta.agentMeta?.usage;
const modelUsed = runResult.meta.agentMeta?.model ?? fallbackModel ?? defaultModel;
const providerUsed =
runResult.meta.agentMeta?.provider ?? fallbackProvider ?? followupRun.run.provider;
const cliSessionId = isCliProvider(providerUsed, cfg)
? runResult.meta.agentMeta?.sessionId?.trim()
: undefined;
const contextTokensUsed =
agentCfgContextTokens ??
lookupContextTokens(modelUsed) ??
activeSessionEntry?.contextTokens ??
DEFAULT_CONTEXT_TOKENS;
await persistSessionUsageUpdate({
storePath,
sessionKey,
usage,
modelUsed,
providerUsed,
contextTokensUsed,
systemPromptReport: runResult.meta.systemPromptReport,
cliSessionId,
});
// Drain any late tool/block deliveries before deciding there's "nothing to send".
// Otherwise, a late typing trigger (e.g. from a tool callback) can outlive the run and
// keep the typing indicator stuck.
if (payloadArray.length === 0) {
return finalizeWithFollowup(undefined, queueKey, runFollowupTurn);
}
const payloadResult = buildReplyPayloads({
payloads: payloadArray,
isHeartbeat,
didLogHeartbeatStrip,
blockStreamingEnabled,
blockReplyPipeline,
directlySentBlockKeys,
replyToMode,
replyToChannel,
currentMessageId: sessionCtx.MessageSidFull ?? sessionCtx.MessageSid,
2026-01-08 08:49:16 +01:00
messageProvider: followupRun.run.messageProvider,
messagingToolSentTexts: runResult.messagingToolSentTexts,
messagingToolSentTargets: runResult.messagingToolSentTargets,
2026-01-08 08:49:16 +01:00
originatingTo: sessionCtx.OriginatingTo ?? sessionCtx.To,
accountId: sessionCtx.AccountId,
});
const { replyPayloads } = payloadResult;
didLogHeartbeatStrip = payloadResult.didLogHeartbeatStrip;
2026-01-04 05:47:21 +01:00
if (replyPayloads.length === 0) {
return finalizeWithFollowup(undefined, queueKey, runFollowupTurn);
}
2026-01-04 05:47:21 +01:00
await signalTypingIfNeeded(replyPayloads, typingSignals);
2026-01-04 05:47:21 +01:00
if (isDiagnosticsEnabled(cfg) && hasNonzeroUsage(usage)) {
const input = usage.input ?? 0;
const output = usage.output ?? 0;
const cacheRead = usage.cacheRead ?? 0;
const cacheWrite = usage.cacheWrite ?? 0;
const promptTokens = input + cacheRead + cacheWrite;
const totalTokens = usage.total ?? promptTokens + output;
const costConfig = resolveModelCostConfig({
provider: providerUsed,
model: modelUsed,
config: cfg,
});
const costUsd = estimateUsageCost({ usage, cost: costConfig });
emitDiagnosticEvent({
type: "model.usage",
sessionKey,
sessionId: followupRun.run.sessionId,
channel: replyToChannel,
provider: providerUsed,
model: modelUsed,
usage: {
input,
output,
cacheRead,
cacheWrite,
promptTokens,
total: totalTokens,
},
context: {
limit: contextTokensUsed,
used: totalTokens,
},
costUsd,
durationMs: Date.now() - runStartedAt,
});
}
const responseUsageRaw =
activeSessionEntry?.responseUsage ??
(sessionKey ? activeSessionStore?.[sessionKey]?.responseUsage : undefined);
const responseUsageMode = resolveResponseUsageMode(responseUsageRaw);
if (responseUsageMode !== "off" && hasNonzeroUsage(usage)) {
2026-01-09 02:21:17 +00:00
const authMode = resolveModelAuthMode(providerUsed, cfg);
const showCost = authMode === "api-key";
const costConfig = showCost
? resolveModelCostConfig({
provider: providerUsed,
model: modelUsed,
config: cfg,
})
: undefined;
let formatted = formatResponseUsageLine({
2026-01-09 02:21:17 +00:00
usage,
showCost,
costConfig,
});
if (formatted && responseUsageMode === "full" && sessionKey) {
formatted = `${formatted} · session ${sessionKey}`;
}
if (formatted) {
responseUsageLine = formatted;
}
2026-01-09 02:21:17 +00:00
}
2026-01-04 05:47:21 +01:00
// If verbose is enabled and this is a new session, prepend a session hint.
2026-01-08 08:49:16 +01:00
let finalPayloads = replyPayloads;
2026-01-17 05:33:27 +00:00
const verboseEnabled = resolvedVerboseLevel !== "off";
if (autoCompactionCompleted) {
const count = await incrementCompactionCount({
sessionEntry: activeSessionEntry,
sessionStore: activeSessionStore,
sessionKey,
storePath,
});
2026-01-17 05:33:27 +00:00
if (verboseEnabled) {
const suffix = typeof count === "number" ? ` (count ${count})` : "";
finalPayloads = [{ text: `🧹 Auto-compaction complete${suffix}.` }, ...finalPayloads];
}
}
2026-01-17 05:33:27 +00:00
if (verboseEnabled && activeIsNewSession) {
finalPayloads = [{ text: `🧭 New session: ${followupRun.run.sessionId}` }, ...finalPayloads];
2026-01-04 05:47:21 +01:00
}
2026-01-09 02:21:17 +00:00
if (responseUsageLine) {
finalPayloads = appendUsageLine(finalPayloads, responseUsageLine);
}
2026-01-04 05:47:21 +01:00
return finalizeWithFollowup(
finalPayloads.length === 1 ? finalPayloads[0] : finalPayloads,
queueKey,
runFollowupTurn,
2026-01-04 05:47:21 +01:00
);
} finally {
blockReplyPipeline?.stop();
2026-01-06 03:05:11 +00:00
typing.markRunComplete();
2026-01-04 05:47:21 +01:00
}
}