@xopcai/xopc 0.0.114 → 0.0.115

package/dist/src/agent/embedded/session-tool-result-guard.js.map

@@ -1 +1 @@
-{"version":3,"file":"session-tool-result-guard.js","names":[],"sources":["../../../../src/agent/embedded/session-tool-result-guard.ts"],"sourcesContent":["/**\n * Session tool-result guard — wraps a pi `SessionManager.appendMessage` to:\n *   1. cap oversized tool-result text + details so they cannot blow up the next\n *      LLM request (size + persistence limits).\n *   2. track pending tool-call IDs so missing tool results can be synthesised\n *      (some providers refuse a turn that has an orphan tool_use block).\n *   3. drop assistant `toolCall` blocks whose tool name is not in the allowlist\n *      (these would also trigger provider 400s).\n *   4. broadcast `xopc:transcript-row` updates so the gateway UI can stream them.\n *\n * Previously this module shipped with three sibling files\n * (`session-tool-result-state.ts`, `session-raw-append-message.ts`,\n * `session-tool-result-guard-wrapper.ts`). They are now consolidated here as\n * private constructs around the `ToolResultGuard` class. Pi-coding-agent owns\n * the `SessionManager` instance and calls `appendMessage` from inside the\n * runtime, so we still need to monkey-patch that method — but the patched\n * implementation is just `guard.guardedAppend.bind(guard)` and all state lives\n * on the class.\n */\n\nimport type { AgentMessage } from '@earendil-works/pi-agent-core';\nimport type { SessionManager } from '@earendil-works/pi-coding-agent';\n\nimport type { Config } from '../../config/schema.js';\nimport {\n  boundedJsonUtf8Bytes,\n  firstEnumerableOwnKeys,\n  jsonUtf8BytesOrInfinity,\n  type BoundedJsonUtf8Bytes,\n} from '../../infra/json-utf8-bytes.js';\nimport { emitSessionTranscriptUpdate } from '../../session/transcript-events.js';\nimport { normalizeOptionalString } from '../../utils/string-coerce.js';\nimport { formatContextLimitTruncationNotice } from './tool-result-context-guard.js';\nimport {\n  DEFAULT_MAX_LIVE_TOOL_RESULT_CHARS,\n  resolveLiveToolResultMaxChars,\n  truncateToolResultMessage,\n} from './tool-result-truncation.js';\nimport {\n  makeMissingToolResult,\n  sanitizeToolCallInputs,\n} from '../transcript/session-transcript-repair.js';\nimport {\n  extractToolCallsFromAssistant,\n  extractToolResultId,\n} from '../transcript/tool-call-id.js';\n\n// ── Public surface ──────────────────────────────────────────────────────────\n\nexport type BeforeMessageWriteHookEvent = { message: AgentMessage };\nexport type BeforeMessageWriteHookResult =\n  | { block?: boolean; message?: AgentMessage }\n  | undefined;\n\nexport interface ToolResultGuardOptions {\n  /** Optional session key for transcript update broadcasts. */\n  sessionKey?: string;\n  /** Optional transform applied to any message before persistence. */\n  transformMessageForPersistence?: (message: AgentMessage) => AgentMessage;\n  /**\n   * Optional, synchronous transform applied to toolResult messages *before* they are\n   * persisted to the session transcript.\n   */\n  transformToolResultForPersistence?: (\n    message: AgentMessage,\n    meta: { toolCallId?: string; toolName?: string; isSynthetic?: boolean },\n  ) => AgentMessage;\n  /**\n   * Whether to synthesize missing tool results to satisfy strict providers.\n   * Defaults to true.\n   */\n  allowSyntheticToolResults?: boolean;\n  missingToolResultText?: string;\n  /**\n   * Optional set/list of tool names accepted for assistant toolCall/toolUse blocks.\n   * When set, tool calls with unknown names are dropped before persistence.\n   */\n  allowedToolNames?: Iterable<string>;\n  /**\n   * Synchronous hook invoked before any message is written to the session JSONL.\n   * If the hook returns { block: true }, the message is silently dropped.\n   * If it returns { message }, the modified message is written instead.\n   */\n  beforeMessageWriteHook?: (event: BeforeMessageWriteHookEvent) => BeforeMessageWriteHookResult;\n  maxToolResultChars?: number;\n}\n\nexport interface InstallSessionToolResultGuardResult {\n  flushPendingToolResults: () => void;\n  clearPendingToolResults: () => void;\n  getPendingIds: () => string[];\n}\n\n/** Idempotent wrapper that also adds the helper methods consumers expect. */\nexport type GuardedPiTranscriptManager = SessionManager & {\n  flushPendingToolResults?: () => void;\n  clearPendingToolResults?: () => void;\n};\n\n/**\n * Install the guard on a SessionManager and return its control API.\n * Subsequent assistant/toolResult writes by pi-coding-agent flow through the\n * guard transparently.\n */\nexport function installSessionToolResultGuard(\n  sessionManager: SessionManager,\n  opts: ToolResultGuardOptions = {},\n): InstallSessionToolResultGuardResult {\n  const guard = new ToolResultGuard(sessionManager, opts);\n  guard.attach();\n  return {\n    flushPendingToolResults: () => guard.flushPending(),\n    clearPendingToolResults: () => guard.clearPending(),\n    getPendingIds: () => guard.getPendingIds(),\n  };\n}\n\n/**\n * Convenience wrapper used by the embedded runner pool: install the guard\n * (idempotent), pin the size cap from the model context window, and expose\n * `flushPendingToolResults` / `clearPendingToolResults` directly on the\n * SessionManager instance so callers do not need to keep the install result.\n */\nexport function guardSessionManager(\n  sessionManager: SessionManager,\n  opts?: {\n    agentId?: string;\n    sessionKey?: string;\n    config?: Config;\n    contextWindowTokens?: number;\n    allowSyntheticToolResults?: boolean;\n    missingToolResultText?: string;\n    allowedToolNames?: Iterable<string>;\n    transformMessageForPersistence?: (message: AgentMessage) => AgentMessage;\n  },\n): GuardedPiTranscriptManager {\n  if (typeof (sessionManager as GuardedPiTranscriptManager).flushPendingToolResults === 'function') {\n    return sessionManager as GuardedPiTranscriptManager;\n  }\n\n  const result = installSessionToolResultGuard(sessionManager, {\n    sessionKey: opts?.sessionKey,\n    transformMessageForPersistence: opts?.transformMessageForPersistence,\n    allowSyntheticToolResults: opts?.allowSyntheticToolResults,\n    missingToolResultText: opts?.missingToolResultText,\n    allowedToolNames: opts?.allowedToolNames,\n    maxToolResultChars:\n      typeof opts?.contextWindowTokens === 'number'\n        ? resolveLiveToolResultMaxChars({\n            contextWindowTokens: opts.contextWindowTokens,\n            cfg: opts?.config,\n            agentId: opts?.agentId,\n          })\n        : undefined,\n  });\n  const tagged = sessionManager as GuardedPiTranscriptManager;\n  tagged.flushPendingToolResults = result.flushPendingToolResults;\n  tagged.clearPendingToolResults = result.clearPendingToolResults;\n  return tagged;\n}\n\n/**\n * Recover the original (un-guarded) appendMessage for a session manager.\n * Useful for callers that need a low-level \"bypass the guard\" write path.\n */\nexport function getRawSessionAppendMessage(\n  sessionManager: SessionManager,\n): SessionManager['appendMessage'] {\n  const stored = (sessionManager as SessionManagerWithRawAppend)[RAW_APPEND_MESSAGE];\n  return stored ?? sessionManager.appendMessage.bind(sessionManager);\n}\n\n// ── Internal: persistence + truncation helpers ──────────────────────────────\n\nfunction resolveMaxToolResultChars(opts: { maxToolResultChars?: number }): number {\n  return Math.max(1, opts.maxToolResultChars ?? DEFAULT_MAX_LIVE_TOOL_RESULT_CHARS);\n}\n\nfunction capToolResultSize(msg: AgentMessage, maxChars: number): AgentMessage {\n  if ((msg as { role?: string }).role !== 'toolResult') {\n    return msg;\n  }\n  return truncateToolResultMessage(msg, maxChars, {\n    suffix: (truncatedChars) => formatContextLimitTruncationNotice(truncatedChars),\n    minKeepChars: 2_000,\n  });\n}\n\n// `details` is runtime/UI metadata, not model-visible tool output. Keep the\n// session JSONL useful for debugging without letting metadata blobs dominate\n// disk, replay repair, transcript broadcasts, or future tooling that reads raw\n// sessions. Model-visible text belongs in tool result `content`.\nconst MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES = 8_192;\nconst MAX_PERSISTED_DETAIL_STRING_CHARS = 2_000;\nconst MAX_PERSISTED_DETAIL_SESSION_COUNT = 10;\nconst MAX_PERSISTED_DETAIL_FALLBACK_STRING_CHARS = 200;\n\nfunction originalDetailsSizeFields(size: BoundedJsonUtf8Bytes): Record<string, number> {\n  return size.complete\n    ? { originalDetailsBytes: size.bytes }\n    : { originalDetailsBytesAtLeast: size.bytes };\n}\n\nfunction truncatePersistedDetailString(\n  value: string,\n  maxChars = MAX_PERSISTED_DETAIL_STRING_CHARS,\n): string {\n  if (value.length <= maxChars) {\n    return value;\n  }\n  return `${value.slice(0, maxChars)}\\n\\n[xopc persisted detail truncated: ${\n    value.length - maxChars\n  } chars omitted]`;\n}\n\nfunction sanitizePersistedSessionDetail(value: unknown): unknown {\n  if (!value || typeof value !== 'object') {\n    return value;\n  }\n  const src = value as Record<string, unknown>;\n  const out: Record<string, unknown> = {};\n  for (const key of [\n    'sessionId',\n    'status',\n    'pid',\n    'startedAt',\n    'endedAt',\n    'runtimeMs',\n    'cwd',\n    'name',\n    'truncated',\n    'exitCode',\n    'exitSignal',\n  ]) {\n    const field = src[key];\n    if (field !== undefined) {\n      out[key] = typeof field === 'string' ? truncatePersistedDetailString(field, 500) : field;\n    }\n  }\n  if (typeof src.command === 'string') {\n    out.command = truncatePersistedDetailString(src.command, 500);\n  }\n  return out;\n}\n\nfunction buildPersistedDetailsFallback(\n  src: Record<string, unknown> | undefined,\n  originalSize: BoundedJsonUtf8Bytes,\n  sanitizedBytes?: number,\n): Record<string, unknown> {\n  const fallback: Record<string, unknown> = {\n    persistedDetailsTruncated: true,\n    finalDetailsTruncated: true,\n    ...originalDetailsSizeFields(originalSize),\n  };\n  if (sanitizedBytes !== undefined) {\n    fallback.sanitizedDetailsBytes = sanitizedBytes;\n  }\n  if (src) {\n    fallback.originalDetailKeys = firstEnumerableOwnKeys(src, 40);\n    for (const key of ['status', 'sessionId', 'pid', 'exitCode', 'exitSignal', 'truncated']) {\n      const field = src[key];\n      if (field !== undefined) {\n        fallback[key] =\n          typeof field === 'string'\n            ? truncatePersistedDetailString(field, MAX_PERSISTED_DETAIL_FALLBACK_STRING_CHARS)\n            : field;\n      }\n    }\n  }\n  return fallback;\n}\n\nfunction enforcePersistedDetailsByteCap(\n  value: Record<string, unknown>,\n  src: Record<string, unknown> | undefined,\n  originalSize: BoundedJsonUtf8Bytes,\n): Record<string, unknown> {\n  const sanitizedBytes = jsonUtf8BytesOrInfinity(value);\n  if (sanitizedBytes <= MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES) {\n    return value;\n  }\n  const fallback = buildPersistedDetailsFallback(src, originalSize, sanitizedBytes);\n  if (jsonUtf8BytesOrInfinity(fallback) <= MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES) {\n    return fallback;\n  }\n  return {\n    persistedDetailsTruncated: true,\n    finalDetailsTruncated: true,\n    ...originalDetailsSizeFields(originalSize),\n    sanitizedDetailsBytes: sanitizedBytes,\n  };\n}\n\nfunction sanitizeToolResultDetailsForPersistence(details: unknown): unknown {\n  if (details === undefined || details === null) {\n    return details;\n  }\n  const originalSize = boundedJsonUtf8Bytes(details, MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES);\n  if (originalSize.complete && originalSize.bytes <= MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES) {\n    return details;\n  }\n  if (typeof details !== 'object') {\n    return enforcePersistedDetailsByteCap(\n      {\n        persistedDetailsTruncated: true,\n        ...originalDetailsSizeFields(originalSize),\n        valueType: typeof details,\n      },\n      undefined,\n      originalSize,\n    );\n  }\n  const src = details as Record<string, unknown>;\n  const out: Record<string, unknown> = {\n    persistedDetailsTruncated: true,\n    ...originalDetailsSizeFields(originalSize),\n    originalDetailKeys: firstEnumerableOwnKeys(src, 40),\n  };\n  for (const key of [\n    'status',\n    'sessionId',\n    'pid',\n    'startedAt',\n    'endedAt',\n    'cwd',\n    'name',\n    'exitCode',\n    'exitSignal',\n    'retryInMs',\n    'total',\n    'totalLines',\n    'totalChars',\n    'truncated',\n    'fullOutputPath',\n    'truncation',\n  ]) {\n    const field = src[key];\n    if (field !== undefined) {\n      out[key] = typeof field === 'string' ? truncatePersistedDetailString(field) : field;\n    }\n  }\n  if (typeof src.tail === 'string') {\n    out.tail = truncatePersistedDetailString(src.tail);\n  }\n  if (Array.isArray(src.sessions)) {\n    out.sessions = src.sessions\n      .slice(0, MAX_PERSISTED_DETAIL_SESSION_COUNT)\n      .map(sanitizePersistedSessionDetail);\n    if (src.sessions.length > MAX_PERSISTED_DETAIL_SESSION_COUNT) {\n      out.sessionsTruncated = src.sessions.length - MAX_PERSISTED_DETAIL_SESSION_COUNT;\n    }\n  }\n  return enforcePersistedDetailsByteCap(out, src, originalSize);\n}\n\nfunction capToolResultDetails(msg: AgentMessage): AgentMessage {\n  if ((msg as { role?: string }).role !== 'toolResult') {\n    return msg;\n  }\n  const details = (msg as { details?: unknown }).details;\n  const sanitizedDetails = sanitizeToolResultDetailsForPersistence(details);\n  if (sanitizedDetails === details) {\n    return msg;\n  }\n  const next = { ...msg } as AgentMessage & { details?: unknown };\n  next.details = sanitizedDetails;\n  return next;\n}\n\nfunction capToolResultForPersistence(msg: AgentMessage, maxChars: number): AgentMessage {\n  return capToolResultDetails(capToolResultSize(msg, maxChars));\n}\n\nfunction normalizePersistedToolResultName(\n  message: AgentMessage,\n  fallbackName?: string,\n): AgentMessage {\n  if ((message as { role?: unknown }).role !== 'toolResult') {\n    return message;\n  }\n  const toolResult = message as Extract<AgentMessage, { role: 'toolResult' }>;\n  const rawToolName = (toolResult as { toolName?: unknown }).toolName;\n  const normalizedToolName = normalizeOptionalString(rawToolName);\n  if (normalizedToolName) {\n    if (rawToolName === normalizedToolName) {\n      return toolResult;\n    }\n    return { ...toolResult, toolName: normalizedToolName };\n  }\n\n  const normalizedFallback = normalizeOptionalString(fallbackName);\n  if (normalizedFallback) {\n    return { ...toolResult, toolName: normalizedFallback };\n  }\n\n  if (typeof rawToolName === 'string') {\n    return { ...toolResult, toolName: 'unknown' };\n  }\n  return toolResult;\n}\n\n// ── Internal: raw-append symbol storage ─────────────────────────────────────\n\nconst RAW_APPEND_MESSAGE = Symbol('xopc.session.rawAppendMessage');\n\ntype SessionManagerWithRawAppend = SessionManager & {\n  [RAW_APPEND_MESSAGE]?: SessionManager['appendMessage'];\n};\n\nfunction rememberOriginalAppend(\n  sessionManager: SessionManager,\n): SessionManager['appendMessage'] {\n  const tagged = sessionManager as SessionManagerWithRawAppend;\n  const stored = tagged[RAW_APPEND_MESSAGE];\n  if (stored) {\n    return stored;\n  }\n  const original = sessionManager.appendMessage.bind(sessionManager);\n  tagged[RAW_APPEND_MESSAGE] = original;\n  return original;\n}\n\n// ── Internal: pending tool-call state ──────────────────────────────────────\n\ntype PendingToolCall = { id: string; name?: string };\n\nclass PendingToolCallTracker {\n  private readonly pending = new Map<string, string | undefined>();\n\n  size(): number {\n    return this.pending.size;\n  }\n\n  getToolName(id: string): string | undefined {\n    return this.pending.get(id);\n  }\n\n  delete(id: string): void {\n    this.pending.delete(id);\n  }\n\n  clear(): void {\n    this.pending.clear();\n  }\n\n  trackToolCalls(calls: readonly PendingToolCall[]): void {\n    for (const call of calls) {\n      this.pending.set(call.id, call.name);\n    }\n  }\n\n  entries(): IterableIterator<[string, string | undefined]> {\n    return this.pending.entries();\n  }\n\n  getPendingIds(): string[] {\n    return Array.from(this.pending.keys());\n  }\n\n  shouldFlushForSanitizedDrop(): boolean {\n    return this.pending.size > 0;\n  }\n\n  shouldFlushBeforeNonToolResult(nextRole: unknown, toolCallCount: number): boolean {\n    return this.pending.size > 0 && (toolCallCount === 0 || nextRole !== 'assistant');\n  }\n\n  shouldFlushBeforeNewToolCalls(toolCallCount: number): boolean {\n    return this.pending.size > 0 && toolCallCount > 0;\n  }\n}\n\n// ── ToolResultGuard class ──────────────────────────────────────────────────\n\nclass ToolResultGuard {\n  private readonly sessionManager: SessionManager;\n  private readonly opts: ToolResultGuardOptions;\n  private readonly pending = new PendingToolCallTracker();\n  private readonly originalAppend: SessionManager['appendMessage'];\n  private readonly allowSyntheticToolResults: boolean;\n  private readonly maxToolResultChars: number;\n\n  constructor(sessionManager: SessionManager, opts: ToolResultGuardOptions) {\n    this.sessionManager = sessionManager;\n    this.opts = opts;\n    this.originalAppend = rememberOriginalAppend(sessionManager);\n    this.allowSyntheticToolResults = opts.allowSyntheticToolResults ?? true;\n    this.maxToolResultChars = resolveMaxToolResultChars(opts);\n  }\n\n  /** Monkey-patch the session manager so pi-coding-agent's internal appendMessage flows through us. */\n  attach(): void {\n    const bound = this.guardedAppend.bind(this);\n    this.sessionManager.appendMessage = bound as SessionManager['appendMessage'];\n  }\n\n  flushPending(): void {\n    if (this.pending.size() === 0) {\n      return;\n    }\n    if (this.allowSyntheticToolResults) {\n      for (const [id, name] of this.pending.entries()) {\n        const synthetic = makeMissingToolResult({\n          toolCallId: id,\n          toolName: name,\n          text: this.opts.missingToolResultText,\n        });\n        const flushed = this.applyBeforeWriteHook(\n          this.persistToolResult(this.persistMessage(synthetic), {\n            toolCallId: id,\n            toolName: name,\n            isSynthetic: true,\n          }),\n        );\n        if (flushed) {\n          this.originalAppend(capToolResultForPersistence(flushed, this.maxToolResultChars) as never);\n        }\n      }\n    }\n    this.pending.clear();\n  }\n\n  clearPending(): void {\n    this.pending.clear();\n  }\n\n  getPendingIds(): string[] {\n    return this.pending.getPendingIds();\n  }\n\n  private persistMessage(message: AgentMessage): AgentMessage {\n    const transformer = this.opts.transformMessageForPersistence;\n    return transformer ? transformer(message) : message;\n  }\n\n  private persistToolResult(\n    message: AgentMessage,\n    meta: { toolCallId?: string; toolName?: string; isSynthetic?: boolean },\n  ): AgentMessage {\n    const transformer = this.opts.transformToolResultForPersistence;\n    return transformer ? transformer(message, meta) : message;\n  }\n\n  /**\n   * Run the before_message_write hook. Returns the (possibly modified) message,\n   * or null if the message should be blocked.\n   */\n  private applyBeforeWriteHook(msg: AgentMessage): AgentMessage | null {\n    const beforeWrite = this.opts.beforeMessageWriteHook;\n    if (!beforeWrite) {\n      return msg;\n    }\n    const result = beforeWrite({ message: msg });\n    if (result?.block) {\n      return null;\n    }\n    if (result?.message) {\n      return result.message;\n    }\n    return msg;\n  }\n\n  private guardedAppend(message: AgentMessage): unknown {\n    let nextMessage = message;\n    const role = (message as { role?: unknown }).role;\n    if (role === 'assistant') {\n      const sanitized = sanitizeToolCallInputs([message], {\n        allowedToolNames: this.opts.allowedToolNames,\n      });\n      if (sanitized.length === 0) {\n        if (this.pending.shouldFlushForSanitizedDrop()) {\n          this.flushPending();\n        }\n        return undefined;\n      }\n      nextMessage = sanitized[0];\n    }\n    const nextRole = (nextMessage as { role?: unknown }).role;\n\n    if (nextRole === 'toolResult') {\n      const id = extractToolResultId(nextMessage as Extract<AgentMessage, { role: 'toolResult' }>);\n      const toolName = id ? this.pending.getToolName(id) : undefined;\n      if (id) {\n        this.pending.delete(id);\n      }\n      const normalizedToolResult = normalizePersistedToolResultName(nextMessage, toolName);\n      // Apply hard size cap before persistence to prevent oversized tool results\n      // from consuming the entire context window on subsequent LLM calls.\n      const capped = capToolResultForPersistence(\n        this.persistMessage(normalizedToolResult),\n        this.maxToolResultChars,\n      );\n      const persisted = this.applyBeforeWriteHook(\n        this.persistToolResult(capped, {\n          toolCallId: id ?? undefined,\n          toolName,\n          isSynthetic: false,\n        }),\n      );\n      if (!persisted) {\n        return undefined;\n      }\n      return this.originalAppend(capToolResultForPersistence(persisted, this.maxToolResultChars) as never);\n    }\n\n    // Skip tool call extraction for aborted/errored assistant messages.\n    // When stopReason is \"error\" or \"aborted\", the tool_use blocks may be incomplete\n    // and should not have synthetic tool_results created. Creating synthetic results\n    // for incomplete tool calls causes API 400 errors:\n    // \"unexpected tool_use_id found in tool_result blocks\"\n    // This matches the behavior in repairToolUseResultPairing (session-transcript-repair.ts)\n    const stopReason = (nextMessage as { stopReason?: string }).stopReason;\n    const toolCalls =\n      nextRole === 'assistant' && stopReason !== 'aborted' && stopReason !== 'error'\n        ? extractToolCallsFromAssistant(nextMessage as Extract<AgentMessage, { role: 'assistant' }>)\n        : [];\n\n    // Always clear pending tool call state before appending non-tool-result messages.\n    // flushPendingToolResults() only inserts synthetic results when allowSyntheticToolResults\n    // is true; it always clears the pending map. Without this, providers that disable\n    // synthetic results (e.g. OpenAI) accumulate stale pending state when a user message\n    // interrupts in-flight tool calls, leaving orphaned tool_use blocks in the transcript\n    // that cause API 400 errors on subsequent requests.\n    if (this.pending.shouldFlushBeforeNonToolResult(nextRole, toolCalls.length)) {\n      this.flushPending();\n    }\n    // If new tool calls arrive while older ones are pending, flush the old ones first.\n    if (this.pending.shouldFlushBeforeNewToolCalls(toolCalls.length)) {\n      this.flushPending();\n    }\n\n    const finalMessage = this.applyBeforeWriteHook(this.persistMessage(nextMessage));\n    if (!finalMessage) {\n      return undefined;\n    }\n    const result = this.originalAppend(finalMessage as never);\n\n    if (this.opts.sessionKey) {\n      emitSessionTranscriptUpdate({\n        sessionKey: this.opts.sessionKey,\n        message: finalMessage,\n        messageId: typeof result === 'string' ? result : undefined,\n      });\n    }\n\n    if (toolCalls.length > 0) {\n      this.pending.trackToolCalls(toolCalls);\n    }\n\n    return result;\n  }\n}\n"],"mappings":";;;;;;;;oBA+BuE;;;;;;AAyEvE,SAAgB,8BACd,gBACA,OAA+B,EAAE,EACI;CACrC,MAAM,QAAQ,IAAI,gBAAgB,gBAAgB,KAAK;AACvD,OAAM,QAAQ;AACd,QAAO;EACL,+BAA+B,MAAM,cAAc;EACnD,+BAA+B,MAAM,cAAc;EACnD,qBAAqB,MAAM,eAAe;EAC3C;;;;;;;;AASH,SAAgB,oBACd,gBACA,MAU4B;AAC5B,KAAI,OAAQ,eAA8C,4BAA4B,WACpF,QAAO;CAGT,MAAM,SAAS,8BAA8B,gBAAgB;EAC3D,YAAY,MAAM;EAClB,gCAAgC,MAAM;EACtC,2BAA2B,MAAM;EACjC,uBAAuB,MAAM;EAC7B,kBAAkB,MAAM;EACxB,oBACE,OAAO,MAAM,wBAAwB,WACjC,8BAA8B;GAC5B,qBAAqB,KAAK;GAC1B,KAAK,MAAM;GACX,SAAS,MAAM;GAChB,CAAC,GACF,KAAA;EACP,CAAC;CACF,MAAM,SAAS;AACf,QAAO,0BAA0B,OAAO;AACxC,QAAO,0BAA0B,OAAO;AACxC,QAAO;;;;;;AAOT,SAAgB,2BACd,gBACiC;AAEjC,QADgB,eAA+C,uBAC9C,eAAe,cAAc,KAAK,eAAe;;AAKpE,SAAS,0BAA0B,MAA+C;AAChF,QAAO,KAAK,IAAI,GAAG,KAAK,sBAAA,KAAyD;;AAGnF,SAAS,kBAAkB,KAAmB,UAAgC;AAC5E,KAAK,IAA0B,SAAS,aACtC,QAAO;AAET,QAAO,0BAA0B,KAAK,UAAU;EAC9C,SAAS,mBAAmB,mCAAmC,eAAe;EAC9E,cAAc;EACf,CAAC;;AAOJ,MAAM,0CAA0C;AAChD,MAAM,oCAAoC;AAC1C,MAAM,qCAAqC;AAC3C,MAAM,6CAA6C;AAEnD,SAAS,0BAA0B,MAAoD;AACrF,QAAO,KAAK,WACR,EAAE,sBAAsB,KAAK,OAAO,GACpC,EAAE,6BAA6B,KAAK,OAAO;;AAGjD,SAAS,8BACP,OACA,WAAW,mCACH;AACR,KAAI,MAAM,UAAU,SAClB,QAAO;AAET,QAAO,GAAG,MAAM,MAAM,GAAG,SAAS,CAAC,wCACjC,MAAM,SAAS,SAChB;;AAGH,SAAS,+BAA+B,OAAyB;AAC/D,KAAI,CAAC,SAAS,OAAO,UAAU,SAC7B,QAAO;CAET,MAAM,MAAM;CACZ,MAAM,MAA+B,EAAE;AACvC,MAAK,MAAM,OAAO;EAChB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACD,EAAE;EACD,MAAM,QAAQ,IAAI;AAClB,MAAI,UAAU,KAAA,EACZ,KAAI,OAAO,OAAO,UAAU,WAAW,8BAA8B,OAAO,IAAI,GAAG;;AAGvF,KAAI,OAAO,IAAI,YAAY,SACzB,KAAI,UAAU,8BAA8B,IAAI,SAAS,IAAI;AAE/D,QAAO;;AAGT,SAAS,8BACP,KACA,cACA,gBACyB;CACzB,MAAM,WAAoC;EACxC,2BAA2B;EAC3B,uBAAuB;EACvB,GAAG,0BAA0B,aAAa;EAC3C;AACD,KAAI,mBAAmB,KAAA,EACrB,UAAS,wBAAwB;AAEnC,KAAI,KAAK;AACP,WAAS,qBAAqB,uBAAuB,KAAK,GAAG;AAC7D,OAAK,MAAM,OAAO;GAAC;GAAU;GAAa;GAAO;GAAY;GAAc;GAAY,EAAE;GACvF,MAAM,QAAQ,IAAI;AAClB,OAAI,UAAU,KAAA,EACZ,UAAS,OACP,OAAO,UAAU,WACb,8BAA8B,OAAO,2CAA2C,GAChF;;;AAIZ,QAAO;;AAGT,SAAS,+BACP,OACA,KACA,cACyB;CACzB,MAAM,iBAAiB,wBAAwB,MAAM;AACrD,KAAI,kBAAkB,wCACpB,QAAO;CAET,MAAM,WAAW,8BAA8B,KAAK,cAAc,eAAe;AACjF,KAAI,wBAAwB,SAAS,IAAI,wCACvC,QAAO;AAET,QAAO;EACL,2BAA2B;EAC3B,uBAAuB;EACvB,GAAG,0BAA0B,aAAa;EAC1C,uBAAuB;EACxB;;AAGH,SAAS,wCAAwC,SAA2B;AAC1E,KAAI,YAAY,KAAA,KAAa,YAAY,KACvC,QAAO;CAET,MAAM,eAAe,qBAAqB,SAAS,wCAAwC;AAC3F,KAAI,aAAa,YAAY,aAAa,SAAS,wCACjD,QAAO;AAET,KAAI,OAAO,YAAY,SACrB,QAAO,+BACL;EACE,2BAA2B;EAC3B,GAAG,0BAA0B,aAAa;EAC1C,WAAW,OAAO;EACnB,EACD,KAAA,GACA,aACD;CAEH,MAAM,MAAM;CACZ,MAAM,MAA+B;EACnC,2BAA2B;EAC3B,GAAG,0BAA0B,aAAa;EAC1C,oBAAoB,uBAAuB,KAAK,GAAG;EACpD;AACD,MAAK,MAAM,OAAO;EAChB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACD,EAAE;EACD,MAAM,QAAQ,IAAI;AAClB,MAAI,UAAU,KAAA,EACZ,KAAI,OAAO,OAAO,UAAU,WAAW,8BAA8B,MAAM,GAAG;;AAGlF,KAAI,OAAO,IAAI,SAAS,SACtB,KAAI,OAAO,8BAA8B,IAAI,KAAK;AAEpD,KAAI,MAAM,QAAQ,IAAI,SAAS,EAAE;AAC/B,MAAI,WAAW,IAAI,SAChB,MAAM,GAAG,mCAAmC,CAC5C,IAAI,+BAA+B;AACtC,MAAI,IAAI,SAAS,SAAS,mCACxB,KAAI,oBAAoB,IAAI,SAAS,SAAS;;AAGlD,QAAO,+BAA+B,KAAK,KAAK,aAAa;;AAG/D,SAAS,qBAAqB,KAAiC;AAC7D,KAAK,IAA0B,SAAS,aACtC,QAAO;CAET,MAAM,UAAW,IAA8B;CAC/C,MAAM,mBAAmB,wCAAwC,QAAQ;AACzE,KAAI,qBAAqB,QACvB,QAAO;CAET,MAAM,OAAO,EAAE,GAAG,KAAK;AACvB,MAAK,UAAU;AACf,QAAO;;AAGT,SAAS,4BAA4B,KAAmB,UAAgC;AACtF,QAAO,qBAAqB,kBAAkB,KAAK,SAAS,CAAC;;AAG/D,SAAS,iCACP,SACA,cACc;AACd,KAAK,QAA+B,SAAS,aAC3C,QAAO;CAET,MAAM,aAAa;CACnB,MAAM,cAAe,WAAsC;CAC3D,MAAM,qBAAqB,wBAAwB,YAAY;AAC/D,KAAI,oBAAoB;AACtB,MAAI,gBAAgB,mBAClB,QAAO;AAET,SAAO;GAAE,GAAG;GAAY,UAAU;GAAoB;;CAGxD,MAAM,qBAAqB,wBAAwB,aAAa;AAChE,KAAI,mBACF,QAAO;EAAE,GAAG;EAAY,UAAU;EAAoB;AAGxD,KAAI,OAAO,gBAAgB,SACzB,QAAO;EAAE,GAAG;EAAY,UAAU;EAAW;AAE/C,QAAO;;AAKT,MAAM,qBAAqB,OAAO,gCAAgC;AAMlE,SAAS,uBACP,gBACiC;CACjC,MAAM,SAAS;CACf,MAAM,SAAS,OAAO;AACtB,KAAI,OACF,QAAO;CAET,MAAM,WAAW,eAAe,cAAc,KAAK,eAAe;AAClE,QAAO,sBAAsB;AAC7B,QAAO;;AAOT,IAAM,yBAAN,MAA6B;CAC3B,0BAA2B,IAAI,KAAiC;CAEhE,OAAe;AACb,SAAO,KAAK,QAAQ;;CAGtB,YAAY,IAAgC;AAC1C,SAAO,KAAK,QAAQ,IAAI,GAAG;;CAG7B,OAAO,IAAkB;AACvB,OAAK,QAAQ,OAAO,GAAG;;CAGzB,QAAc;AACZ,OAAK,QAAQ,OAAO;;CAGtB,eAAe,OAAyC;AACtD,OAAK,MAAM,QAAQ,MACjB,MAAK,QAAQ,IAAI,KAAK,IAAI,KAAK,KAAK;;CAIxC,UAA0D;AACxD,SAAO,KAAK,QAAQ,SAAS;;CAG/B,gBAA0B;AACxB,SAAO,MAAM,KAAK,KAAK,QAAQ,MAAM,CAAC;;CAGxC,8BAAuC;AACrC,SAAO,KAAK,QAAQ,OAAO;;CAG7B,+BAA+B,UAAmB,eAAgC;AAChF,SAAO,KAAK,QAAQ,OAAO,MAAM,kBAAkB,KAAK,aAAa;;CAGvE,8BAA8B,eAAgC;AAC5D,SAAO,KAAK,QAAQ,OAAO,KAAK,gBAAgB;;;AAMpD,IAAM,kBAAN,MAAsB;CACpB;CACA;CACA,UAA2B,IAAI,wBAAwB;CACvD;CACA;CACA;CAEA,YAAY,gBAAgC,MAA8B;AACxE,OAAK,iBAAiB;AACtB,OAAK,OAAO;AACZ,OAAK,iBAAiB,uBAAuB,eAAe;AAC5D,OAAK,4BAA4B,KAAK,6BAA6B;AACnE,OAAK,qBAAqB,0BAA0B,KAAK;;;CAI3D,SAAe;EACb,MAAM,QAAQ,KAAK,cAAc,KAAK,KAAK;AAC3C,OAAK,eAAe,gBAAgB;;CAGtC,eAAqB;AACnB,MAAI,KAAK,QAAQ,MAAM,KAAK,EAC1B;AAEF,MAAI,KAAK,0BACP,MAAK,MAAM,CAAC,IAAI,SAAS,KAAK,QAAQ,SAAS,EAAE;GAC/C,MAAM,YAAY,sBAAsB;IACtC,YAAY;IACZ,UAAU;IACV,MAAM,KAAK,KAAK;IACjB,CAAC;GACF,MAAM,UAAU,KAAK,qBACnB,KAAK,kBAAkB,KAAK,eAAe,UAAU,EAAE;IACrD,YAAY;IACZ,UAAU;IACV,aAAa;IACd,CAAC,CACH;AACD,OAAI,QACF,MAAK,eAAe,4BAA4B,SAAS,KAAK,mBAAmB,CAAU;;AAIjG,OAAK,QAAQ,OAAO;;CAGtB,eAAqB;AACnB,OAAK,QAAQ,OAAO;;CAGtB,gBAA0B;AACxB,SAAO,KAAK,QAAQ,eAAe;;CAGrC,eAAuB,SAAqC;EAC1D,MAAM,cAAc,KAAK,KAAK;AAC9B,SAAO,cAAc,YAAY,QAAQ,GAAG;;CAG9C,kBACE,SACA,MACc;EACd,MAAM,cAAc,KAAK,KAAK;AAC9B,SAAO,cAAc,YAAY,SAAS,KAAK,GAAG;;;;;;CAOpD,qBAA6B,KAAwC;EACnE,MAAM,cAAc,KAAK,KAAK;AAC9B,MAAI,CAAC,YACH,QAAO;EAET,MAAM,SAAS,YAAY,EAAE,SAAS,KAAK,CAAC;AAC5C,MAAI,QAAQ,MACV,QAAO;AAET,MAAI,QAAQ,QACV,QAAO,OAAO;AAEhB,SAAO;;CAGT,cAAsB,SAAgC;EACpD,IAAI,cAAc;AAElB,MADc,QAA+B,SAChC,aAAa;GACxB,MAAM,YAAY,uBAAuB,CAAC,QAAQ,EAAE,EAClD,kBAAkB,KAAK,KAAK,kBAC7B,CAAC;AACF,OAAI,UAAU,WAAW,GAAG;AAC1B,QAAI,KAAK,QAAQ,6BAA6B,CAC5C,MAAK,cAAc;AAErB;;AAEF,iBAAc,UAAU;;EAE1B,MAAM,WAAY,YAAmC;AAErD,MAAI,aAAa,cAAc;GAC7B,MAAM,KAAK,oBAAoB,YAA6D;GAC5F,MAAM,WAAW,KAAK,KAAK,QAAQ,YAAY,GAAG,GAAG,KAAA;AACrD,OAAI,GACF,MAAK,QAAQ,OAAO,GAAG;GAEzB,MAAM,uBAAuB,iCAAiC,aAAa,SAAS;GAGpF,MAAM,SAAS,4BACb,KAAK,eAAe,qBAAqB,EACzC,KAAK,mBACN;GACD,MAAM,YAAY,KAAK,qBACrB,KAAK,kBAAkB,QAAQ;IAC7B,YAAY,MAAM,KAAA;IAClB;IACA,aAAa;IACd,CAAC,CACH;AACD,OAAI,CAAC,UACH;AAEF,UAAO,KAAK,eAAe,4BAA4B,WAAW,KAAK,mBAAmB,CAAU;;EAStG,MAAM,aAAc,YAAwC;EAC5D,MAAM,YACJ,aAAa,eAAe,eAAe,aAAa,eAAe,UACnE,8BAA8B,YAA4D,GAC1F,EAAE;AAQR,MAAI,KAAK,QAAQ,+BAA+B,UAAU,UAAU,OAAO,CACzE,MAAK,cAAc;AAGrB,MAAI,KAAK,QAAQ,8BAA8B,UAAU,OAAO,CAC9D,MAAK,cAAc;EAGrB,MAAM,eAAe,KAAK,qBAAqB,KAAK,eAAe,YAAY,CAAC;AAChF,MAAI,CAAC,aACH;EAEF,MAAM,SAAS,KAAK,eAAe,aAAsB;AAEzD,MAAI,KAAK,KAAK,WACZ,6BAA4B;GAC1B,YAAY,KAAK,KAAK;GACtB,SAAS;GACT,WAAW,OAAO,WAAW,WAAW,SAAS,KAAA;GAClD,CAAC;AAGJ,MAAI,UAAU,SAAS,EACrB,MAAK,QAAQ,eAAe,UAAU;AAGxC,SAAO"}
+{"version":3,"file":"session-tool-result-guard.js","names":[],"sources":["../../../../src/agent/embedded/session-tool-result-guard.ts"],"sourcesContent":["/**\n * Session tool-result guard — wraps a pi `SessionManager.appendMessage` to:\n *   1. cap oversized tool-result text + details so they cannot blow up the next\n *      LLM request (size + persistence limits).\n *   2. track pending tool-call IDs so missing tool results can be synthesised\n *      (some providers refuse a turn that has an orphan tool_use block).\n *   3. drop assistant `toolCall` blocks whose tool name is not in the allowlist\n *      (these would also trigger provider 400s).\n *   4. broadcast `xopc:transcript-row` updates so the gateway UI can stream them.\n *\n * Previously this module shipped with three sibling files\n * (`session-tool-result-state.ts`, `session-raw-append-message.ts`,\n * `session-tool-result-guard-wrapper.ts`). They are now consolidated here as\n * private constructs around the `ToolResultGuard` class. Pi-coding-agent owns\n * the `SessionManager` instance and calls `appendMessage` from inside the\n * runtime, so we still need to monkey-patch that method — but the patched\n * implementation is just `guard.guardedAppend.bind(guard)` and all state lives\n * on the class.\n */\n\nimport type { AgentMessage } from '@earendil-works/pi-agent-core';\nimport type { SessionManager } from '@earendil-works/pi-coding-agent';\n\nimport type { Config } from '../../config/schema.js';\nimport {\n  boundedJsonUtf8Bytes,\n  firstEnumerableOwnKeys,\n  jsonUtf8BytesOrInfinity,\n  type BoundedJsonUtf8Bytes,\n} from '../../infra/json-utf8-bytes.js';\nimport { emitSessionTranscriptUpdate } from '../../session/transcript-events.js';\nimport { normalizeOptionalString } from '../../utils/string-coerce.js';\nimport { formatContextLimitTruncationNotice } from './tool-result-context-guard.js';\nimport {\n  DEFAULT_MAX_LIVE_TOOL_RESULT_CHARS,\n  resolveLiveToolResultMaxChars,\n  truncateToolResultMessage,\n} from './tool-result-truncation.js';\nimport {\n  makeMissingToolResult,\n  sanitizeToolCallInputs,\n} from '../transcript/session-transcript-repair.js';\nimport {\n  extractToolCallsFromAssistant,\n  extractToolResultId,\n} from '../transcript/tool-call-id.js';\n\n// ── Public surface ──────────────────────────────────────────────────────────\n\nexport type BeforeMessageWriteHookEvent = { message: AgentMessage };\nexport type BeforeMessageWriteHookResult =\n  | { block?: boolean; message?: AgentMessage }\n  | undefined;\n\nexport interface ToolResultGuardOptions {\n  /** Optional session key for transcript update broadcasts. */\n  sessionKey?: string;\n  /** Optional transform applied to any message before persistence. */\n  transformMessageForPersistence?: (message: AgentMessage) => AgentMessage;\n  /**\n   * Optional, synchronous transform applied to toolResult messages *before* they are\n   * persisted to the session transcript.\n   */\n  transformToolResultForPersistence?: (\n    message: AgentMessage,\n    meta: { toolCallId?: string; toolName?: string; isSynthetic?: boolean },\n  ) => AgentMessage;\n  /**\n   * Whether to synthesize missing tool results to satisfy strict providers.\n   * Defaults to true.\n   */\n  allowSyntheticToolResults?: boolean;\n  missingToolResultText?: string;\n  /**\n   * Optional set/list of tool names accepted for assistant toolCall/toolUse blocks.\n   * When set, tool calls with unknown names are dropped before persistence.\n   */\n  allowedToolNames?: Iterable<string>;\n  /**\n   * Synchronous hook invoked before any message is written to the session JSONL.\n   * If the hook returns { block: true }, the message is silently dropped.\n   * If it returns { message }, the modified message is written instead.\n   */\n  beforeMessageWriteHook?: (event: BeforeMessageWriteHookEvent) => BeforeMessageWriteHookResult;\n  maxToolResultChars?: number;\n}\n\nexport interface InstallSessionToolResultGuardResult {\n  flushPendingToolResults: () => void;\n  clearPendingToolResults: () => void;\n  getPendingIds: () => string[];\n}\n\n/** Idempotent wrapper that also adds the helper methods consumers expect. */\nexport type GuardedPiTranscriptManager = SessionManager & {\n  flushPendingToolResults?: () => void;\n  clearPendingToolResults?: () => void;\n};\n\n/**\n * Install the guard on a SessionManager and return its control API.\n * Subsequent assistant/toolResult writes by pi-coding-agent flow through the\n * guard transparently.\n */\nexport function installSessionToolResultGuard(\n  sessionManager: SessionManager,\n  opts: ToolResultGuardOptions = {},\n): InstallSessionToolResultGuardResult {\n  const guard = new ToolResultGuard(sessionManager, opts);\n  guard.attach();\n  return {\n    flushPendingToolResults: () => guard.flushPending(),\n    clearPendingToolResults: () => guard.clearPending(),\n    getPendingIds: () => guard.getPendingIds(),\n  };\n}\n\n/**\n * Convenience wrapper used by the embedded runner pool: install the guard\n * (idempotent), pin the size cap from the model context window, and expose\n * `flushPendingToolResults` / `clearPendingToolResults` directly on the\n * SessionManager instance so callers do not need to keep the install result.\n */\nexport function guardSessionManager(\n  sessionManager: SessionManager,\n  opts?: {\n    agentId?: string;\n    sessionKey?: string;\n    config?: Config;\n    contextWindowTokens?: number;\n    allowSyntheticToolResults?: boolean;\n    missingToolResultText?: string;\n    allowedToolNames?: Iterable<string>;\n    transformMessageForPersistence?: (message: AgentMessage) => AgentMessage;\n  },\n): GuardedPiTranscriptManager {\n  if (typeof (sessionManager as GuardedPiTranscriptManager).flushPendingToolResults === 'function') {\n    return sessionManager as GuardedPiTranscriptManager;\n  }\n\n  const result = installSessionToolResultGuard(sessionManager, {\n    sessionKey: opts?.sessionKey,\n    transformMessageForPersistence: opts?.transformMessageForPersistence,\n    allowSyntheticToolResults: opts?.allowSyntheticToolResults,\n    missingToolResultText: opts?.missingToolResultText,\n    allowedToolNames: opts?.allowedToolNames,\n    maxToolResultChars:\n      typeof opts?.contextWindowTokens === 'number'\n        ? resolveLiveToolResultMaxChars({\n            contextWindowTokens: opts.contextWindowTokens,\n            cfg: opts?.config,\n            agentId: opts?.agentId,\n          })\n        : undefined,\n  });\n  const tagged = sessionManager as GuardedPiTranscriptManager;\n  tagged.flushPendingToolResults = result.flushPendingToolResults;\n  tagged.clearPendingToolResults = result.clearPendingToolResults;\n  return tagged;\n}\n\n/**\n * Recover the original (un-guarded) appendMessage for a session manager.\n * Useful for callers that need a low-level \"bypass the guard\" write path.\n */\nexport function getRawSessionAppendMessage(\n  sessionManager: SessionManager,\n): SessionManager['appendMessage'] {\n  const stored = (sessionManager as SessionManagerWithRawAppend)[RAW_APPEND_MESSAGE];\n  return stored ?? sessionManager.appendMessage.bind(sessionManager);\n}\n\n// ── Internal: persistence + truncation helpers ──────────────────────────────\n\nfunction resolveMaxToolResultChars(opts: { maxToolResultChars?: number }): number {\n  return Math.max(1, opts.maxToolResultChars ?? DEFAULT_MAX_LIVE_TOOL_RESULT_CHARS);\n}\n\nfunction capToolResultSize(msg: AgentMessage, maxChars: number): AgentMessage {\n  if ((msg as { role?: string }).role !== 'toolResult') {\n    return msg;\n  }\n  return truncateToolResultMessage(msg, maxChars, {\n    suffix: (truncatedChars) => formatContextLimitTruncationNotice(truncatedChars),\n    minKeepChars: 2_000,\n  });\n}\n\n// `details` is runtime/UI metadata, not model-visible tool output. Keep the\n// session JSONL useful for debugging without letting metadata blobs dominate\n// disk, replay repair, transcript broadcasts, or future tooling that reads raw\n// sessions. Model-visible text belongs in tool result `content`.\nconst MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES = 8_192;\nconst MAX_PERSISTED_DETAIL_STRING_CHARS = 2_000;\nconst MAX_PERSISTED_DETAIL_SESSION_COUNT = 10;\nconst MAX_PERSISTED_DETAIL_FALLBACK_STRING_CHARS = 200;\n\nfunction originalDetailsSizeFields(size: BoundedJsonUtf8Bytes): Record<string, number> {\n  return size.complete\n    ? { originalDetailsBytes: size.bytes }\n    : { originalDetailsBytesAtLeast: size.bytes };\n}\n\nfunction truncatePersistedDetailString(\n  value: string,\n  maxChars = MAX_PERSISTED_DETAIL_STRING_CHARS,\n): string {\n  if (value.length <= maxChars) {\n    return value;\n  }\n  return `${value.slice(0, maxChars)}\\n\\n[xopc persisted detail truncated: ${\n    value.length - maxChars\n  } chars omitted]`;\n}\n\nfunction sanitizePersistedSessionDetail(value: unknown): unknown {\n  if (!value || typeof value !== 'object') {\n    return value;\n  }\n  const src = value as Record<string, unknown>;\n  const out: Record<string, unknown> = {};\n  for (const key of [\n    'sessionId',\n    'status',\n    'pid',\n    'startedAt',\n    'endedAt',\n    'runtimeMs',\n    'cwd',\n    'name',\n    'truncated',\n    'exitCode',\n    'exitSignal',\n  ]) {\n    const field = src[key];\n    if (field !== undefined) {\n      out[key] = typeof field === 'string' ? truncatePersistedDetailString(field, 500) : field;\n    }\n  }\n  if (typeof src.command === 'string') {\n    out.command = truncatePersistedDetailString(src.command, 500);\n  }\n  return out;\n}\n\nfunction buildPersistedDetailsFallback(\n  src: Record<string, unknown> | undefined,\n  originalSize: BoundedJsonUtf8Bytes,\n  sanitizedBytes?: number,\n): Record<string, unknown> {\n  const fallback: Record<string, unknown> = {\n    persistedDetailsTruncated: true,\n    finalDetailsTruncated: true,\n    ...originalDetailsSizeFields(originalSize),\n  };\n  if (sanitizedBytes !== undefined) {\n    fallback.sanitizedDetailsBytes = sanitizedBytes;\n  }\n  if (src) {\n    fallback.originalDetailKeys = firstEnumerableOwnKeys(src, 40);\n    for (const key of ['status', 'sessionId', 'pid', 'exitCode', 'exitSignal', 'truncated']) {\n      const field = src[key];\n      if (field !== undefined) {\n        fallback[key] =\n          typeof field === 'string'\n            ? truncatePersistedDetailString(field, MAX_PERSISTED_DETAIL_FALLBACK_STRING_CHARS)\n            : field;\n      }\n    }\n  }\n  return fallback;\n}\n\nfunction enforcePersistedDetailsByteCap(\n  value: Record<string, unknown>,\n  src: Record<string, unknown> | undefined,\n  originalSize: BoundedJsonUtf8Bytes,\n): Record<string, unknown> {\n  const sanitizedBytes = jsonUtf8BytesOrInfinity(value);\n  if (sanitizedBytes <= MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES) {\n    return value;\n  }\n  const fallback = buildPersistedDetailsFallback(src, originalSize, sanitizedBytes);\n  if (jsonUtf8BytesOrInfinity(fallback) <= MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES) {\n    return fallback;\n  }\n  return {\n    persistedDetailsTruncated: true,\n    finalDetailsTruncated: true,\n    ...originalDetailsSizeFields(originalSize),\n    sanitizedDetailsBytes: sanitizedBytes,\n  };\n}\n\nfunction sanitizeToolResultDetailsForPersistence(details: unknown): unknown {\n  if (details === undefined || details === null) {\n    return details;\n  }\n  const originalSize = boundedJsonUtf8Bytes(details, MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES);\n  if (originalSize.complete && originalSize.bytes <= MAX_PERSISTED_TOOL_RESULT_DETAILS_BYTES) {\n    return details;\n  }\n  if (typeof details !== 'object') {\n    return enforcePersistedDetailsByteCap(\n      {\n        persistedDetailsTruncated: true,\n        ...originalDetailsSizeFields(originalSize),\n        valueType: typeof details,\n      },\n      undefined,\n      originalSize,\n    );\n  }\n  const src = details as Record<string, unknown>;\n  const out: Record<string, unknown> = {\n    persistedDetailsTruncated: true,\n    ...originalDetailsSizeFields(originalSize),\n    originalDetailKeys: firstEnumerableOwnKeys(src, 40),\n  };\n  for (const key of [\n    'status',\n    'sessionId',\n    'pid',\n    'startedAt',\n    'endedAt',\n    'cwd',\n    'name',\n    'exitCode',\n    'exitSignal',\n    'retryInMs',\n    'total',\n    'totalLines',\n    'totalChars',\n    'truncated',\n    'fullOutputPath',\n    'truncation',\n  ]) {\n    const field = src[key];\n    if (field !== undefined) {\n      out[key] = typeof field === 'string' ? truncatePersistedDetailString(field) : field;\n    }\n  }\n  if (typeof src.tail === 'string') {\n    out.tail = truncatePersistedDetailString(src.tail);\n  }\n  if (Array.isArray(src.sessions)) {\n    out.sessions = src.sessions\n      .slice(0, MAX_PERSISTED_DETAIL_SESSION_COUNT)\n      .map(sanitizePersistedSessionDetail);\n    if (src.sessions.length > MAX_PERSISTED_DETAIL_SESSION_COUNT) {\n      out.sessionsTruncated = src.sessions.length - MAX_PERSISTED_DETAIL_SESSION_COUNT;\n    }\n  }\n  return enforcePersistedDetailsByteCap(out, src, originalSize);\n}\n\nfunction capToolResultDetails(msg: AgentMessage): AgentMessage {\n  if ((msg as { role?: string }).role !== 'toolResult') {\n    return msg;\n  }\n  const details = (msg as { details?: unknown }).details;\n  const sanitizedDetails = sanitizeToolResultDetailsForPersistence(details);\n  if (sanitizedDetails === details) {\n    return msg;\n  }\n  const next = { ...msg } as AgentMessage & { details?: unknown };\n  next.details = sanitizedDetails;\n  return next;\n}\n\nfunction capToolResultForPersistence(msg: AgentMessage, maxChars: number): AgentMessage {\n  return capToolResultDetails(capToolResultSize(msg, maxChars));\n}\n\nfunction normalizePersistedToolResultName(\n  message: AgentMessage,\n  fallbackName?: string,\n): AgentMessage {\n  if ((message as { role?: unknown }).role !== 'toolResult') {\n    return message;\n  }\n  const toolResult = message as Extract<AgentMessage, { role: 'toolResult' }>;\n  const rawToolName = (toolResult as { toolName?: unknown }).toolName;\n  const normalizedToolName = normalizeOptionalString(rawToolName);\n  if (normalizedToolName) {\n    if (rawToolName === normalizedToolName) {\n      return toolResult;\n    }\n    return { ...toolResult, toolName: normalizedToolName };\n  }\n\n  const normalizedFallback = normalizeOptionalString(fallbackName);\n  if (normalizedFallback) {\n    return { ...toolResult, toolName: normalizedFallback };\n  }\n\n  if (typeof rawToolName === 'string') {\n    return { ...toolResult, toolName: 'unknown' };\n  }\n  return toolResult;\n}\n\n// ── Internal: raw-append symbol storage ─────────────────────────────────────\n\nconst RAW_APPEND_MESSAGE = Symbol('xopc.session.rawAppendMessage');\n\ntype SessionManagerWithRawAppend = SessionManager & {\n  [RAW_APPEND_MESSAGE]?: SessionManager['appendMessage'];\n};\n\nfunction rememberOriginalAppend(\n  sessionManager: SessionManager,\n): SessionManager['appendMessage'] {\n  const tagged = sessionManager as SessionManagerWithRawAppend;\n  const stored = tagged[RAW_APPEND_MESSAGE];\n  if (stored) {\n    return stored;\n  }\n  const original = sessionManager.appendMessage.bind(sessionManager);\n  tagged[RAW_APPEND_MESSAGE] = original;\n  return original;\n}\n\n// ── Internal: pending tool-call state ──────────────────────────────────────\n\ntype PendingToolCall = { id: string; name?: string };\n\nclass PendingToolCallTracker {\n  private readonly pending = new Map<string, string | undefined>();\n\n  size(): number {\n    return this.pending.size;\n  }\n\n  getToolName(id: string): string | undefined {\n    return this.pending.get(id);\n  }\n\n  delete(id: string): void {\n    this.pending.delete(id);\n  }\n\n  clear(): void {\n    this.pending.clear();\n  }\n\n  trackToolCalls(calls: readonly PendingToolCall[]): void {\n    for (const call of calls) {\n      this.pending.set(call.id, call.name);\n    }\n  }\n\n  entries(): IterableIterator<[string, string | undefined]> {\n    return this.pending.entries();\n  }\n\n  getPendingIds(): string[] {\n    return Array.from(this.pending.keys());\n  }\n\n  shouldFlushForSanitizedDrop(): boolean {\n    return this.pending.size > 0;\n  }\n\n  shouldFlushBeforeNonToolResult(nextRole: unknown, toolCallCount: number): boolean {\n    return this.pending.size > 0 && (toolCallCount === 0 || nextRole !== 'assistant');\n  }\n\n  shouldFlushBeforeNewToolCalls(toolCallCount: number): boolean {\n    return this.pending.size > 0 && toolCallCount > 0;\n  }\n}\n\n// ── ToolResultGuard class ──────────────────────────────────────────────────\n\nclass ToolResultGuard {\n  private readonly sessionManager: SessionManager;\n  private readonly opts: ToolResultGuardOptions;\n  private readonly pending = new PendingToolCallTracker();\n  private readonly originalAppend: SessionManager['appendMessage'];\n  private readonly allowSyntheticToolResults: boolean;\n  private readonly maxToolResultChars: number;\n\n  constructor(sessionManager: SessionManager, opts: ToolResultGuardOptions) {\n    this.sessionManager = sessionManager;\n    this.opts = opts;\n    this.originalAppend = rememberOriginalAppend(sessionManager);\n    this.allowSyntheticToolResults = opts.allowSyntheticToolResults ?? true;\n    this.maxToolResultChars = resolveMaxToolResultChars(opts);\n  }\n\n  /** Monkey-patch the session manager so pi-coding-agent's internal appendMessage flows through us. */\n  attach(): void {\n    const bound = this.guardedAppend.bind(this);\n    this.sessionManager.appendMessage = bound as SessionManager['appendMessage'];\n  }\n\n  flushPending(): void {\n    if (this.pending.size() === 0) {\n      return;\n    }\n    if (this.allowSyntheticToolResults) {\n      for (const [id, name] of this.pending.entries()) {\n        const synthetic = makeMissingToolResult({\n          toolCallId: id,\n          toolName: name,\n          text: this.opts.missingToolResultText,\n        });\n        const flushed = this.applyBeforeWriteHook(\n          this.persistToolResult(this.persistMessage(synthetic), {\n            toolCallId: id,\n            toolName: name,\n            isSynthetic: true,\n          }),\n        );\n        if (flushed) {\n          this.originalAppend(capToolResultForPersistence(flushed, this.maxToolResultChars) as never);\n        }\n      }\n    }\n    this.pending.clear();\n  }\n\n  clearPending(): void {\n    this.pending.clear();\n  }\n\n  getPendingIds(): string[] {\n    return this.pending.getPendingIds();\n  }\n\n  private persistMessage(message: AgentMessage): AgentMessage {\n    const transformer = this.opts.transformMessageForPersistence;\n    return transformer ? transformer(message) : message;\n  }\n\n  private persistToolResult(\n    message: AgentMessage,\n    meta: { toolCallId?: string; toolName?: string; isSynthetic?: boolean },\n  ): AgentMessage {\n    const transformer = this.opts.transformToolResultForPersistence;\n    return transformer ? transformer(message, meta) : message;\n  }\n\n  /**\n   * Run the before_message_write hook. Returns the (possibly modified) message,\n   * or null if the message should be blocked.\n   */\n  private applyBeforeWriteHook(msg: AgentMessage): AgentMessage | null {\n    const beforeWrite = this.opts.beforeMessageWriteHook;\n    if (!beforeWrite) {\n      return msg;\n    }\n    const result = beforeWrite({ message: msg });\n    if (result?.block) {\n      return null;\n    }\n    if (result?.message) {\n      return result.message;\n    }\n    return msg;\n  }\n\n  private guardedAppend(message: AgentMessage): unknown {\n    let nextMessage = message;\n    const role = (message as { role?: unknown }).role;\n    if (role === 'assistant') {\n      const sanitized = sanitizeToolCallInputs([message], {\n        allowedToolNames: this.opts.allowedToolNames,\n      });\n      if (sanitized.length === 0) {\n        if (this.pending.shouldFlushForSanitizedDrop()) {\n          this.flushPending();\n        }\n        return undefined;\n      }\n      nextMessage = sanitized[0];\n    }\n    const nextRole = (nextMessage as { role?: unknown }).role;\n\n    if (nextRole === 'toolResult') {\n      const id = extractToolResultId(nextMessage as Extract<AgentMessage, { role: 'toolResult' }>);\n      const toolName = id ? this.pending.getToolName(id) : undefined;\n      if (id) {\n        this.pending.delete(id);\n      }\n      const normalizedToolResult = normalizePersistedToolResultName(nextMessage, toolName);\n      // Apply hard size cap before persistence to prevent oversized tool results\n      // from consuming the entire context window on subsequent LLM calls.\n      const capped = capToolResultForPersistence(\n        this.persistMessage(normalizedToolResult),\n        this.maxToolResultChars,\n      );\n      const persisted = this.applyBeforeWriteHook(\n        this.persistToolResult(capped, {\n          toolCallId: id ?? undefined,\n          toolName,\n          isSynthetic: false,\n        }),\n      );\n      if (!persisted) {\n        return undefined;\n      }\n      const finalToolResult = capToolResultForPersistence(persisted, this.maxToolResultChars);\n      const result = this.originalAppend(finalToolResult as never);\n\n      if (this.opts.sessionKey) {\n        emitSessionTranscriptUpdate({\n          sessionKey: this.opts.sessionKey,\n          message: finalToolResult,\n          messageId: typeof result === 'string' ? result : undefined,\n        });\n      }\n\n      return result;\n    }\n\n    // Skip tool call extraction for aborted/errored assistant messages.\n    // When stopReason is \"error\" or \"aborted\", the tool_use blocks may be incomplete\n    // and should not have synthetic tool_results created. Creating synthetic results\n    // for incomplete tool calls causes API 400 errors:\n    // \"unexpected tool_use_id found in tool_result blocks\"\n    // This matches the behavior in repairToolUseResultPairing (session-transcript-repair.ts)\n    const stopReason = (nextMessage as { stopReason?: string }).stopReason;\n    const toolCalls =\n      nextRole === 'assistant' && stopReason !== 'aborted' && stopReason !== 'error'\n        ? extractToolCallsFromAssistant(nextMessage as Extract<AgentMessage, { role: 'assistant' }>)\n        : [];\n\n    // Always clear pending tool call state before appending non-tool-result messages.\n    // flushPendingToolResults() only inserts synthetic results when allowSyntheticToolResults\n    // is true; it always clears the pending map. Without this, providers that disable\n    // synthetic results (e.g. OpenAI) accumulate stale pending state when a user message\n    // interrupts in-flight tool calls, leaving orphaned tool_use blocks in the transcript\n    // that cause API 400 errors on subsequent requests.\n    if (this.pending.shouldFlushBeforeNonToolResult(nextRole, toolCalls.length)) {\n      this.flushPending();\n    }\n    // If new tool calls arrive while older ones are pending, flush the old ones first.\n    if (this.pending.shouldFlushBeforeNewToolCalls(toolCalls.length)) {\n      this.flushPending();\n    }\n\n    const finalMessage = this.applyBeforeWriteHook(this.persistMessage(nextMessage));\n    if (!finalMessage) {\n      return undefined;\n    }\n    const result = this.originalAppend(finalMessage as never);\n\n    if (this.opts.sessionKey) {\n      emitSessionTranscriptUpdate({\n        sessionKey: this.opts.sessionKey,\n        message: finalMessage,\n        messageId: typeof result === 'string' ? result : undefined,\n      });\n    }\n\n    if (toolCalls.length > 0) {\n      this.pending.trackToolCalls(toolCalls);\n    }\n\n    return result;\n  }\n}\n"],"mappings":";;;;;;;;oBA+BuE;;;;;;AAyEvE,SAAgB,8BACd,gBACA,OAA+B,EAAE,EACI;CACrC,MAAM,QAAQ,IAAI,gBAAgB,gBAAgB,KAAK;AACvD,OAAM,QAAQ;AACd,QAAO;EACL,+BAA+B,MAAM,cAAc;EACnD,+BAA+B,MAAM,cAAc;EACnD,qBAAqB,MAAM,eAAe;EAC3C;;;;;;;;AASH,SAAgB,oBACd,gBACA,MAU4B;AAC5B,KAAI,OAAQ,eAA8C,4BAA4B,WACpF,QAAO;CAGT,MAAM,SAAS,8BAA8B,gBAAgB;EAC3D,YAAY,MAAM;EAClB,gCAAgC,MAAM;EACtC,2BAA2B,MAAM;EACjC,uBAAuB,MAAM;EAC7B,kBAAkB,MAAM;EACxB,oBACE,OAAO,MAAM,wBAAwB,WACjC,8BAA8B;GAC5B,qBAAqB,KAAK;GAC1B,KAAK,MAAM;GACX,SAAS,MAAM;GAChB,CAAC,GACF,KAAA;EACP,CAAC;CACF,MAAM,SAAS;AACf,QAAO,0BAA0B,OAAO;AACxC,QAAO,0BAA0B,OAAO;AACxC,QAAO;;;;;;AAOT,SAAgB,2BACd,gBACiC;AAEjC,QADgB,eAA+C,uBAC9C,eAAe,cAAc,KAAK,eAAe;;AAKpE,SAAS,0BAA0B,MAA+C;AAChF,QAAO,KAAK,IAAI,GAAG,KAAK,sBAAA,KAAyD;;AAGnF,SAAS,kBAAkB,KAAmB,UAAgC;AAC5E,KAAK,IAA0B,SAAS,aACtC,QAAO;AAET,QAAO,0BAA0B,KAAK,UAAU;EAC9C,SAAS,mBAAmB,mCAAmC,eAAe;EAC9E,cAAc;EACf,CAAC;;AAOJ,MAAM,0CAA0C;AAChD,MAAM,oCAAoC;AAC1C,MAAM,qCAAqC;AAC3C,MAAM,6CAA6C;AAEnD,SAAS,0BAA0B,MAAoD;AACrF,QAAO,KAAK,WACR,EAAE,sBAAsB,KAAK,OAAO,GACpC,EAAE,6BAA6B,KAAK,OAAO;;AAGjD,SAAS,8BACP,OACA,WAAW,mCACH;AACR,KAAI,MAAM,UAAU,SAClB,QAAO;AAET,QAAO,GAAG,MAAM,MAAM,GAAG,SAAS,CAAC,wCACjC,MAAM,SAAS,SAChB;;AAGH,SAAS,+BAA+B,OAAyB;AAC/D,KAAI,CAAC,SAAS,OAAO,UAAU,SAC7B,QAAO;CAET,MAAM,MAAM;CACZ,MAAM,MAA+B,EAAE;AACvC,MAAK,MAAM,OAAO;EAChB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACD,EAAE;EACD,MAAM,QAAQ,IAAI;AAClB,MAAI,UAAU,KAAA,EACZ,KAAI,OAAO,OAAO,UAAU,WAAW,8BAA8B,OAAO,IAAI,GAAG;;AAGvF,KAAI,OAAO,IAAI,YAAY,SACzB,KAAI,UAAU,8BAA8B,IAAI,SAAS,IAAI;AAE/D,QAAO;;AAGT,SAAS,8BACP,KACA,cACA,gBACyB;CACzB,MAAM,WAAoC;EACxC,2BAA2B;EAC3B,uBAAuB;EACvB,GAAG,0BAA0B,aAAa;EAC3C;AACD,KAAI,mBAAmB,KAAA,EACrB,UAAS,wBAAwB;AAEnC,KAAI,KAAK;AACP,WAAS,qBAAqB,uBAAuB,KAAK,GAAG;AAC7D,OAAK,MAAM,OAAO;GAAC;GAAU;GAAa;GAAO;GAAY;GAAc;GAAY,EAAE;GACvF,MAAM,QAAQ,IAAI;AAClB,OAAI,UAAU,KAAA,EACZ,UAAS,OACP,OAAO,UAAU,WACb,8BAA8B,OAAO,2CAA2C,GAChF;;;AAIZ,QAAO;;AAGT,SAAS,+BACP,OACA,KACA,cACyB;CACzB,MAAM,iBAAiB,wBAAwB,MAAM;AACrD,KAAI,kBAAkB,wCACpB,QAAO;CAET,MAAM,WAAW,8BAA8B,KAAK,cAAc,eAAe;AACjF,KAAI,wBAAwB,SAAS,IAAI,wCACvC,QAAO;AAET,QAAO;EACL,2BAA2B;EAC3B,uBAAuB;EACvB,GAAG,0BAA0B,aAAa;EAC1C,uBAAuB;EACxB;;AAGH,SAAS,wCAAwC,SAA2B;AAC1E,KAAI,YAAY,KAAA,KAAa,YAAY,KACvC,QAAO;CAET,MAAM,eAAe,qBAAqB,SAAS,wCAAwC;AAC3F,KAAI,aAAa,YAAY,aAAa,SAAS,wCACjD,QAAO;AAET,KAAI,OAAO,YAAY,SACrB,QAAO,+BACL;EACE,2BAA2B;EAC3B,GAAG,0BAA0B,aAAa;EAC1C,WAAW,OAAO;EACnB,EACD,KAAA,GACA,aACD;CAEH,MAAM,MAAM;CACZ,MAAM,MAA+B;EACnC,2BAA2B;EAC3B,GAAG,0BAA0B,aAAa;EAC1C,oBAAoB,uBAAuB,KAAK,GAAG;EACpD;AACD,MAAK,MAAM,OAAO;EAChB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACD,EAAE;EACD,MAAM,QAAQ,IAAI;AAClB,MAAI,UAAU,KAAA,EACZ,KAAI,OAAO,OAAO,UAAU,WAAW,8BAA8B,MAAM,GAAG;;AAGlF,KAAI,OAAO,IAAI,SAAS,SACtB,KAAI,OAAO,8BAA8B,IAAI,KAAK;AAEpD,KAAI,MAAM,QAAQ,IAAI,SAAS,EAAE;AAC/B,MAAI,WAAW,IAAI,SAChB,MAAM,GAAG,mCAAmC,CAC5C,IAAI,+BAA+B;AACtC,MAAI,IAAI,SAAS,SAAS,mCACxB,KAAI,oBAAoB,IAAI,SAAS,SAAS;;AAGlD,QAAO,+BAA+B,KAAK,KAAK,aAAa;;AAG/D,SAAS,qBAAqB,KAAiC;AAC7D,KAAK,IAA0B,SAAS,aACtC,QAAO;CAET,MAAM,UAAW,IAA8B;CAC/C,MAAM,mBAAmB,wCAAwC,QAAQ;AACzE,KAAI,qBAAqB,QACvB,QAAO;CAET,MAAM,OAAO,EAAE,GAAG,KAAK;AACvB,MAAK,UAAU;AACf,QAAO;;AAGT,SAAS,4BAA4B,KAAmB,UAAgC;AACtF,QAAO,qBAAqB,kBAAkB,KAAK,SAAS,CAAC;;AAG/D,SAAS,iCACP,SACA,cACc;AACd,KAAK,QAA+B,SAAS,aAC3C,QAAO;CAET,MAAM,aAAa;CACnB,MAAM,cAAe,WAAsC;CAC3D,MAAM,qBAAqB,wBAAwB,YAAY;AAC/D,KAAI,oBAAoB;AACtB,MAAI,gBAAgB,mBAClB,QAAO;AAET,SAAO;GAAE,GAAG;GAAY,UAAU;GAAoB;;CAGxD,MAAM,qBAAqB,wBAAwB,aAAa;AAChE,KAAI,mBACF,QAAO;EAAE,GAAG;EAAY,UAAU;EAAoB;AAGxD,KAAI,OAAO,gBAAgB,SACzB,QAAO;EAAE,GAAG;EAAY,UAAU;EAAW;AAE/C,QAAO;;AAKT,MAAM,qBAAqB,OAAO,gCAAgC;AAMlE,SAAS,uBACP,gBACiC;CACjC,MAAM,SAAS;CACf,MAAM,SAAS,OAAO;AACtB,KAAI,OACF,QAAO;CAET,MAAM,WAAW,eAAe,cAAc,KAAK,eAAe;AAClE,QAAO,sBAAsB;AAC7B,QAAO;;AAOT,IAAM,yBAAN,MAA6B;CAC3B,0BAA2B,IAAI,KAAiC;CAEhE,OAAe;AACb,SAAO,KAAK,QAAQ;;CAGtB,YAAY,IAAgC;AAC1C,SAAO,KAAK,QAAQ,IAAI,GAAG;;CAG7B,OAAO,IAAkB;AACvB,OAAK,QAAQ,OAAO,GAAG;;CAGzB,QAAc;AACZ,OAAK,QAAQ,OAAO;;CAGtB,eAAe,OAAyC;AACtD,OAAK,MAAM,QAAQ,MACjB,MAAK,QAAQ,IAAI,KAAK,IAAI,KAAK,KAAK;;CAIxC,UAA0D;AACxD,SAAO,KAAK,QAAQ,SAAS;;CAG/B,gBAA0B;AACxB,SAAO,MAAM,KAAK,KAAK,QAAQ,MAAM,CAAC;;CAGxC,8BAAuC;AACrC,SAAO,KAAK,QAAQ,OAAO;;CAG7B,+BAA+B,UAAmB,eAAgC;AAChF,SAAO,KAAK,QAAQ,OAAO,MAAM,kBAAkB,KAAK,aAAa;;CAGvE,8BAA8B,eAAgC;AAC5D,SAAO,KAAK,QAAQ,OAAO,KAAK,gBAAgB;;;AAMpD,IAAM,kBAAN,MAAsB;CACpB;CACA;CACA,UAA2B,IAAI,wBAAwB;CACvD;CACA;CACA;CAEA,YAAY,gBAAgC,MAA8B;AACxE,OAAK,iBAAiB;AACtB,OAAK,OAAO;AACZ,OAAK,iBAAiB,uBAAuB,eAAe;AAC5D,OAAK,4BAA4B,KAAK,6BAA6B;AACnE,OAAK,qBAAqB,0BAA0B,KAAK;;;CAI3D,SAAe;EACb,MAAM,QAAQ,KAAK,cAAc,KAAK,KAAK;AAC3C,OAAK,eAAe,gBAAgB;;CAGtC,eAAqB;AACnB,MAAI,KAAK,QAAQ,MAAM,KAAK,EAC1B;AAEF,MAAI,KAAK,0BACP,MAAK,MAAM,CAAC,IAAI,SAAS,KAAK,QAAQ,SAAS,EAAE;GAC/C,MAAM,YAAY,sBAAsB;IACtC,YAAY;IACZ,UAAU;IACV,MAAM,KAAK,KAAK;IACjB,CAAC;GACF,MAAM,UAAU,KAAK,qBACnB,KAAK,kBAAkB,KAAK,eAAe,UAAU,EAAE;IACrD,YAAY;IACZ,UAAU;IACV,aAAa;IACd,CAAC,CACH;AACD,OAAI,QACF,MAAK,eAAe,4BAA4B,SAAS,KAAK,mBAAmB,CAAU;;AAIjG,OAAK,QAAQ,OAAO;;CAGtB,eAAqB;AACnB,OAAK,QAAQ,OAAO;;CAGtB,gBAA0B;AACxB,SAAO,KAAK,QAAQ,eAAe;;CAGrC,eAAuB,SAAqC;EAC1D,MAAM,cAAc,KAAK,KAAK;AAC9B,SAAO,cAAc,YAAY,QAAQ,GAAG;;CAG9C,kBACE,SACA,MACc;EACd,MAAM,cAAc,KAAK,KAAK;AAC9B,SAAO,cAAc,YAAY,SAAS,KAAK,GAAG;;;;;;CAOpD,qBAA6B,KAAwC;EACnE,MAAM,cAAc,KAAK,KAAK;AAC9B,MAAI,CAAC,YACH,QAAO;EAET,MAAM,SAAS,YAAY,EAAE,SAAS,KAAK,CAAC;AAC5C,MAAI,QAAQ,MACV,QAAO;AAET,MAAI,QAAQ,QACV,QAAO,OAAO;AAEhB,SAAO;;CAGT,cAAsB,SAAgC;EACpD,IAAI,cAAc;AAElB,MADc,QAA+B,SAChC,aAAa;GACxB,MAAM,YAAY,uBAAuB,CAAC,QAAQ,EAAE,EAClD,kBAAkB,KAAK,KAAK,kBAC7B,CAAC;AACF,OAAI,UAAU,WAAW,GAAG;AAC1B,QAAI,KAAK,QAAQ,6BAA6B,CAC5C,MAAK,cAAc;AAErB;;AAEF,iBAAc,UAAU;;EAE1B,MAAM,WAAY,YAAmC;AAErD,MAAI,aAAa,cAAc;GAC7B,MAAM,KAAK,oBAAoB,YAA6D;GAC5F,MAAM,WAAW,KAAK,KAAK,QAAQ,YAAY,GAAG,GAAG,KAAA;AACrD,OAAI,GACF,MAAK,QAAQ,OAAO,GAAG;GAEzB,MAAM,uBAAuB,iCAAiC,aAAa,SAAS;GAGpF,MAAM,SAAS,4BACb,KAAK,eAAe,qBAAqB,EACzC,KAAK,mBACN;GACD,MAAM,YAAY,KAAK,qBACrB,KAAK,kBAAkB,QAAQ;IAC7B,YAAY,MAAM,KAAA;IAClB;IACA,aAAa;IACd,CAAC,CACH;AACD,OAAI,CAAC,UACH;GAEF,MAAM,kBAAkB,4BAA4B,WAAW,KAAK,mBAAmB;GACvF,MAAM,SAAS,KAAK,eAAe,gBAAyB;AAE5D,OAAI,KAAK,KAAK,WACZ,6BAA4B;IAC1B,YAAY,KAAK,KAAK;IACtB,SAAS;IACT,WAAW,OAAO,WAAW,WAAW,SAAS,KAAA;IAClD,CAAC;AAGJ,UAAO;;EAST,MAAM,aAAc,YAAwC;EAC5D,MAAM,YACJ,aAAa,eAAe,eAAe,aAAa,eAAe,UACnE,8BAA8B,YAA4D,GAC1F,EAAE;AAQR,MAAI,KAAK,QAAQ,+BAA+B,UAAU,UAAU,OAAO,CACzE,MAAK,cAAc;AAGrB,MAAI,KAAK,QAAQ,8BAA8B,UAAU,OAAO,CAC9D,MAAK,cAAc;EAGrB,MAAM,eAAe,KAAK,qBAAqB,KAAK,eAAe,YAAY,CAAC;AAChF,MAAI,CAAC,aACH;EAEF,MAAM,SAAS,KAAK,eAAe,aAAsB;AAEzD,MAAI,KAAK,KAAK,WACZ,6BAA4B;GAC1B,YAAY,KAAK,KAAK;GACtB,SAAS;GACT,WAAW,OAAO,WAAW,WAAW,SAAS,KAAA;GAClD,CAAC;AAGJ,MAAI,UAAU,SAAS,EACrB,MAAK,QAAQ,eAAe,UAAU;AAGxC,SAAO"}

package/dist/src/agent/goals/checklist-judge.d.ts

@@ -8,6 +8,7 @@ export type DecomposeChecklistResult = {
 };
 export declare function decomposeGoalChecklist(opts: {
     goal: string;
+    existingChecklist?: string;
     judgeModelRef: string;
     signal?: AbortSignal;
     judgeTimeoutMs?: number;

package/dist/src/agent/goals/checklist-judge.js

@@ -8,7 +8,7 @@ init_providers();
 const GOAL_TRUNC = 4e3;
 const RESPONSE_SNIPPET = 4e3;
 const HISTORY_TRUNC = 24e3;
-const DECOMPOSE_SYSTEM = "You break a user goal into an EXTREMELY detailed checklist of concrete, verifiable completion criteria. Bias toward more items. Each item is one factual statement about finished work.\n\nReply ONLY with a single JSON object on one line:\n{\"items\":[{\"text\":\"...\"},...]}\nUse at least 3 items when the goal warrants it.";
+const DECOMPOSE_SYSTEM = "You break a user goal into an EXTREMELY detailed checklist of concrete, verifiable completion criteria. Bias toward more items. Each item is one factual statement about finished work.\n\nIf existing acceptance criteria are provided, return only missing supplemental criteria. Do not repeat, rewrite, or replace existing criteria.\n\nReply ONLY with a single JSON object on one line:\n{\"items\":[{\"text\":\"...\"},...]}\nUse at least 3 items when the goal warrants it.";
 const DECOMPOSE_USER = "Goal:\n{goal}\n\nReturn JSON only.";
 const EVALUATE_CHECKLIST_SYSTEM = "You evaluate an autonomous agent's progress on a goal with a numbered checklist.\nFor each pending item, decide if evidence in the agent snippet and/or history excerpt shows it is satisfied.\nFlip pending→completed only with clear evidence; pending→impossible only if truly unachievable here.\nDo not regress completed/impossible items — omit them from updates.\n\nReply ONLY with one JSON object on one line:\n{\"verdict\":\"continue|done|blocked|needs_input\",\"confidence\":0.0,\"reason\":\"...\",\"nextAction\":\"...\",\"missingEvidence\":[\"...\"],\"userQuestion\":\"...\",\"updates\":[{\"index\":1,\"status\":\"completed|impossible\",\"evidence\":\"short citation\"}],\"new_items\":[{\"text\":\"...\"}]}\nUse verdict=done only when all checklist criteria are complete/impossible. Use needs_input when the next step requires user input. Use blocked when external progress is impossible.";
 function extractJsonObject(raw) {
@@ -52,9 +52,10 @@ async function decomposeGoalChecklist(opts) {
 	const timer = setTimeout(() => controller.abort(), timeoutMs);
 	const merged = opts.signal ? AbortSignal.any([opts.signal, controller.signal]) : controller.signal;
 	try {
+		const existingChecklist = opts.existingChecklist?.trim();
 		const user = {
 			role: "user",
-			content: `${DECOMPOSE_SYSTEM}\n\n${DECOMPOSE_USER.replace("{goal}", truncateGoalText(goal, GOAL_TRUNC))}` + judgeResponseLanguageNote(locale),
+			content: `${DECOMPOSE_SYSTEM}\n\n${DECOMPOSE_USER.replace("{goal}", truncateGoalText(goal, GOAL_TRUNC)) + (existingChecklist ? `\n\nExisting acceptance criteria:\n${truncateGoalText(existingChecklist, GOAL_TRUNC)}\n\nReturn only missing supplemental criteria as JSON.` : "")}` + judgeResponseLanguageNote(locale),
 			timestamp: Date.now()
 		};
 		const apiKey = await resolveGoalJudgeApiKey(model);

package/dist/src/agent/goals/checklist-judge.js.map

@@ -1 +1 @@
-{"version":3,"file":"checklist-judge.js","names":[],"sources":["../../../../src/agent/goals/checklist-judge.ts"],"sourcesContent":["import type { UserMessage } from '@earendil-works/pi-ai';\nimport { complete } from '@earendil-works/pi-ai';\n\nimport { resolveModel } from '../../providers/index.js';\n\nimport {\n  DEFAULT_JUDGE_TIMEOUT_MS,\n  extractAssistantText,\n  getAssistantMessageErrorReason,\n  resolveGoalJudgeApiKey,\n  stripCodeFences,\n  truncateGoalText,\n} from './judge.js';\nimport type { GoalUiLocale } from './goal-locale.js';\nimport {\n  goalUiLocaleOrFallback,\n  judgeReasonText,\n  judgeResponseLanguageNote,\n  JUDGE_REASON_EN,\n  localizeJudgeReasonText,\n} from './goal-locale.js';\n\nconst GOAL_TRUNC = 4000;\nconst RESPONSE_SNIPPET = 4000;\nconst HISTORY_TRUNC = 24_000;\n\nconst DECOMPOSE_SYSTEM = (\n  'You break a user goal into an EXTREMELY detailed checklist of concrete, verifiable completion criteria. ' +\n  'Bias toward more items. Each item is one factual statement about finished work.\\n\\n' +\n  'Reply ONLY with a single JSON object on one line:\\n' +\n  '{\"items\":[{\"text\":\"...\"},...]}\\n' +\n  'Use at least 3 items when the goal warrants it.'\n);\n\nconst DECOMPOSE_USER = 'Goal:\\n{goal}\\n\\nReturn JSON only.';\n\nconst EVALUATE_CHECKLIST_SYSTEM = (\n  'You evaluate an autonomous agent\\'s progress on a goal with a numbered checklist.\\n' +\n  'For each pending item, decide if evidence in the agent snippet and/or history excerpt shows it is satisfied.\\n' +\n  'Flip pending→completed only with clear evidence; pending→impossible only if truly unachievable here.\\n' +\n  'Do not regress completed/impossible items — omit them from updates.\\n\\n' +\n  'Reply ONLY with one JSON object on one line:\\n' +\n  '{\"verdict\":\"continue|done|blocked|needs_input\",\"confidence\":0.0,\"reason\":\"...\",' +\n  '\"nextAction\":\"...\",\"missingEvidence\":[\"...\"],\"userQuestion\":\"...\",' +\n  '\"updates\":[{\"index\":1,\"status\":\"completed|impossible\",\"evidence\":\"short citation\"}],' +\n  '\"new_items\":[{\"text\":\"...\"}]}\\n' +\n  'Use verdict=done only when all checklist criteria are complete/impossible. ' +\n  'Use needs_input when the next step requires user input. Use blocked when external progress is impossible.'\n);\n\nfunction extractJsonObject(raw: string): Record<string, unknown> | null {\n  if (!raw?.trim()) return null;\n  const text = stripCodeFences(raw);\n  try {\n    const data = JSON.parse(text) as unknown;\n    return data && typeof data === 'object' && !Array.isArray(data) ? (data as Record<string, unknown>) : null;\n  } catch {\n    const start = text.indexOf('{');\n    const end = text.lastIndexOf('}');\n    if (start === -1 || end <= start) return null;\n    try {\n      const data = JSON.parse(text.slice(start, end + 1)) as unknown;\n      return data && typeof data === 'object' && !Array.isArray(data) ? (data as Record<string, unknown>) : null;\n    } catch {\n      return null;\n    }\n  }\n}\n\nexport type DecomposeChecklistResult = {\n  items: { text: string }[];\n  parseFailed: boolean;\n  errorReason?: string;\n};\n\nexport async function decomposeGoalChecklist(opts: {\n  goal: string;\n  judgeModelRef: string;\n  signal?: AbortSignal;\n  judgeTimeoutMs?: number;\n  uiLocale?: GoalUiLocale;\n}): Promise<DecomposeChecklistResult> {\n  const locale = goalUiLocaleOrFallback(opts.uiLocale);\n  const goal = opts.goal.trim();\n  if (!goal)\n    return { items: [], parseFailed: false, errorReason: judgeReasonText('empty_goal', locale) };\n\n  let model: ReturnType<typeof resolveModel>;\n  try {\n    model = resolveModel(opts.judgeModelRef);\n  } catch {\n    return {\n      items: [],\n      parseFailed: false,\n      errorReason: judgeReasonText('judge_model_not_configured', locale),\n    };\n  }\n\n  const timeoutMs =\n    typeof opts.judgeTimeoutMs === 'number' && Number.isFinite(opts.judgeTimeoutMs)\n      ? Math.max(5_000, Math.min(120_000, Math.floor(opts.judgeTimeoutMs)))\n      : DEFAULT_JUDGE_TIMEOUT_MS;\n  const controller = new AbortController();\n  const timer = setTimeout(() => controller.abort(), timeoutMs);\n  const merged = opts.signal ? AbortSignal.any([opts.signal, controller.signal]) : controller.signal;\n\n  try {\n    const user: UserMessage = {\n      role: 'user',\n      content:\n        `${DECOMPOSE_SYSTEM}\\n\\n${DECOMPOSE_USER.replace('{goal}', truncateGoalText(goal, GOAL_TRUNC))}` +\n        judgeResponseLanguageNote(locale),\n      timestamp: Date.now(),\n    };\n    const apiKey = await resolveGoalJudgeApiKey(model);\n    const result = await complete(\n      model,\n      { messages: [user] },\n      { apiKey, maxTokens: 2000, temperature: 0, signal: merged },\n    );\n    const errorReason = getAssistantMessageErrorReason(result);\n    if (errorReason) {\n      return {\n        items: [],\n        parseFailed: false,\n        errorReason: judgeReasonText('decompose_call_failed', locale),\n      };\n    }\n\n    const text = extractAssistantText(result.content);\n    const data = extractJsonObject(text);\n    if (!data)\n      return {\n        items: [],\n        parseFailed: true,\n        errorReason: judgeReasonText('decompose_reply_not_json', locale),\n      };\n    const rawItems = data.items ?? data.checklist;\n    if (!Array.isArray(rawItems))\n      return {\n        items: [],\n        parseFailed: true,\n        errorReason: judgeReasonText('missing_items_array', locale),\n      };\n    const items: { text: string }[] = [];\n    for (const row of rawItems) {\n      if (typeof row === 'string') {\n        const t = row.trim();\n        if (t) items.push({ text: t });\n      } else if (row && typeof row === 'object' && !Array.isArray(row)) {\n        const t = String((row as { text?: string }).text ?? '').trim();\n        if (t) items.push({ text: t });\n      }\n    }\n    if (!items.length)\n      return { items: [], parseFailed: true, errorReason: judgeReasonText('empty_checklist', locale) };\n    return { items, parseFailed: false };\n  } catch {\n    return {\n      items: [],\n      parseFailed: false,\n      errorReason: judgeReasonText('decompose_call_failed', locale),\n    };\n  } finally {\n    clearTimeout(timer);\n  }\n}\n\nexport type ChecklistJudgeUpdate = {\n  index: number;\n  status: 'completed' | 'impossible';\n  evidence?: string | null;\n};\n\nexport type ChecklistEvaluateParsed = {\n  verdict: 'continue' | 'done' | 'blocked' | 'needs_input';\n  confidence: number;\n  updates: ChecklistJudgeUpdate[];\n  newItems: { text: string }[];\n  reason: string;\n  nextAction?: string;\n  missingEvidence?: string[];\n  userQuestion?: string;\n};\n\nexport type EvaluateChecklistResult = {\n  parsed: ChecklistEvaluateParsed;\n  parseFailed: boolean;\n};\n\nexport async function evaluateGoalChecklistJudge(opts: {\n  goal: string;\n  numberedChecklist: string;\n  lastResponse: string;\n  historyExcerpt: string;\n  judgeModelRef: string;\n  signal?: AbortSignal;\n  judgeTimeoutMs?: number;\n  uiLocale?: GoalUiLocale;\n}): Promise<EvaluateChecklistResult> {\n  const locale = goalUiLocaleOrFallback(opts.uiLocale);\n  let model: ReturnType<typeof resolveModel>;\n  try {\n    model = resolveModel(opts.judgeModelRef);\n  } catch {\n    return {\n      parsed: {\n        verdict: 'continue',\n        confidence: 0,\n        updates: [],\n        newItems: [],\n        reason: judgeReasonText('judge_model_not_configured', locale),\n      },\n      parseFailed: false,\n    };\n  }\n\n  const userBody =\n    `Goal:\\n${truncateGoalText(opts.goal, 2000)}\\n\\n` +\n    `Current checklist (1-based indices):\\n${opts.numberedChecklist}\\n\\n` +\n    `Agent's most recent response (snippet):\\n${truncateGoalText(opts.lastResponse, RESPONSE_SNIPPET)}\\n\\n` +\n    `Recent conversation excerpt (JSON or text, may be truncated):\\n` +\n    `${truncateGoalText(opts.historyExcerpt, HISTORY_TRUNC) || '(none)'}`;\n\n  const timeoutMs =\n    typeof opts.judgeTimeoutMs === 'number' && Number.isFinite(opts.judgeTimeoutMs)\n      ? Math.max(5_000, Math.min(120_000, Math.floor(opts.judgeTimeoutMs)))\n      : DEFAULT_JUDGE_TIMEOUT_MS;\n  const controller = new AbortController();\n  const timer = setTimeout(() => controller.abort(), timeoutMs);\n  const merged = opts.signal ? AbortSignal.any([opts.signal, controller.signal]) : controller.signal;\n\n  try {\n    const user: UserMessage = {\n      role: 'user',\n      content: `${EVALUATE_CHECKLIST_SYSTEM}\\n\\n${userBody}` + judgeResponseLanguageNote(locale),\n      timestamp: Date.now(),\n    };\n    const apiKey = await resolveGoalJudgeApiKey(model);\n    const result = await complete(\n      model,\n      { messages: [user] },\n      { apiKey, maxTokens: 1500, temperature: 0, signal: merged },\n    );\n    const errorReason = getAssistantMessageErrorReason(result);\n    if (errorReason) {\n      return {\n        parsed: {\n          verdict: 'continue',\n          confidence: 0,\n          updates: [],\n          newItems: [],\n          reason: judgeReasonText('judge_call_failed', locale),\n        },\n        parseFailed: false,\n      };\n    }\n\n    const text = extractAssistantText(result.content);\n    const data = extractJsonObject(text);\n    if (!data) {\n      return {\n        parsed: {\n          verdict: 'continue',\n          confidence: 0,\n          updates: [],\n          newItems: [],\n          reason: judgeReasonText('judge_reply_not_json', locale),\n        },\n        parseFailed: true,\n      };\n    }\n    const updatesRaw = data.updates;\n    const newRaw = data.new_items ?? data.newItems;\n    const verdictRaw = String(data.verdict ?? '').trim().toLowerCase();\n    const verdict =\n      verdictRaw === 'done' || verdictRaw === 'blocked' || verdictRaw === 'needs_input'\n        ? verdictRaw\n        : 'continue';\n    const confidenceRaw = Number(data.confidence);\n    const confidence = Number.isFinite(confidenceRaw) ? Math.max(0, Math.min(1, confidenceRaw)) : 0;\n    const nextAction = typeof data.nextAction === 'string' ? data.nextAction.trim() : undefined;\n    const userQuestion = typeof data.userQuestion === 'string' ? data.userQuestion.trim() : undefined;\n    const missingEvidenceRaw = data.missingEvidence ?? data.missing_evidence;\n    const trimmed = typeof data.reason === 'string' ? data.reason.trim() : '';\n    const inner = trimmed || JUDGE_REASON_EN.no_reason_provided;\n    const reason = localizeJudgeReasonText(inner, locale);\n    const updates: ChecklistJudgeUpdate[] = [];\n    if (Array.isArray(updatesRaw)) {\n      for (const u of updatesRaw) {\n        if (!u || typeof u !== 'object' || Array.isArray(u)) continue;\n        const rec = u as Record<string, unknown>;\n        const idx1 = Number(rec.index);\n        if (!Number.isFinite(idx1)) continue;\n        const idx0 = Math.floor(idx1) - 1;\n        const st = String(rec.status ?? '').trim().toLowerCase();\n        if (st !== 'completed' && st !== 'impossible') continue;\n        const evidence = typeof rec.evidence === 'string' ? rec.evidence : null;\n        updates.push({ index: idx0, status: st, evidence });\n      }\n    }\n    const newItems: { text: string }[] = [];\n    if (Array.isArray(newRaw)) {\n      for (const row of newRaw) {\n        if (typeof row === 'string') {\n          const t = row.trim();\n          if (t) newItems.push({ text: t });\n        } else if (row && typeof row === 'object' && !Array.isArray(row)) {\n          const t = String((row as { text?: string }).text ?? '').trim();\n          if (t) newItems.push({ text: t });\n        }\n      }\n    }\n    const missingEvidence: string[] = [];\n    if (Array.isArray(missingEvidenceRaw)) {\n      for (const item of missingEvidenceRaw) {\n        if (typeof item === 'string' && item.trim()) missingEvidence.push(item.trim());\n      }\n    }\n    return {\n      parsed: {\n        verdict,\n        confidence,\n        updates,\n        newItems,\n        reason: reason || judgeReasonText('no_reason_provided', locale),\n        ...(nextAction ? { nextAction } : {}),\n        ...(missingEvidence.length ? { missingEvidence } : {}),\n        ...(userQuestion ? { userQuestion } : {}),\n      },\n      parseFailed: false,\n    };\n  } catch {\n    return {\n      parsed: {\n        verdict: 'continue',\n        confidence: 0,\n        updates: [],\n        newItems: [],\n        reason: judgeReasonText('judge_call_failed', locale),\n      },\n      parseFailed: false,\n    };\n  } finally {\n    clearTimeout(timer);\n  }\n}\n"],"mappings":";;;;;;gBAGwD;AAmBxD,MAAM,aAAa;AACnB,MAAM,mBAAmB;AACzB,MAAM,gBAAgB;AAEtB,MAAM,mBACJ;AAOF,MAAM,iBAAiB;AAEvB,MAAM,4BACJ;AAaF,SAAS,kBAAkB,KAA6C;AACtE,KAAI,CAAC,KAAK,MAAM,CAAE,QAAO;CACzB,MAAM,OAAO,gBAAgB,IAAI;AACjC,KAAI;EACF,MAAM,OAAO,KAAK,MAAM,KAAK;AAC7B,SAAO,QAAQ,OAAO,SAAS,YAAY,CAAC,MAAM,QAAQ,KAAK,GAAI,OAAmC;SAChG;EACN,MAAM,QAAQ,KAAK,QAAQ,IAAI;EAC/B,MAAM,MAAM,KAAK,YAAY,IAAI;AACjC,MAAI,UAAU,MAAM,OAAO,MAAO,QAAO;AACzC,MAAI;GACF,MAAM,OAAO,KAAK,MAAM,KAAK,MAAM,OAAO,MAAM,EAAE,CAAC;AACnD,UAAO,QAAQ,OAAO,SAAS,YAAY,CAAC,MAAM,QAAQ,KAAK,GAAI,OAAmC;UAChG;AACN,UAAO;;;;AAWb,eAAsB,uBAAuB,MAMP;CACpC,MAAM,SAAS,uBAAuB,KAAK,SAAS;CACpD,MAAM,OAAO,KAAK,KAAK,MAAM;AAC7B,KAAI,CAAC,KACH,QAAO;EAAE,OAAO,EAAE;EAAE,aAAa;EAAO,aAAa,gBAAgB,cAAc,OAAO;EAAE;CAE9F,IAAI;AACJ,KAAI;AACF,UAAQ,aAAa,KAAK,cAAc;SAClC;AACN,SAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,8BAA8B,OAAO;GACnE;;CAGH,MAAM,YACJ,OAAO,KAAK,mBAAmB,YAAY,OAAO,SAAS,KAAK,eAAe,GAC3E,KAAK,IAAI,KAAO,KAAK,IAAI,MAAS,KAAK,MAAM,KAAK,eAAe,CAAC,CAAC,GACnE;CACN,MAAM,aAAa,IAAI,iBAAiB;CACxC,MAAM,QAAQ,iBAAiB,WAAW,OAAO,EAAE,UAAU;CAC7D,MAAM,SAAS,KAAK,SAAS,YAAY,IAAI,CAAC,KAAK,QAAQ,WAAW,OAAO,CAAC,GAAG,WAAW;AAE5F,KAAI;EACF,MAAM,OAAoB;GACxB,MAAM;GACN,SACE,GAAG,iBAAiB,MAAM,eAAe,QAAQ,UAAU,iBAAiB,MAAM,WAAW,CAAC,KAC9F,0BAA0B,OAAO;GACnC,WAAW,KAAK,KAAK;GACtB;EACD,MAAM,SAAS,MAAM,uBAAuB,MAAM;EAClD,MAAM,SAAS,MAAM,SACnB,OACA,EAAE,UAAU,CAAC,KAAK,EAAE,EACpB;GAAE;GAAQ,WAAW;GAAM,aAAa;GAAG,QAAQ;GAAQ,CAC5D;AAED,MADoB,+BAA+B,OACpC,CACb,QAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,yBAAyB,OAAO;GAC9D;EAIH,MAAM,OAAO,kBADA,qBAAqB,OAAO,QACN,CAAC;AACpC,MAAI,CAAC,KACH,QAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,4BAA4B,OAAO;GACjE;EACH,MAAM,WAAW,KAAK,SAAS,KAAK;AACpC,MAAI,CAAC,MAAM,QAAQ,SAAS,CAC1B,QAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,uBAAuB,OAAO;GAC5D;EACH,MAAM,QAA4B,EAAE;AACpC,OAAK,MAAM,OAAO,SAChB,KAAI,OAAO,QAAQ,UAAU;GAC3B,MAAM,IAAI,IAAI,MAAM;AACpB,OAAI,EAAG,OAAM,KAAK,EAAE,MAAM,GAAG,CAAC;aACrB,OAAO,OAAO,QAAQ,YAAY,CAAC,MAAM,QAAQ,IAAI,EAAE;GAChE,MAAM,IAAI,OAAQ,IAA0B,QAAQ,GAAG,CAAC,MAAM;AAC9D,OAAI,EAAG,OAAM,KAAK,EAAE,MAAM,GAAG,CAAC;;AAGlC,MAAI,CAAC,MAAM,OACT,QAAO;GAAE,OAAO,EAAE;GAAE,aAAa;GAAM,aAAa,gBAAgB,mBAAmB,OAAO;GAAE;AAClG,SAAO;GAAE;GAAO,aAAa;GAAO;SAC9B;AACN,SAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,yBAAyB,OAAO;GAC9D;WACO;AACR,eAAa,MAAM;;;AA0BvB,eAAsB,2BAA2B,MASZ;CACnC,MAAM,SAAS,uBAAuB,KAAK,SAAS;CACpD,IAAI;AACJ,KAAI;AACF,UAAQ,aAAa,KAAK,cAAc;SAClC;AACN,SAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,8BAA8B,OAAO;IAC9D;GACD,aAAa;GACd;;CAGH,MAAM,WACJ,UAAU,iBAAiB,KAAK,MAAM,IAAK,CAAC,4CACH,KAAK,kBAAkB,+CACpB,iBAAiB,KAAK,cAAc,iBAAiB,CAAC,qEAE/F,iBAAiB,KAAK,gBAAgB,cAAc,IAAI;CAE7D,MAAM,YACJ,OAAO,KAAK,mBAAmB,YAAY,OAAO,SAAS,KAAK,eAAe,GAC3E,KAAK,IAAI,KAAO,KAAK,IAAI,MAAS,KAAK,MAAM,KAAK,eAAe,CAAC,CAAC,GACnE;CACN,MAAM,aAAa,IAAI,iBAAiB;CACxC,MAAM,QAAQ,iBAAiB,WAAW,OAAO,EAAE,UAAU;CAC7D,MAAM,SAAS,KAAK,SAAS,YAAY,IAAI,CAAC,KAAK,QAAQ,WAAW,OAAO,CAAC,GAAG,WAAW;AAE5F,KAAI;EACF,MAAM,OAAoB;GACxB,MAAM;GACN,SAAS,GAAG,0BAA0B,MAAM,aAAa,0BAA0B,OAAO;GAC1F,WAAW,KAAK,KAAK;GACtB;EACD,MAAM,SAAS,MAAM,uBAAuB,MAAM;EAClD,MAAM,SAAS,MAAM,SACnB,OACA,EAAE,UAAU,CAAC,KAAK,EAAE,EACpB;GAAE;GAAQ,WAAW;GAAM,aAAa;GAAG,QAAQ;GAAQ,CAC5D;AAED,MADoB,+BAA+B,OACpC,CACb,QAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,qBAAqB,OAAO;IACrD;GACD,aAAa;GACd;EAIH,MAAM,OAAO,kBADA,qBAAqB,OAAO,QACN,CAAC;AACpC,MAAI,CAAC,KACH,QAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,wBAAwB,OAAO;IACxD;GACD,aAAa;GACd;EAEH,MAAM,aAAa,KAAK;EACxB,MAAM,SAAS,KAAK,aAAa,KAAK;EACtC,MAAM,aAAa,OAAO,KAAK,WAAW,GAAG,CAAC,MAAM,CAAC,aAAa;EAClE,MAAM,UACJ,eAAe,UAAU,eAAe,aAAa,eAAe,gBAChE,aACA;EACN,MAAM,gBAAgB,OAAO,KAAK,WAAW;EAC7C,MAAM,aAAa,OAAO,SAAS,cAAc,GAAG,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,cAAc,CAAC,GAAG;EAC9F,MAAM,aAAa,OAAO,KAAK,eAAe,WAAW,KAAK,WAAW,MAAM,GAAG,KAAA;EAClF,MAAM,eAAe,OAAO,KAAK,iBAAiB,WAAW,KAAK,aAAa,MAAM,GAAG,KAAA;EACxF,MAAM,qBAAqB,KAAK,mBAAmB,KAAK;EAGxD,MAAM,SAAS,yBAFC,OAAO,KAAK,WAAW,WAAW,KAAK,OAAO,MAAM,GAAG,OAC9C,gBAAgB,oBACK,OAAO;EACrD,MAAM,UAAkC,EAAE;AAC1C,MAAI,MAAM,QAAQ,WAAW,CAC3B,MAAK,MAAM,KAAK,YAAY;AAC1B,OAAI,CAAC,KAAK,OAAO,MAAM,YAAY,MAAM,QAAQ,EAAE,CAAE;GACrD,MAAM,MAAM;GACZ,MAAM,OAAO,OAAO,IAAI,MAAM;AAC9B,OAAI,CAAC,OAAO,SAAS,KAAK,CAAE;GAC5B,MAAM,OAAO,KAAK,MAAM,KAAK,GAAG;GAChC,MAAM,KAAK,OAAO,IAAI,UAAU,GAAG,CAAC,MAAM,CAAC,aAAa;AACxD,OAAI,OAAO,eAAe,OAAO,aAAc;GAC/C,MAAM,WAAW,OAAO,IAAI,aAAa,WAAW,IAAI,WAAW;AACnE,WAAQ,KAAK;IAAE,OAAO;IAAM,QAAQ;IAAI;IAAU,CAAC;;EAGvD,MAAM,WAA+B,EAAE;AACvC,MAAI,MAAM,QAAQ,OAAO;QAClB,MAAM,OAAO,OAChB,KAAI,OAAO,QAAQ,UAAU;IAC3B,MAAM,IAAI,IAAI,MAAM;AACpB,QAAI,EAAG,UAAS,KAAK,EAAE,MAAM,GAAG,CAAC;cACxB,OAAO,OAAO,QAAQ,YAAY,CAAC,MAAM,QAAQ,IAAI,EAAE;IAChE,MAAM,IAAI,OAAQ,IAA0B,QAAQ,GAAG,CAAC,MAAM;AAC9D,QAAI,EAAG,UAAS,KAAK,EAAE,MAAM,GAAG,CAAC;;;EAIvC,MAAM,kBAA4B,EAAE;AACpC,MAAI,MAAM,QAAQ,mBAAmB;QAC9B,MAAM,QAAQ,mBACjB,KAAI,OAAO,SAAS,YAAY,KAAK,MAAM,CAAE,iBAAgB,KAAK,KAAK,MAAM,CAAC;;AAGlF,SAAO;GACL,QAAQ;IACN;IACA;IACA;IACA;IACA,QAAQ,UAAU,gBAAgB,sBAAsB,OAAO;IAC/D,GAAI,aAAa,EAAE,YAAY,GAAG,EAAE;IACpC,GAAI,gBAAgB,SAAS,EAAE,iBAAiB,GAAG,EAAE;IACrD,GAAI,eAAe,EAAE,cAAc,GAAG,EAAE;IACzC;GACD,aAAa;GACd;SACK;AACN,SAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,qBAAqB,OAAO;IACrD;GACD,aAAa;GACd;WACO;AACR,eAAa,MAAM"}
+{"version":3,"file":"checklist-judge.js","names":[],"sources":["../../../../src/agent/goals/checklist-judge.ts"],"sourcesContent":["import type { UserMessage } from '@earendil-works/pi-ai';\nimport { complete } from '@earendil-works/pi-ai';\n\nimport { resolveModel } from '../../providers/index.js';\n\nimport {\n  DEFAULT_JUDGE_TIMEOUT_MS,\n  extractAssistantText,\n  getAssistantMessageErrorReason,\n  resolveGoalJudgeApiKey,\n  stripCodeFences,\n  truncateGoalText,\n} from './judge.js';\nimport type { GoalUiLocale } from './goal-locale.js';\nimport {\n  goalUiLocaleOrFallback,\n  judgeReasonText,\n  judgeResponseLanguageNote,\n  JUDGE_REASON_EN,\n  localizeJudgeReasonText,\n} from './goal-locale.js';\n\nconst GOAL_TRUNC = 4000;\nconst RESPONSE_SNIPPET = 4000;\nconst HISTORY_TRUNC = 24_000;\n\nconst DECOMPOSE_SYSTEM = (\n  'You break a user goal into an EXTREMELY detailed checklist of concrete, verifiable completion criteria. ' +\n  'Bias toward more items. Each item is one factual statement about finished work.\\n\\n' +\n  'If existing acceptance criteria are provided, return only missing supplemental criteria. ' +\n  'Do not repeat, rewrite, or replace existing criteria.\\n\\n' +\n  'Reply ONLY with a single JSON object on one line:\\n' +\n  '{\"items\":[{\"text\":\"...\"},...]}\\n' +\n  'Use at least 3 items when the goal warrants it.'\n);\n\nconst DECOMPOSE_USER = 'Goal:\\n{goal}\\n\\nReturn JSON only.';\n\nconst EVALUATE_CHECKLIST_SYSTEM = (\n  'You evaluate an autonomous agent\\'s progress on a goal with a numbered checklist.\\n' +\n  'For each pending item, decide if evidence in the agent snippet and/or history excerpt shows it is satisfied.\\n' +\n  'Flip pending→completed only with clear evidence; pending→impossible only if truly unachievable here.\\n' +\n  'Do not regress completed/impossible items — omit them from updates.\\n\\n' +\n  'Reply ONLY with one JSON object on one line:\\n' +\n  '{\"verdict\":\"continue|done|blocked|needs_input\",\"confidence\":0.0,\"reason\":\"...\",' +\n  '\"nextAction\":\"...\",\"missingEvidence\":[\"...\"],\"userQuestion\":\"...\",' +\n  '\"updates\":[{\"index\":1,\"status\":\"completed|impossible\",\"evidence\":\"short citation\"}],' +\n  '\"new_items\":[{\"text\":\"...\"}]}\\n' +\n  'Use verdict=done only when all checklist criteria are complete/impossible. ' +\n  'Use needs_input when the next step requires user input. Use blocked when external progress is impossible.'\n);\n\nfunction extractJsonObject(raw: string): Record<string, unknown> | null {\n  if (!raw?.trim()) return null;\n  const text = stripCodeFences(raw);\n  try {\n    const data = JSON.parse(text) as unknown;\n    return data && typeof data === 'object' && !Array.isArray(data) ? (data as Record<string, unknown>) : null;\n  } catch {\n    const start = text.indexOf('{');\n    const end = text.lastIndexOf('}');\n    if (start === -1 || end <= start) return null;\n    try {\n      const data = JSON.parse(text.slice(start, end + 1)) as unknown;\n      return data && typeof data === 'object' && !Array.isArray(data) ? (data as Record<string, unknown>) : null;\n    } catch {\n      return null;\n    }\n  }\n}\n\nexport type DecomposeChecklistResult = {\n  items: { text: string }[];\n  parseFailed: boolean;\n  errorReason?: string;\n};\n\nexport async function decomposeGoalChecklist(opts: {\n  goal: string;\n  existingChecklist?: string;\n  judgeModelRef: string;\n  signal?: AbortSignal;\n  judgeTimeoutMs?: number;\n  uiLocale?: GoalUiLocale;\n}): Promise<DecomposeChecklistResult> {\n  const locale = goalUiLocaleOrFallback(opts.uiLocale);\n  const goal = opts.goal.trim();\n  if (!goal)\n    return { items: [], parseFailed: false, errorReason: judgeReasonText('empty_goal', locale) };\n\n  let model: ReturnType<typeof resolveModel>;\n  try {\n    model = resolveModel(opts.judgeModelRef);\n  } catch {\n    return {\n      items: [],\n      parseFailed: false,\n      errorReason: judgeReasonText('judge_model_not_configured', locale),\n    };\n  }\n\n  const timeoutMs =\n    typeof opts.judgeTimeoutMs === 'number' && Number.isFinite(opts.judgeTimeoutMs)\n      ? Math.max(5_000, Math.min(120_000, Math.floor(opts.judgeTimeoutMs)))\n      : DEFAULT_JUDGE_TIMEOUT_MS;\n  const controller = new AbortController();\n  const timer = setTimeout(() => controller.abort(), timeoutMs);\n  const merged = opts.signal ? AbortSignal.any([opts.signal, controller.signal]) : controller.signal;\n\n  try {\n    const existingChecklist = opts.existingChecklist?.trim();\n    const decomposeUser = DECOMPOSE_USER.replace('{goal}', truncateGoalText(goal, GOAL_TRUNC)) +\n      (existingChecklist\n        ? `\\n\\nExisting acceptance criteria:\\n${truncateGoalText(existingChecklist, GOAL_TRUNC)}\\n\\nReturn only missing supplemental criteria as JSON.`\n        : '');\n    const user: UserMessage = {\n      role: 'user',\n      content:\n        `${DECOMPOSE_SYSTEM}\\n\\n${decomposeUser}` +\n        judgeResponseLanguageNote(locale),\n      timestamp: Date.now(),\n    };\n    const apiKey = await resolveGoalJudgeApiKey(model);\n    const result = await complete(\n      model,\n      { messages: [user] },\n      { apiKey, maxTokens: 2000, temperature: 0, signal: merged },\n    );\n    const errorReason = getAssistantMessageErrorReason(result);\n    if (errorReason) {\n      return {\n        items: [],\n        parseFailed: false,\n        errorReason: judgeReasonText('decompose_call_failed', locale),\n      };\n    }\n\n    const text = extractAssistantText(result.content);\n    const data = extractJsonObject(text);\n    if (!data)\n      return {\n        items: [],\n        parseFailed: true,\n        errorReason: judgeReasonText('decompose_reply_not_json', locale),\n      };\n    const rawItems = data.items ?? data.checklist;\n    if (!Array.isArray(rawItems))\n      return {\n        items: [],\n        parseFailed: true,\n        errorReason: judgeReasonText('missing_items_array', locale),\n      };\n    const items: { text: string }[] = [];\n    for (const row of rawItems) {\n      if (typeof row === 'string') {\n        const t = row.trim();\n        if (t) items.push({ text: t });\n      } else if (row && typeof row === 'object' && !Array.isArray(row)) {\n        const t = String((row as { text?: string }).text ?? '').trim();\n        if (t) items.push({ text: t });\n      }\n    }\n    if (!items.length)\n      return { items: [], parseFailed: true, errorReason: judgeReasonText('empty_checklist', locale) };\n    return { items, parseFailed: false };\n  } catch {\n    return {\n      items: [],\n      parseFailed: false,\n      errorReason: judgeReasonText('decompose_call_failed', locale),\n    };\n  } finally {\n    clearTimeout(timer);\n  }\n}\n\nexport type ChecklistJudgeUpdate = {\n  index: number;\n  status: 'completed' | 'impossible';\n  evidence?: string | null;\n};\n\nexport type ChecklistEvaluateParsed = {\n  verdict: 'continue' | 'done' | 'blocked' | 'needs_input';\n  confidence: number;\n  updates: ChecklistJudgeUpdate[];\n  newItems: { text: string }[];\n  reason: string;\n  nextAction?: string;\n  missingEvidence?: string[];\n  userQuestion?: string;\n};\n\nexport type EvaluateChecklistResult = {\n  parsed: ChecklistEvaluateParsed;\n  parseFailed: boolean;\n};\n\nexport async function evaluateGoalChecklistJudge(opts: {\n  goal: string;\n  numberedChecklist: string;\n  lastResponse: string;\n  historyExcerpt: string;\n  judgeModelRef: string;\n  signal?: AbortSignal;\n  judgeTimeoutMs?: number;\n  uiLocale?: GoalUiLocale;\n}): Promise<EvaluateChecklistResult> {\n  const locale = goalUiLocaleOrFallback(opts.uiLocale);\n  let model: ReturnType<typeof resolveModel>;\n  try {\n    model = resolveModel(opts.judgeModelRef);\n  } catch {\n    return {\n      parsed: {\n        verdict: 'continue',\n        confidence: 0,\n        updates: [],\n        newItems: [],\n        reason: judgeReasonText('judge_model_not_configured', locale),\n      },\n      parseFailed: false,\n    };\n  }\n\n  const userBody =\n    `Goal:\\n${truncateGoalText(opts.goal, 2000)}\\n\\n` +\n    `Current checklist (1-based indices):\\n${opts.numberedChecklist}\\n\\n` +\n    `Agent's most recent response (snippet):\\n${truncateGoalText(opts.lastResponse, RESPONSE_SNIPPET)}\\n\\n` +\n    `Recent conversation excerpt (JSON or text, may be truncated):\\n` +\n    `${truncateGoalText(opts.historyExcerpt, HISTORY_TRUNC) || '(none)'}`;\n\n  const timeoutMs =\n    typeof opts.judgeTimeoutMs === 'number' && Number.isFinite(opts.judgeTimeoutMs)\n      ? Math.max(5_000, Math.min(120_000, Math.floor(opts.judgeTimeoutMs)))\n      : DEFAULT_JUDGE_TIMEOUT_MS;\n  const controller = new AbortController();\n  const timer = setTimeout(() => controller.abort(), timeoutMs);\n  const merged = opts.signal ? AbortSignal.any([opts.signal, controller.signal]) : controller.signal;\n\n  try {\n    const user: UserMessage = {\n      role: 'user',\n      content: `${EVALUATE_CHECKLIST_SYSTEM}\\n\\n${userBody}` + judgeResponseLanguageNote(locale),\n      timestamp: Date.now(),\n    };\n    const apiKey = await resolveGoalJudgeApiKey(model);\n    const result = await complete(\n      model,\n      { messages: [user] },\n      { apiKey, maxTokens: 1500, temperature: 0, signal: merged },\n    );\n    const errorReason = getAssistantMessageErrorReason(result);\n    if (errorReason) {\n      return {\n        parsed: {\n          verdict: 'continue',\n          confidence: 0,\n          updates: [],\n          newItems: [],\n          reason: judgeReasonText('judge_call_failed', locale),\n        },\n        parseFailed: false,\n      };\n    }\n\n    const text = extractAssistantText(result.content);\n    const data = extractJsonObject(text);\n    if (!data) {\n      return {\n        parsed: {\n          verdict: 'continue',\n          confidence: 0,\n          updates: [],\n          newItems: [],\n          reason: judgeReasonText('judge_reply_not_json', locale),\n        },\n        parseFailed: true,\n      };\n    }\n    const updatesRaw = data.updates;\n    const newRaw = data.new_items ?? data.newItems;\n    const verdictRaw = String(data.verdict ?? '').trim().toLowerCase();\n    const verdict =\n      verdictRaw === 'done' || verdictRaw === 'blocked' || verdictRaw === 'needs_input'\n        ? verdictRaw\n        : 'continue';\n    const confidenceRaw = Number(data.confidence);\n    const confidence = Number.isFinite(confidenceRaw) ? Math.max(0, Math.min(1, confidenceRaw)) : 0;\n    const nextAction = typeof data.nextAction === 'string' ? data.nextAction.trim() : undefined;\n    const userQuestion = typeof data.userQuestion === 'string' ? data.userQuestion.trim() : undefined;\n    const missingEvidenceRaw = data.missingEvidence ?? data.missing_evidence;\n    const trimmed = typeof data.reason === 'string' ? data.reason.trim() : '';\n    const inner = trimmed || JUDGE_REASON_EN.no_reason_provided;\n    const reason = localizeJudgeReasonText(inner, locale);\n    const updates: ChecklistJudgeUpdate[] = [];\n    if (Array.isArray(updatesRaw)) {\n      for (const u of updatesRaw) {\n        if (!u || typeof u !== 'object' || Array.isArray(u)) continue;\n        const rec = u as Record<string, unknown>;\n        const idx1 = Number(rec.index);\n        if (!Number.isFinite(idx1)) continue;\n        const idx0 = Math.floor(idx1) - 1;\n        const st = String(rec.status ?? '').trim().toLowerCase();\n        if (st !== 'completed' && st !== 'impossible') continue;\n        const evidence = typeof rec.evidence === 'string' ? rec.evidence : null;\n        updates.push({ index: idx0, status: st, evidence });\n      }\n    }\n    const newItems: { text: string }[] = [];\n    if (Array.isArray(newRaw)) {\n      for (const row of newRaw) {\n        if (typeof row === 'string') {\n          const t = row.trim();\n          if (t) newItems.push({ text: t });\n        } else if (row && typeof row === 'object' && !Array.isArray(row)) {\n          const t = String((row as { text?: string }).text ?? '').trim();\n          if (t) newItems.push({ text: t });\n        }\n      }\n    }\n    const missingEvidence: string[] = [];\n    if (Array.isArray(missingEvidenceRaw)) {\n      for (const item of missingEvidenceRaw) {\n        if (typeof item === 'string' && item.trim()) missingEvidence.push(item.trim());\n      }\n    }\n    return {\n      parsed: {\n        verdict,\n        confidence,\n        updates,\n        newItems,\n        reason: reason || judgeReasonText('no_reason_provided', locale),\n        ...(nextAction ? { nextAction } : {}),\n        ...(missingEvidence.length ? { missingEvidence } : {}),\n        ...(userQuestion ? { userQuestion } : {}),\n      },\n      parseFailed: false,\n    };\n  } catch {\n    return {\n      parsed: {\n        verdict: 'continue',\n        confidence: 0,\n        updates: [],\n        newItems: [],\n        reason: judgeReasonText('judge_call_failed', locale),\n      },\n      parseFailed: false,\n    };\n  } finally {\n    clearTimeout(timer);\n  }\n}\n"],"mappings":";;;;;;gBAGwD;AAmBxD,MAAM,aAAa;AACnB,MAAM,mBAAmB;AACzB,MAAM,gBAAgB;AAEtB,MAAM,mBACJ;AASF,MAAM,iBAAiB;AAEvB,MAAM,4BACJ;AAaF,SAAS,kBAAkB,KAA6C;AACtE,KAAI,CAAC,KAAK,MAAM,CAAE,QAAO;CACzB,MAAM,OAAO,gBAAgB,IAAI;AACjC,KAAI;EACF,MAAM,OAAO,KAAK,MAAM,KAAK;AAC7B,SAAO,QAAQ,OAAO,SAAS,YAAY,CAAC,MAAM,QAAQ,KAAK,GAAI,OAAmC;SAChG;EACN,MAAM,QAAQ,KAAK,QAAQ,IAAI;EAC/B,MAAM,MAAM,KAAK,YAAY,IAAI;AACjC,MAAI,UAAU,MAAM,OAAO,MAAO,QAAO;AACzC,MAAI;GACF,MAAM,OAAO,KAAK,MAAM,KAAK,MAAM,OAAO,MAAM,EAAE,CAAC;AACnD,UAAO,QAAQ,OAAO,SAAS,YAAY,CAAC,MAAM,QAAQ,KAAK,GAAI,OAAmC;UAChG;AACN,UAAO;;;;AAWb,eAAsB,uBAAuB,MAOP;CACpC,MAAM,SAAS,uBAAuB,KAAK,SAAS;CACpD,MAAM,OAAO,KAAK,KAAK,MAAM;AAC7B,KAAI,CAAC,KACH,QAAO;EAAE,OAAO,EAAE;EAAE,aAAa;EAAO,aAAa,gBAAgB,cAAc,OAAO;EAAE;CAE9F,IAAI;AACJ,KAAI;AACF,UAAQ,aAAa,KAAK,cAAc;SAClC;AACN,SAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,8BAA8B,OAAO;GACnE;;CAGH,MAAM,YACJ,OAAO,KAAK,mBAAmB,YAAY,OAAO,SAAS,KAAK,eAAe,GAC3E,KAAK,IAAI,KAAO,KAAK,IAAI,MAAS,KAAK,MAAM,KAAK,eAAe,CAAC,CAAC,GACnE;CACN,MAAM,aAAa,IAAI,iBAAiB;CACxC,MAAM,QAAQ,iBAAiB,WAAW,OAAO,EAAE,UAAU;CAC7D,MAAM,SAAS,KAAK,SAAS,YAAY,IAAI,CAAC,KAAK,QAAQ,WAAW,OAAO,CAAC,GAAG,WAAW;AAE5F,KAAI;EACF,MAAM,oBAAoB,KAAK,mBAAmB,MAAM;EAKxD,MAAM,OAAoB;GACxB,MAAM;GACN,SACE,GAAG,iBAAiB,MAPF,eAAe,QAAQ,UAAU,iBAAiB,MAAM,WAAW,CAAC,IACvF,oBACG,sCAAsC,iBAAiB,mBAAmB,WAAW,CAAC,0DACtF,QAKF,0BAA0B,OAAO;GACnC,WAAW,KAAK,KAAK;GACtB;EACD,MAAM,SAAS,MAAM,uBAAuB,MAAM;EAClD,MAAM,SAAS,MAAM,SACnB,OACA,EAAE,UAAU,CAAC,KAAK,EAAE,EACpB;GAAE;GAAQ,WAAW;GAAM,aAAa;GAAG,QAAQ;GAAQ,CAC5D;AAED,MADoB,+BAA+B,OACpC,CACb,QAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,yBAAyB,OAAO;GAC9D;EAIH,MAAM,OAAO,kBADA,qBAAqB,OAAO,QACN,CAAC;AACpC,MAAI,CAAC,KACH,QAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,4BAA4B,OAAO;GACjE;EACH,MAAM,WAAW,KAAK,SAAS,KAAK;AACpC,MAAI,CAAC,MAAM,QAAQ,SAAS,CAC1B,QAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,uBAAuB,OAAO;GAC5D;EACH,MAAM,QAA4B,EAAE;AACpC,OAAK,MAAM,OAAO,SAChB,KAAI,OAAO,QAAQ,UAAU;GAC3B,MAAM,IAAI,IAAI,MAAM;AACpB,OAAI,EAAG,OAAM,KAAK,EAAE,MAAM,GAAG,CAAC;aACrB,OAAO,OAAO,QAAQ,YAAY,CAAC,MAAM,QAAQ,IAAI,EAAE;GAChE,MAAM,IAAI,OAAQ,IAA0B,QAAQ,GAAG,CAAC,MAAM;AAC9D,OAAI,EAAG,OAAM,KAAK,EAAE,MAAM,GAAG,CAAC;;AAGlC,MAAI,CAAC,MAAM,OACT,QAAO;GAAE,OAAO,EAAE;GAAE,aAAa;GAAM,aAAa,gBAAgB,mBAAmB,OAAO;GAAE;AAClG,SAAO;GAAE;GAAO,aAAa;GAAO;SAC9B;AACN,SAAO;GACL,OAAO,EAAE;GACT,aAAa;GACb,aAAa,gBAAgB,yBAAyB,OAAO;GAC9D;WACO;AACR,eAAa,MAAM;;;AA0BvB,eAAsB,2BAA2B,MASZ;CACnC,MAAM,SAAS,uBAAuB,KAAK,SAAS;CACpD,IAAI;AACJ,KAAI;AACF,UAAQ,aAAa,KAAK,cAAc;SAClC;AACN,SAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,8BAA8B,OAAO;IAC9D;GACD,aAAa;GACd;;CAGH,MAAM,WACJ,UAAU,iBAAiB,KAAK,MAAM,IAAK,CAAC,4CACH,KAAK,kBAAkB,+CACpB,iBAAiB,KAAK,cAAc,iBAAiB,CAAC,qEAE/F,iBAAiB,KAAK,gBAAgB,cAAc,IAAI;CAE7D,MAAM,YACJ,OAAO,KAAK,mBAAmB,YAAY,OAAO,SAAS,KAAK,eAAe,GAC3E,KAAK,IAAI,KAAO,KAAK,IAAI,MAAS,KAAK,MAAM,KAAK,eAAe,CAAC,CAAC,GACnE;CACN,MAAM,aAAa,IAAI,iBAAiB;CACxC,MAAM,QAAQ,iBAAiB,WAAW,OAAO,EAAE,UAAU;CAC7D,MAAM,SAAS,KAAK,SAAS,YAAY,IAAI,CAAC,KAAK,QAAQ,WAAW,OAAO,CAAC,GAAG,WAAW;AAE5F,KAAI;EACF,MAAM,OAAoB;GACxB,MAAM;GACN,SAAS,GAAG,0BAA0B,MAAM,aAAa,0BAA0B,OAAO;GAC1F,WAAW,KAAK,KAAK;GACtB;EACD,MAAM,SAAS,MAAM,uBAAuB,MAAM;EAClD,MAAM,SAAS,MAAM,SACnB,OACA,EAAE,UAAU,CAAC,KAAK,EAAE,EACpB;GAAE;GAAQ,WAAW;GAAM,aAAa;GAAG,QAAQ;GAAQ,CAC5D;AAED,MADoB,+BAA+B,OACpC,CACb,QAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,qBAAqB,OAAO;IACrD;GACD,aAAa;GACd;EAIH,MAAM,OAAO,kBADA,qBAAqB,OAAO,QACN,CAAC;AACpC,MAAI,CAAC,KACH,QAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,wBAAwB,OAAO;IACxD;GACD,aAAa;GACd;EAEH,MAAM,aAAa,KAAK;EACxB,MAAM,SAAS,KAAK,aAAa,KAAK;EACtC,MAAM,aAAa,OAAO,KAAK,WAAW,GAAG,CAAC,MAAM,CAAC,aAAa;EAClE,MAAM,UACJ,eAAe,UAAU,eAAe,aAAa,eAAe,gBAChE,aACA;EACN,MAAM,gBAAgB,OAAO,KAAK,WAAW;EAC7C,MAAM,aAAa,OAAO,SAAS,cAAc,GAAG,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,cAAc,CAAC,GAAG;EAC9F,MAAM,aAAa,OAAO,KAAK,eAAe,WAAW,KAAK,WAAW,MAAM,GAAG,KAAA;EAClF,MAAM,eAAe,OAAO,KAAK,iBAAiB,WAAW,KAAK,aAAa,MAAM,GAAG,KAAA;EACxF,MAAM,qBAAqB,KAAK,mBAAmB,KAAK;EAGxD,MAAM,SAAS,yBAFC,OAAO,KAAK,WAAW,WAAW,KAAK,OAAO,MAAM,GAAG,OAC9C,gBAAgB,oBACK,OAAO;EACrD,MAAM,UAAkC,EAAE;AAC1C,MAAI,MAAM,QAAQ,WAAW,CAC3B,MAAK,MAAM,KAAK,YAAY;AAC1B,OAAI,CAAC,KAAK,OAAO,MAAM,YAAY,MAAM,QAAQ,EAAE,CAAE;GACrD,MAAM,MAAM;GACZ,MAAM,OAAO,OAAO,IAAI,MAAM;AAC9B,OAAI,CAAC,OAAO,SAAS,KAAK,CAAE;GAC5B,MAAM,OAAO,KAAK,MAAM,KAAK,GAAG;GAChC,MAAM,KAAK,OAAO,IAAI,UAAU,GAAG,CAAC,MAAM,CAAC,aAAa;AACxD,OAAI,OAAO,eAAe,OAAO,aAAc;GAC/C,MAAM,WAAW,OAAO,IAAI,aAAa,WAAW,IAAI,WAAW;AACnE,WAAQ,KAAK;IAAE,OAAO;IAAM,QAAQ;IAAI;IAAU,CAAC;;EAGvD,MAAM,WAA+B,EAAE;AACvC,MAAI,MAAM,QAAQ,OAAO;QAClB,MAAM,OAAO,OAChB,KAAI,OAAO,QAAQ,UAAU;IAC3B,MAAM,IAAI,IAAI,MAAM;AACpB,QAAI,EAAG,UAAS,KAAK,EAAE,MAAM,GAAG,CAAC;cACxB,OAAO,OAAO,QAAQ,YAAY,CAAC,MAAM,QAAQ,IAAI,EAAE;IAChE,MAAM,IAAI,OAAQ,IAA0B,QAAQ,GAAG,CAAC,MAAM;AAC9D,QAAI,EAAG,UAAS,KAAK,EAAE,MAAM,GAAG,CAAC;;;EAIvC,MAAM,kBAA4B,EAAE;AACpC,MAAI,MAAM,QAAQ,mBAAmB;QAC9B,MAAM,QAAQ,mBACjB,KAAI,OAAO,SAAS,YAAY,KAAK,MAAM,CAAE,iBAAgB,KAAK,KAAK,MAAM,CAAC;;AAGlF,SAAO;GACL,QAAQ;IACN;IACA;IACA;IACA;IACA,QAAQ,UAAU,gBAAgB,sBAAsB,OAAO;IAC/D,GAAI,aAAa,EAAE,YAAY,GAAG,EAAE;IACpC,GAAI,gBAAgB,SAAS,EAAE,iBAAiB,GAAG,EAAE;IACrD,GAAI,eAAe,EAAE,cAAc,GAAG,EAAE;IACzC;GACD,aAAa;GACd;SACK;AACN,SAAO;GACL,QAAQ;IACN,SAAS;IACT,YAAY;IACZ,SAAS,EAAE;IACX,UAAU,EAAE;IACZ,QAAQ,gBAAgB,qBAAqB,OAAO;IACrD;GACD,aAAa;GACd;WACO;AACR,eAAa,MAAM"}

package/dist/src/agent/goals/evaluate-turn.d.ts

@@ -5,7 +5,7 @@ import { type PersistentGoalState } from './state.js';
 export declare const CONTINUATION_PROMPT_TEMPLATE: string;
 /** Hermes `CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE` (English; parity / tests). */
 export declare const CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE: string;
-export type GoalsEvaluateConfigSlice = Pick<GoalsConfig, 'checklistMode' | 'maxConsecutiveParseFailures' | 'judgeTimeoutMs' | 'checklistHistoryChars'>;
+export type GoalsEvaluateConfigSlice = Pick<GoalsConfig, 'checklistMode' | 'checklistDecomposePolicy' | 'maxConsecutiveParseFailures' | 'judgeTimeoutMs' | 'checklistHistoryChars'>;
 export type GoalPostTurnDecision = {
     newState: PersistentGoalState | null;
     shouldContinue: boolean;

package/dist/src/agent/goals/evaluate-turn.js

@@ -71,6 +71,7 @@ async function evaluateAfterTurnHermesLike(state, lastResponse, judgeModelRef, s
 	if (checklistMode && !next.decomposed) {
 		const dec = await decomposeGoalChecklist({
 			goal: next.goal,
+			existingChecklist: next.checklist?.length ? renderChecklistNumbered(next.checklist) : void 0,
 			judgeModelRef,
 			signal,
 			judgeTimeoutMs,

package/dist/src/agent/goals/evaluate-turn.js.map

@@ -1 +1 @@
-{"version":3,"file":"evaluate-turn.js","names":[],"sources":["../../../../src/agent/goals/evaluate-turn.ts"],"sourcesContent":["import type { GoalsConfig } from '../../config/schema.js';\n\nimport { decomposeGoalChecklist, evaluateGoalChecklistJudge } from './checklist-judge.js';\nimport { allChecklistTerminal, checklistCounts } from './checklist-types.js';\nimport {\n  buildContinuationPromptFromState,\n  checklistProgressSuffix,\n  CONTINUATION_PROMPT_TEMPLATE_EN,\n  CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE_EN,\n  goalEvaluateUserCopy,\n  resolveGoalUiLocale,\n  type GoalUiLocale,\n} from './goal-locale.js';\nimport { judgeGoalHermesStyle } from './judge.js';\nimport {\n  applyJudgeChecklistUpdates,\n  mergeDecomposedChecklistItems,\n  renderChecklistNumbered,\n  type PersistentGoalState,\n} from './state.js';\n\n/** Hermes `CONTINUATION_PROMPT_TEMPLATE` (English; parity / tests). */\nexport const CONTINUATION_PROMPT_TEMPLATE = CONTINUATION_PROMPT_TEMPLATE_EN;\n\n/** Hermes `CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE` (English; parity / tests). */\nexport const CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE = CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE_EN;\n\nexport type GoalsEvaluateConfigSlice = Pick<\n  GoalsConfig,\n  'checklistMode' | 'maxConsecutiveParseFailures' | 'judgeTimeoutMs' | 'checklistHistoryChars'\n>;\n\nexport type GoalPostTurnDecision = {\n  newState: PersistentGoalState | null;\n  shouldContinue: boolean;\n  continuationPrompt: string | null;\n  verdict: 'done' | 'continue' | 'blocked' | 'needs_input' | 'skipped' | 'inactive' | 'decompose';\n  reason: string;\n  message: string;\n  confidence?: number;\n  missingEvidence?: string[];\n  userQuestion?: string;\n  completedChecklistItemIndexes?: number[];\n};\n\nfunction cloneState(s: PersistentGoalState): PersistentGoalState {\n  return {\n    ...s,\n    checklist: s.checklist?.map((it) => ({ ...it })),\n  };\n}\n\n/**\n * Hermes `GoalManager.evaluate_after_turn` equivalent (minus SessionDB I/O).\n */\nexport async function evaluateAfterTurnHermesLike(\n  state: PersistentGoalState,\n  lastResponse: string,\n  judgeModelRef: string,\n  signal?: AbortSignal,\n  opts?: { goalsSlice?: Partial<GoalsEvaluateConfigSlice>; historyExcerpt?: string; uiLocale?: GoalUiLocale },\n): Promise<GoalPostTurnDecision> {\n  const locale = opts?.uiLocale ?? resolveGoalUiLocale(state);\n  const copy = goalEvaluateUserCopy(locale);\n\n  if (state.status !== 'active') {\n    return {\n      newState: state,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'inactive',\n      reason: 'no active goal',\n      message: '',\n    };\n  }\n\n  const slice = opts?.goalsSlice;\n  const checklistMode = slice?.checklistMode !== false;\n  const maxParseFailures = slice?.maxConsecutiveParseFailures ?? 3;\n  const judgeTimeoutMs = slice?.judgeTimeoutMs;\n\n  const next = cloneState(state);\n  next.turnsUsed += 1;\n  next.lastTurnAt = Date.now();\n\n  const bumpParse = () => {\n    next.consecutiveParseFailures = (next.consecutiveParseFailures ?? 0) + 1;\n  };\n  const resetParse = () => {\n    next.consecutiveParseFailures = 0;\n  };\n\n  const pauseIfParseStorm = (reason: string): GoalPostTurnDecision | null => {\n    const n = next.consecutiveParseFailures ?? 0;\n    if (n < maxParseFailures) return null;\n    next.status = 'paused';\n    next.pausedReason = `judge parse failures (${n} consecutive)`;\n    return {\n      newState: next,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'continue',\n      reason,\n      message: copy.pauseParse(n),\n    };\n  };\n\n  const pauseIfBudget = (reason: string): GoalPostTurnDecision | null => {\n    if (next.turnsUsed < next.maxTurns) return null;\n    next.status = 'paused';\n    next.pausedReason = `turn budget exhausted (${next.turnsUsed}/${next.maxTurns})`;\n    return {\n      newState: next,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'continue',\n      reason,\n      message: copy.pauseBudget(next.turnsUsed, next.maxTurns),\n    };\n  };\n\n  // ── Phase A: decompose (once per goal when checklist mode is on) ──\n  if (checklistMode && !next.decomposed) {\n    const dec = await decomposeGoalChecklist({\n      goal: next.goal,\n      judgeModelRef,\n      signal,\n      judgeTimeoutMs,\n      uiLocale: locale,\n    });\n    next.decomposed = true;\n    if (!dec.parseFailed && dec.items.length > 0) {\n      const prior = next.checklist ?? [];\n      next.checklist = mergeDecomposedChecklistItems(prior, dec.items);\n      const mergedCount = next.checklist.length;\n      next.lastVerdict = 'decompose';\n      next.lastReason = copy.decomposedReason(mergedCount);\n      resetParse();\n      const budgetEarly = pauseIfBudget(next.lastReason);\n      if (budgetEarly) return budgetEarly;\n      return {\n        newState: next,\n        shouldContinue: true,\n        continuationPrompt: buildContinuationPromptFromState(next, locale),\n        verdict: 'decompose',\n        reason: next.lastReason ?? '',\n        message: copy.checklistReady(mergedCount),\n      };\n    }\n    next.lastReason = dec.errorReason\n      ? copy.decomposeFallbackReason(dec.errorReason)\n      : copy.noChecklistFallback();\n  }\n\n  // ── Phase B: checklist judge ──\n  const checklist = next.checklist ?? [];\n  if (checklist.length > 0) {\n    const evalResult = await evaluateGoalChecklistJudge({\n      goal: next.goal,\n      numberedChecklist: renderChecklistNumbered(checklist),\n      lastResponse,\n      historyExcerpt: opts?.historyExcerpt ?? '',\n      judgeModelRef,\n      signal,\n      judgeTimeoutMs,\n      uiLocale: locale,\n    });\n\n    if (evalResult.parseFailed) {\n      bumpParse();\n    } else {\n      resetParse();\n      next.checklist = applyJudgeChecklistUpdates(checklist, {\n        updates: evalResult.parsed.updates,\n        newItems: evalResult.parsed.newItems,\n      });\n    }\n\n    next.lastVerdict = 'continue';\n    next.lastReason = evalResult.parsed.reason;\n\n    const completedChecklistItemIndexes = evalResult.parsed.updates\n      .filter((update) => update.status === 'completed' || update.status === 'impossible')\n      .map((update) => update.index);\n    const structuredNextAction = evalResult.parsed.nextAction?.trim();\n    const continuationPrompt = structuredNextAction || buildContinuationPromptFromState(next, locale);\n\n    if (!evalResult.parseFailed && allChecklistTerminal(next.checklist ?? [])) {\n      next.status = 'done';\n      next.lastVerdict = 'done';\n      return {\n        newState: next,\n        shouldContinue: false,\n        continuationPrompt: null,\n        verdict: 'done',\n        reason: evalResult.parsed.reason,\n        message: copy.goalAchieved(evalResult.parsed.reason),\n        confidence: evalResult.parsed.confidence,\n        missingEvidence: evalResult.parsed.missingEvidence,\n        userQuestion: evalResult.parsed.userQuestion,\n        completedChecklistItemIndexes,\n      };\n    }\n\n    if (!evalResult.parseFailed && (evalResult.parsed.verdict === 'blocked' || evalResult.parsed.verdict === 'needs_input')) {\n      next.status = 'paused';\n      next.pausedReason = evalResult.parsed.reason;\n      return {\n        newState: next,\n        shouldContinue: false,\n        continuationPrompt: null,\n        verdict: evalResult.parsed.verdict,\n        reason: evalResult.parsed.reason,\n        message: evalResult.parsed.verdict === 'needs_input'\n          ? `Goal needs input: ${evalResult.parsed.userQuestion || evalResult.parsed.reason}`\n          : `Goal blocked: ${evalResult.parsed.reason}`,\n        confidence: evalResult.parsed.confidence,\n        missingEvidence: evalResult.parsed.missingEvidence,\n        userQuestion: evalResult.parsed.userQuestion,\n        completedChecklistItemIndexes,\n      };\n    }\n\n    const parseStorm = pauseIfParseStorm(evalResult.parsed.reason);\n    if (parseStorm) return parseStorm;\n\n    const budget = pauseIfBudget(evalResult.parsed.reason);\n    if (budget) return budget;\n\n    const { total, completed, impossible } = checklistCounts(next.checklist ?? []);\n    const done = completed + impossible;\n    const progressSuffix = checklistProgressSuffix(done, total, locale);\n\n    return {\n      newState: next,\n      shouldContinue: true,\n      continuationPrompt,\n      verdict: 'continue',\n      reason: evalResult.parsed.reason,\n      message: copy.continuingWithProgress(\n        next.turnsUsed,\n        next.maxTurns,\n        progressSuffix,\n        evalResult.parsed.reason,\n      ),\n      confidence: evalResult.parsed.confidence,\n      missingEvidence: evalResult.parsed.missingEvidence,\n      userQuestion: evalResult.parsed.userQuestion,\n      completedChecklistItemIndexes,\n    };\n  }\n\n  // ── Freeform judge (no checklist) ──\n  const j = await judgeGoalHermesStyle(state.goal, lastResponse, judgeModelRef, signal, {\n    judgeTimeoutMs,\n    uiLocale: locale,\n  });\n  next.lastVerdict = j.verdict === 'skipped' ? 'skipped' : j.verdict;\n  next.lastReason = j.reason;\n\n  if (j.parseFailed) bumpParse();\n  else resetParse();\n\n  if (j.verdict === 'done') {\n    next.status = 'done';\n    return {\n      newState: next,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'done',\n      reason: j.reason,\n      message: copy.goalAchieved(j.reason),\n    };\n  }\n\n  const parseStorm = pauseIfParseStorm(j.reason);\n  if (parseStorm) return parseStorm;\n\n  const budget = pauseIfBudget(j.reason);\n  if (budget) return budget;\n\n  return {\n    newState: next,\n    shouldContinue: true,\n    continuationPrompt: buildContinuationPromptFromState(next, locale),\n    verdict: 'continue',\n    reason: j.reason,\n    message: copy.continuing(next.turnsUsed, next.maxTurns, j.reason),\n  };\n}\n"],"mappings":";;;;;;;AAsBA,MAAa,+BAA+B;;AAG5C,MAAa,8CAA8C;AAoB3D,SAAS,WAAW,GAA6C;AAC/D,QAAO;EACL,GAAG;EACH,WAAW,EAAE,WAAW,KAAK,QAAQ,EAAE,GAAG,IAAI,EAAE;EACjD;;;;;AAMH,eAAsB,4BACpB,OACA,cACA,eACA,QACA,MAC+B;CAC/B,MAAM,SAAS,MAAM,YAAY,oBAAoB,MAAM;CAC3D,MAAM,OAAO,qBAAqB,OAAO;AAEzC,KAAI,MAAM,WAAW,SACnB,QAAO;EACL,UAAU;EACV,gBAAgB;EAChB,oBAAoB;EACpB,SAAS;EACT,QAAQ;EACR,SAAS;EACV;CAGH,MAAM,QAAQ,MAAM;CACpB,MAAM,gBAAgB,OAAO,kBAAkB;CAC/C,MAAM,mBAAmB,OAAO,+BAA+B;CAC/D,MAAM,iBAAiB,OAAO;CAE9B,MAAM,OAAO,WAAW,MAAM;AAC9B,MAAK,aAAa;AAClB,MAAK,aAAa,KAAK,KAAK;CAE5B,MAAM,kBAAkB;AACtB,OAAK,4BAA4B,KAAK,4BAA4B,KAAK;;CAEzE,MAAM,mBAAmB;AACvB,OAAK,2BAA2B;;CAGlC,MAAM,qBAAqB,WAAgD;EACzE,MAAM,IAAI,KAAK,4BAA4B;AAC3C,MAAI,IAAI,iBAAkB,QAAO;AACjC,OAAK,SAAS;AACd,OAAK,eAAe,yBAAyB,EAAE;AAC/C,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB,oBAAoB;GACpB,SAAS;GACT;GACA,SAAS,KAAK,WAAW,EAAE;GAC5B;;CAGH,MAAM,iBAAiB,WAAgD;AACrE,MAAI,KAAK,YAAY,KAAK,SAAU,QAAO;AAC3C,OAAK,SAAS;AACd,OAAK,eAAe,0BAA0B,KAAK,UAAU,GAAG,KAAK,SAAS;AAC9E,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB,oBAAoB;GACpB,SAAS;GACT;GACA,SAAS,KAAK,YAAY,KAAK,WAAW,KAAK,SAAS;GACzD;;AAIH,KAAI,iBAAiB,CAAC,KAAK,YAAY;EACrC,MAAM,MAAM,MAAM,uBAAuB;GACvC,MAAM,KAAK;GACX;GACA;GACA;GACA,UAAU;GACX,CAAC;AACF,OAAK,aAAa;AAClB,MAAI,CAAC,IAAI,eAAe,IAAI,MAAM,SAAS,GAAG;AAE5C,QAAK,YAAY,8BADH,KAAK,aAAa,EAAE,EACoB,IAAI,MAAM;GAChE,MAAM,cAAc,KAAK,UAAU;AACnC,QAAK,cAAc;AACnB,QAAK,aAAa,KAAK,iBAAiB,YAAY;AACpD,eAAY;GACZ,MAAM,cAAc,cAAc,KAAK,WAAW;AAClD,OAAI,YAAa,QAAO;AACxB,UAAO;IACL,UAAU;IACV,gBAAgB;IAChB,oBAAoB,iCAAiC,MAAM,OAAO;IAClE,SAAS;IACT,QAAQ,KAAK,cAAc;IAC3B,SAAS,KAAK,eAAe,YAAY;IAC1C;;AAEH,OAAK,aAAa,IAAI,cAClB,KAAK,wBAAwB,IAAI,YAAY,GAC7C,KAAK,qBAAqB;;CAIhC,MAAM,YAAY,KAAK,aAAa,EAAE;AACtC,KAAI,UAAU,SAAS,GAAG;EACxB,MAAM,aAAa,MAAM,2BAA2B;GAClD,MAAM,KAAK;GACX,mBAAmB,wBAAwB,UAAU;GACrD;GACA,gBAAgB,MAAM,kBAAkB;GACxC;GACA;GACA;GACA,UAAU;GACX,CAAC;AAEF,MAAI,WAAW,YACb,YAAW;OACN;AACL,eAAY;AACZ,QAAK,YAAY,2BAA2B,WAAW;IACrD,SAAS,WAAW,OAAO;IAC3B,UAAU,WAAW,OAAO;IAC7B,CAAC;;AAGJ,OAAK,cAAc;AACnB,OAAK,aAAa,WAAW,OAAO;EAEpC,MAAM,gCAAgC,WAAW,OAAO,QACrD,QAAQ,WAAW,OAAO,WAAW,eAAe,OAAO,WAAW,aAAa,CACnF,KAAK,WAAW,OAAO,MAAM;EAEhC,MAAM,qBADuB,WAAW,OAAO,YAAY,MAAM,IACd,iCAAiC,MAAM,OAAO;AAEjG,MAAI,CAAC,WAAW,eAAe,qBAAqB,KAAK,aAAa,EAAE,CAAC,EAAE;AACzE,QAAK,SAAS;AACd,QAAK,cAAc;AACnB,UAAO;IACL,UAAU;IACV,gBAAgB;IAChB,oBAAoB;IACpB,SAAS;IACT,QAAQ,WAAW,OAAO;IAC1B,SAAS,KAAK,aAAa,WAAW,OAAO,OAAO;IACpD,YAAY,WAAW,OAAO;IAC9B,iBAAiB,WAAW,OAAO;IACnC,cAAc,WAAW,OAAO;IAChC;IACD;;AAGH,MAAI,CAAC,WAAW,gBAAgB,WAAW,OAAO,YAAY,aAAa,WAAW,OAAO,YAAY,gBAAgB;AACvH,QAAK,SAAS;AACd,QAAK,eAAe,WAAW,OAAO;AACtC,UAAO;IACL,UAAU;IACV,gBAAgB;IAChB,oBAAoB;IACpB,SAAS,WAAW,OAAO;IAC3B,QAAQ,WAAW,OAAO;IAC1B,SAAS,WAAW,OAAO,YAAY,gBACnC,qBAAqB,WAAW,OAAO,gBAAgB,WAAW,OAAO,WACzE,iBAAiB,WAAW,OAAO;IACvC,YAAY,WAAW,OAAO;IAC9B,iBAAiB,WAAW,OAAO;IACnC,cAAc,WAAW,OAAO;IAChC;IACD;;EAGH,MAAM,aAAa,kBAAkB,WAAW,OAAO,OAAO;AAC9D,MAAI,WAAY,QAAO;EAEvB,MAAM,SAAS,cAAc,WAAW,OAAO,OAAO;AACtD,MAAI,OAAQ,QAAO;EAEnB,MAAM,EAAE,OAAO,WAAW,eAAe,gBAAgB,KAAK,aAAa,EAAE,CAAC;EAE9E,MAAM,iBAAiB,wBADV,YAAY,YAC4B,OAAO,OAAO;AAEnE,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB;GACA,SAAS;GACT,QAAQ,WAAW,OAAO;GAC1B,SAAS,KAAK,uBACZ,KAAK,WACL,KAAK,UACL,gBACA,WAAW,OAAO,OACnB;GACD,YAAY,WAAW,OAAO;GAC9B,iBAAiB,WAAW,OAAO;GACnC,cAAc,WAAW,OAAO;GAChC;GACD;;CAIH,MAAM,IAAI,MAAM,qBAAqB,MAAM,MAAM,cAAc,eAAe,QAAQ;EACpF;EACA,UAAU;EACX,CAAC;AACF,MAAK,cAAc,EAAE,YAAY,YAAY,YAAY,EAAE;AAC3D,MAAK,aAAa,EAAE;AAEpB,KAAI,EAAE,YAAa,YAAW;KACzB,aAAY;AAEjB,KAAI,EAAE,YAAY,QAAQ;AACxB,OAAK,SAAS;AACd,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB,oBAAoB;GACpB,SAAS;GACT,QAAQ,EAAE;GACV,SAAS,KAAK,aAAa,EAAE,OAAO;GACrC;;CAGH,MAAM,aAAa,kBAAkB,EAAE,OAAO;AAC9C,KAAI,WAAY,QAAO;CAEvB,MAAM,SAAS,cAAc,EAAE,OAAO;AACtC,KAAI,OAAQ,QAAO;AAEnB,QAAO;EACL,UAAU;EACV,gBAAgB;EAChB,oBAAoB,iCAAiC,MAAM,OAAO;EAClE,SAAS;EACT,QAAQ,EAAE;EACV,SAAS,KAAK,WAAW,KAAK,WAAW,KAAK,UAAU,EAAE,OAAO;EAClE"}
+{"version":3,"file":"evaluate-turn.js","names":[],"sources":["../../../../src/agent/goals/evaluate-turn.ts"],"sourcesContent":["import type { GoalsConfig } from '../../config/schema.js';\n\nimport { decomposeGoalChecklist, evaluateGoalChecklistJudge } from './checklist-judge.js';\nimport { allChecklistTerminal, checklistCounts } from './checklist-types.js';\nimport {\n  buildContinuationPromptFromState,\n  checklistProgressSuffix,\n  CONTINUATION_PROMPT_TEMPLATE_EN,\n  CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE_EN,\n  goalEvaluateUserCopy,\n  resolveGoalUiLocale,\n  type GoalUiLocale,\n} from './goal-locale.js';\nimport { judgeGoalHermesStyle } from './judge.js';\nimport {\n  applyJudgeChecklistUpdates,\n  mergeDecomposedChecklistItems,\n  renderChecklistNumbered,\n  type PersistentGoalState,\n} from './state.js';\n\n/** Hermes `CONTINUATION_PROMPT_TEMPLATE` (English; parity / tests). */\nexport const CONTINUATION_PROMPT_TEMPLATE = CONTINUATION_PROMPT_TEMPLATE_EN;\n\n/** Hermes `CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE` (English; parity / tests). */\nexport const CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE = CONTINUATION_PROMPT_WITH_CHECKLIST_TEMPLATE_EN;\n\nexport type GoalsEvaluateConfigSlice = Pick<\n  GoalsConfig,\n  | 'checklistMode'\n  | 'checklistDecomposePolicy'\n  | 'maxConsecutiveParseFailures'\n  | 'judgeTimeoutMs'\n  | 'checklistHistoryChars'\n>;\n\nexport type GoalPostTurnDecision = {\n  newState: PersistentGoalState | null;\n  shouldContinue: boolean;\n  continuationPrompt: string | null;\n  verdict: 'done' | 'continue' | 'blocked' | 'needs_input' | 'skipped' | 'inactive' | 'decompose';\n  reason: string;\n  message: string;\n  confidence?: number;\n  missingEvidence?: string[];\n  userQuestion?: string;\n  completedChecklistItemIndexes?: number[];\n};\n\nfunction cloneState(s: PersistentGoalState): PersistentGoalState {\n  return {\n    ...s,\n    checklist: s.checklist?.map((it) => ({ ...it })),\n  };\n}\n\n/**\n * Hermes `GoalManager.evaluate_after_turn` equivalent (minus SessionDB I/O).\n */\nexport async function evaluateAfterTurnHermesLike(\n  state: PersistentGoalState,\n  lastResponse: string,\n  judgeModelRef: string,\n  signal?: AbortSignal,\n  opts?: { goalsSlice?: Partial<GoalsEvaluateConfigSlice>; historyExcerpt?: string; uiLocale?: GoalUiLocale },\n): Promise<GoalPostTurnDecision> {\n  const locale = opts?.uiLocale ?? resolveGoalUiLocale(state);\n  const copy = goalEvaluateUserCopy(locale);\n\n  if (state.status !== 'active') {\n    return {\n      newState: state,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'inactive',\n      reason: 'no active goal',\n      message: '',\n    };\n  }\n\n  const slice = opts?.goalsSlice;\n  const checklistMode = slice?.checklistMode !== false;\n  const maxParseFailures = slice?.maxConsecutiveParseFailures ?? 3;\n  const judgeTimeoutMs = slice?.judgeTimeoutMs;\n\n  const next = cloneState(state);\n  next.turnsUsed += 1;\n  next.lastTurnAt = Date.now();\n\n  const bumpParse = () => {\n    next.consecutiveParseFailures = (next.consecutiveParseFailures ?? 0) + 1;\n  };\n  const resetParse = () => {\n    next.consecutiveParseFailures = 0;\n  };\n\n  const pauseIfParseStorm = (reason: string): GoalPostTurnDecision | null => {\n    const n = next.consecutiveParseFailures ?? 0;\n    if (n < maxParseFailures) return null;\n    next.status = 'paused';\n    next.pausedReason = `judge parse failures (${n} consecutive)`;\n    return {\n      newState: next,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'continue',\n      reason,\n      message: copy.pauseParse(n),\n    };\n  };\n\n  const pauseIfBudget = (reason: string): GoalPostTurnDecision | null => {\n    if (next.turnsUsed < next.maxTurns) return null;\n    next.status = 'paused';\n    next.pausedReason = `turn budget exhausted (${next.turnsUsed}/${next.maxTurns})`;\n    return {\n      newState: next,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'continue',\n      reason,\n      message: copy.pauseBudget(next.turnsUsed, next.maxTurns),\n    };\n  };\n\n  // ── Phase A: decompose (once per goal when checklist mode is on) ──\n  if (checklistMode && !next.decomposed) {\n    const dec = await decomposeGoalChecklist({\n      goal: next.goal,\n      existingChecklist: next.checklist?.length ? renderChecklistNumbered(next.checklist) : undefined,\n      judgeModelRef,\n      signal,\n      judgeTimeoutMs,\n      uiLocale: locale,\n    });\n    next.decomposed = true;\n    if (!dec.parseFailed && dec.items.length > 0) {\n      const prior = next.checklist ?? [];\n      next.checklist = mergeDecomposedChecklistItems(prior, dec.items);\n      const mergedCount = next.checklist.length;\n      next.lastVerdict = 'decompose';\n      next.lastReason = copy.decomposedReason(mergedCount);\n      resetParse();\n      const budgetEarly = pauseIfBudget(next.lastReason);\n      if (budgetEarly) return budgetEarly;\n      return {\n        newState: next,\n        shouldContinue: true,\n        continuationPrompt: buildContinuationPromptFromState(next, locale),\n        verdict: 'decompose',\n        reason: next.lastReason ?? '',\n        message: copy.checklistReady(mergedCount),\n      };\n    }\n    next.lastReason = dec.errorReason\n      ? copy.decomposeFallbackReason(dec.errorReason)\n      : copy.noChecklistFallback();\n  }\n\n  // ── Phase B: checklist judge ──\n  const checklist = next.checklist ?? [];\n  if (checklist.length > 0) {\n    const evalResult = await evaluateGoalChecklistJudge({\n      goal: next.goal,\n      numberedChecklist: renderChecklistNumbered(checklist),\n      lastResponse,\n      historyExcerpt: opts?.historyExcerpt ?? '',\n      judgeModelRef,\n      signal,\n      judgeTimeoutMs,\n      uiLocale: locale,\n    });\n\n    if (evalResult.parseFailed) {\n      bumpParse();\n    } else {\n      resetParse();\n      next.checklist = applyJudgeChecklistUpdates(checklist, {\n        updates: evalResult.parsed.updates,\n        newItems: evalResult.parsed.newItems,\n      });\n    }\n\n    next.lastVerdict = 'continue';\n    next.lastReason = evalResult.parsed.reason;\n\n    const completedChecklistItemIndexes = evalResult.parsed.updates\n      .filter((update) => update.status === 'completed' || update.status === 'impossible')\n      .map((update) => update.index);\n    const structuredNextAction = evalResult.parsed.nextAction?.trim();\n    const continuationPrompt = structuredNextAction || buildContinuationPromptFromState(next, locale);\n\n    if (!evalResult.parseFailed && allChecklistTerminal(next.checklist ?? [])) {\n      next.status = 'done';\n      next.lastVerdict = 'done';\n      return {\n        newState: next,\n        shouldContinue: false,\n        continuationPrompt: null,\n        verdict: 'done',\n        reason: evalResult.parsed.reason,\n        message: copy.goalAchieved(evalResult.parsed.reason),\n        confidence: evalResult.parsed.confidence,\n        missingEvidence: evalResult.parsed.missingEvidence,\n        userQuestion: evalResult.parsed.userQuestion,\n        completedChecklistItemIndexes,\n      };\n    }\n\n    if (!evalResult.parseFailed && (evalResult.parsed.verdict === 'blocked' || evalResult.parsed.verdict === 'needs_input')) {\n      next.status = 'paused';\n      next.pausedReason = evalResult.parsed.reason;\n      return {\n        newState: next,\n        shouldContinue: false,\n        continuationPrompt: null,\n        verdict: evalResult.parsed.verdict,\n        reason: evalResult.parsed.reason,\n        message: evalResult.parsed.verdict === 'needs_input'\n          ? `Goal needs input: ${evalResult.parsed.userQuestion || evalResult.parsed.reason}`\n          : `Goal blocked: ${evalResult.parsed.reason}`,\n        confidence: evalResult.parsed.confidence,\n        missingEvidence: evalResult.parsed.missingEvidence,\n        userQuestion: evalResult.parsed.userQuestion,\n        completedChecklistItemIndexes,\n      };\n    }\n\n    const parseStorm = pauseIfParseStorm(evalResult.parsed.reason);\n    if (parseStorm) return parseStorm;\n\n    const budget = pauseIfBudget(evalResult.parsed.reason);\n    if (budget) return budget;\n\n    const { total, completed, impossible } = checklistCounts(next.checklist ?? []);\n    const done = completed + impossible;\n    const progressSuffix = checklistProgressSuffix(done, total, locale);\n\n    return {\n      newState: next,\n      shouldContinue: true,\n      continuationPrompt,\n      verdict: 'continue',\n      reason: evalResult.parsed.reason,\n      message: copy.continuingWithProgress(\n        next.turnsUsed,\n        next.maxTurns,\n        progressSuffix,\n        evalResult.parsed.reason,\n      ),\n      confidence: evalResult.parsed.confidence,\n      missingEvidence: evalResult.parsed.missingEvidence,\n      userQuestion: evalResult.parsed.userQuestion,\n      completedChecklistItemIndexes,\n    };\n  }\n\n  // ── Freeform judge (no checklist) ──\n  const j = await judgeGoalHermesStyle(state.goal, lastResponse, judgeModelRef, signal, {\n    judgeTimeoutMs,\n    uiLocale: locale,\n  });\n  next.lastVerdict = j.verdict === 'skipped' ? 'skipped' : j.verdict;\n  next.lastReason = j.reason;\n\n  if (j.parseFailed) bumpParse();\n  else resetParse();\n\n  if (j.verdict === 'done') {\n    next.status = 'done';\n    return {\n      newState: next,\n      shouldContinue: false,\n      continuationPrompt: null,\n      verdict: 'done',\n      reason: j.reason,\n      message: copy.goalAchieved(j.reason),\n    };\n  }\n\n  const parseStorm = pauseIfParseStorm(j.reason);\n  if (parseStorm) return parseStorm;\n\n  const budget = pauseIfBudget(j.reason);\n  if (budget) return budget;\n\n  return {\n    newState: next,\n    shouldContinue: true,\n    continuationPrompt: buildContinuationPromptFromState(next, locale),\n    verdict: 'continue',\n    reason: j.reason,\n    message: copy.continuing(next.turnsUsed, next.maxTurns, j.reason),\n  };\n}\n"],"mappings":";;;;;;;AAsBA,MAAa,+BAA+B;;AAG5C,MAAa,8CAA8C;AAwB3D,SAAS,WAAW,GAA6C;AAC/D,QAAO;EACL,GAAG;EACH,WAAW,EAAE,WAAW,KAAK,QAAQ,EAAE,GAAG,IAAI,EAAE;EACjD;;;;;AAMH,eAAsB,4BACpB,OACA,cACA,eACA,QACA,MAC+B;CAC/B,MAAM,SAAS,MAAM,YAAY,oBAAoB,MAAM;CAC3D,MAAM,OAAO,qBAAqB,OAAO;AAEzC,KAAI,MAAM,WAAW,SACnB,QAAO;EACL,UAAU;EACV,gBAAgB;EAChB,oBAAoB;EACpB,SAAS;EACT,QAAQ;EACR,SAAS;EACV;CAGH,MAAM,QAAQ,MAAM;CACpB,MAAM,gBAAgB,OAAO,kBAAkB;CAC/C,MAAM,mBAAmB,OAAO,+BAA+B;CAC/D,MAAM,iBAAiB,OAAO;CAE9B,MAAM,OAAO,WAAW,MAAM;AAC9B,MAAK,aAAa;AAClB,MAAK,aAAa,KAAK,KAAK;CAE5B,MAAM,kBAAkB;AACtB,OAAK,4BAA4B,KAAK,4BAA4B,KAAK;;CAEzE,MAAM,mBAAmB;AACvB,OAAK,2BAA2B;;CAGlC,MAAM,qBAAqB,WAAgD;EACzE,MAAM,IAAI,KAAK,4BAA4B;AAC3C,MAAI,IAAI,iBAAkB,QAAO;AACjC,OAAK,SAAS;AACd,OAAK,eAAe,yBAAyB,EAAE;AAC/C,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB,oBAAoB;GACpB,SAAS;GACT;GACA,SAAS,KAAK,WAAW,EAAE;GAC5B;;CAGH,MAAM,iBAAiB,WAAgD;AACrE,MAAI,KAAK,YAAY,KAAK,SAAU,QAAO;AAC3C,OAAK,SAAS;AACd,OAAK,eAAe,0BAA0B,KAAK,UAAU,GAAG,KAAK,SAAS;AAC9E,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB,oBAAoB;GACpB,SAAS;GACT;GACA,SAAS,KAAK,YAAY,KAAK,WAAW,KAAK,SAAS;GACzD;;AAIH,KAAI,iBAAiB,CAAC,KAAK,YAAY;EACrC,MAAM,MAAM,MAAM,uBAAuB;GACvC,MAAM,KAAK;GACX,mBAAmB,KAAK,WAAW,SAAS,wBAAwB,KAAK,UAAU,GAAG,KAAA;GACtF;GACA;GACA;GACA,UAAU;GACX,CAAC;AACF,OAAK,aAAa;AAClB,MAAI,CAAC,IAAI,eAAe,IAAI,MAAM,SAAS,GAAG;AAE5C,QAAK,YAAY,8BADH,KAAK,aAAa,EAAE,EACoB,IAAI,MAAM;GAChE,MAAM,cAAc,KAAK,UAAU;AACnC,QAAK,cAAc;AACnB,QAAK,aAAa,KAAK,iBAAiB,YAAY;AACpD,eAAY;GACZ,MAAM,cAAc,cAAc,KAAK,WAAW;AAClD,OAAI,YAAa,QAAO;AACxB,UAAO;IACL,UAAU;IACV,gBAAgB;IAChB,oBAAoB,iCAAiC,MAAM,OAAO;IAClE,SAAS;IACT,QAAQ,KAAK,cAAc;IAC3B,SAAS,KAAK,eAAe,YAAY;IAC1C;;AAEH,OAAK,aAAa,IAAI,cAClB,KAAK,wBAAwB,IAAI,YAAY,GAC7C,KAAK,qBAAqB;;CAIhC,MAAM,YAAY,KAAK,aAAa,EAAE;AACtC,KAAI,UAAU,SAAS,GAAG;EACxB,MAAM,aAAa,MAAM,2BAA2B;GAClD,MAAM,KAAK;GACX,mBAAmB,wBAAwB,UAAU;GACrD;GACA,gBAAgB,MAAM,kBAAkB;GACxC;GACA;GACA;GACA,UAAU;GACX,CAAC;AAEF,MAAI,WAAW,YACb,YAAW;OACN;AACL,eAAY;AACZ,QAAK,YAAY,2BAA2B,WAAW;IACrD,SAAS,WAAW,OAAO;IAC3B,UAAU,WAAW,OAAO;IAC7B,CAAC;;AAGJ,OAAK,cAAc;AACnB,OAAK,aAAa,WAAW,OAAO;EAEpC,MAAM,gCAAgC,WAAW,OAAO,QACrD,QAAQ,WAAW,OAAO,WAAW,eAAe,OAAO,WAAW,aAAa,CACnF,KAAK,WAAW,OAAO,MAAM;EAEhC,MAAM,qBADuB,WAAW,OAAO,YAAY,MAAM,IACd,iCAAiC,MAAM,OAAO;AAEjG,MAAI,CAAC,WAAW,eAAe,qBAAqB,KAAK,aAAa,EAAE,CAAC,EAAE;AACzE,QAAK,SAAS;AACd,QAAK,cAAc;AACnB,UAAO;IACL,UAAU;IACV,gBAAgB;IAChB,oBAAoB;IACpB,SAAS;IACT,QAAQ,WAAW,OAAO;IAC1B,SAAS,KAAK,aAAa,WAAW,OAAO,OAAO;IACpD,YAAY,WAAW,OAAO;IAC9B,iBAAiB,WAAW,OAAO;IACnC,cAAc,WAAW,OAAO;IAChC;IACD;;AAGH,MAAI,CAAC,WAAW,gBAAgB,WAAW,OAAO,YAAY,aAAa,WAAW,OAAO,YAAY,gBAAgB;AACvH,QAAK,SAAS;AACd,QAAK,eAAe,WAAW,OAAO;AACtC,UAAO;IACL,UAAU;IACV,gBAAgB;IAChB,oBAAoB;IACpB,SAAS,WAAW,OAAO;IAC3B,QAAQ,WAAW,OAAO;IAC1B,SAAS,WAAW,OAAO,YAAY,gBACnC,qBAAqB,WAAW,OAAO,gBAAgB,WAAW,OAAO,WACzE,iBAAiB,WAAW,OAAO;IACvC,YAAY,WAAW,OAAO;IAC9B,iBAAiB,WAAW,OAAO;IACnC,cAAc,WAAW,OAAO;IAChC;IACD;;EAGH,MAAM,aAAa,kBAAkB,WAAW,OAAO,OAAO;AAC9D,MAAI,WAAY,QAAO;EAEvB,MAAM,SAAS,cAAc,WAAW,OAAO,OAAO;AACtD,MAAI,OAAQ,QAAO;EAEnB,MAAM,EAAE,OAAO,WAAW,eAAe,gBAAgB,KAAK,aAAa,EAAE,CAAC;EAE9E,MAAM,iBAAiB,wBADV,YAAY,YAC4B,OAAO,OAAO;AAEnE,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB;GACA,SAAS;GACT,QAAQ,WAAW,OAAO;GAC1B,SAAS,KAAK,uBACZ,KAAK,WACL,KAAK,UACL,gBACA,WAAW,OAAO,OACnB;GACD,YAAY,WAAW,OAAO;GAC9B,iBAAiB,WAAW,OAAO;GACnC,cAAc,WAAW,OAAO;GAChC;GACD;;CAIH,MAAM,IAAI,MAAM,qBAAqB,MAAM,MAAM,cAAc,eAAe,QAAQ;EACpF;EACA,UAAU;EACX,CAAC;AACF,MAAK,cAAc,EAAE,YAAY,YAAY,YAAY,EAAE;AAC3D,MAAK,aAAa,EAAE;AAEpB,KAAI,EAAE,YAAa,YAAW;KACzB,aAAY;AAEjB,KAAI,EAAE,YAAY,QAAQ;AACxB,OAAK,SAAS;AACd,SAAO;GACL,UAAU;GACV,gBAAgB;GAChB,oBAAoB;GACpB,SAAS;GACT,QAAQ,EAAE;GACV,SAAS,KAAK,aAAa,EAAE,OAAO;GACrC;;CAGH,MAAM,aAAa,kBAAkB,EAAE,OAAO;AAC9C,KAAI,WAAY,QAAO;CAEvB,MAAM,SAAS,cAAc,EAAE,OAAO;AACtC,KAAI,OAAQ,QAAO;AAEnB,QAAO;EACL,UAAU;EACV,gBAAgB;EAChB,oBAAoB,iCAAiC,MAAM,OAAO;EAClE,SAAS;EACT,QAAQ,EAAE;EACV,SAAS,KAAK,WAAW,KAAK,WAAW,KAAK,UAAU,EAAE,OAAO;EAClE"}

package/dist/src/agent/goals/goal-locale.js

@@ -1,7 +1,7 @@
-import { checklistCounts } from "./checklist-types.js";
-import { renderChecklistPlain } from "./state.js";
 import { formatI18n } from "../../i18n/format.js";
 import { SERVER_LOCALES, isServerLocale, normalizeServerLocale, serverLocaleOrFallback } from "../../i18n/locale.js";
+import { checklistCounts } from "./checklist-types.js";
+import { renderChecklistPlain } from "./state.js";
 import { JUDGE_REASON_EN, goalsChecklistProgressSuffix, goalsContinuationChecklistTemplate, goalsContinuationPlainTemplate, goalsEvaluateCopy, goalsJudgeReason, goalsJudgeResponseLanguageNote } from "../../i18n/goals-bundle.js";
 //#region src/agent/goals/goal-locale.ts
 const GOAL_UI_LOCALES = SERVER_LOCALES;

package/dist/src/agent/goals/post-turn.d.ts

@@ -1,5 +1,10 @@
-import type { Config } from '../../config/schema.js';
+import type { Config, GoalsConfig } from '../../config/schema.js';
 import type { PersistentGoalApis } from './persistent-goal-apis.js';
+export declare function shouldRunInitialChecklistDecomposition(input: {
+    checklistLength: number;
+    turnsUsed: number;
+    checklistDecomposePolicy?: GoalsConfig['checklistDecomposePolicy'];
+}): boolean;
 export declare function handlePersistentGoalPostTurn(opts: {
     apis: PersistentGoalApis;
     sessionKey: string;

package/dist/src/agent/goals/post-turn.js

@@ -1,9 +1,9 @@
-import { createLogger } from "../../utils/logger/index.js";
-import { init_logger } from "../../utils/logger.js";
 import { getAgentDefaultModelRef, init_schema } from "../../config/schema.js";
-import { resolveGoalUiLocale } from "./goal-locale.js";
 import { resolveEffectiveAgentProfileForSession } from "../../config/agent-profile.js";
+import { createLogger } from "../../utils/logger/index.js";
+import { init_logger } from "../../utils/logger.js";
 import { GoalService } from "../../goals/goal-service.js";
+import { resolveGoalUiLocale } from "./goal-locale.js";
 import { evaluateAfterTurnHermesLike } from "./evaluate-turn.js";
 import "../../goals/index.js";
 //#region src/agent/goals/post-turn.ts
@@ -32,6 +32,10 @@ function resolveJudgeModelRef(config, sessionKey, storedJudge, runtimeSessionMod
 	if (runtimeSessionModelRef?.trim()) return runtimeSessionModelRef.trim();
 	if (config) return getAgentDefaultModelRef(config);
 }
+function shouldRunInitialChecklistDecomposition(input) {
+	if (input.checklistLength === 0) return true;
+	return input.checklistDecomposePolicy === "supplement_existing" && input.turnsUsed === 0;
+}
 async function appendAssistantLine(apis, sessionKey, text) {
 	await apis.appendAssistantReceipt(sessionKey, text);
 }
@@ -42,6 +46,12 @@ async function handlePersistentGoalPostTurn(opts) {
 	if (!await apis.getSessionMetadata(sessionKey)) return;
 	const goal = goalService.getActiveForSession(sessionKey);
 	if (!goal || goal.status !== "active") return;
+	const goalsCfg = config?.goals;
+	const shouldDecomposeChecklist = shouldRunInitialChecklistDecomposition({
+		checklistLength: goal.checklist.length,
+		turnsUsed: goal.turnsUsed,
+		checklistDecomposePolicy: goalsCfg?.checklistDecomposePolicy
+	});
 	const state = {
 		goal: goal.title,
 		status: "active",
@@ -51,7 +61,7 @@ async function handlePersistentGoalPostTurn(opts) {
 		lastTurnAt: goal.updatedAt,
 		lastReason: goal.blockedReason,
 		judgeModelRef: goal.judgeModelRef,
-		decomposed: goal.checklist.length > 0 ? true : void 0,
+		decomposed: shouldDecomposeChecklist ? void 0 : true,
 		uiLocale: goal.uiLocale,
 		checklist: goal.checklist.map((it) => ({
 			text: it.text,
@@ -67,7 +77,6 @@ async function handlePersistentGoalPostTurn(opts) {
 		log.warn({ sessionKey }, "Persistent goal: no judge model ref; skipping post-turn");
 		return;
 	}
-	const goalsCfg = config?.goals;
 	const historyCap = goalsCfg?.checklistHistoryChars ?? 24e3;
 	let historyExcerpt = "";
 	try {
@@ -145,6 +154,6 @@ async function handlePersistentGoalPostTurn(opts) {
 	if (decision.shouldContinue && decision.continuationPrompt) apis.scheduleContinuation(sessionKey, decision.continuationPrompt);
 }
 //#endregion
-export { handlePersistentGoalPostTurn };
+export { handlePersistentGoalPostTurn, shouldRunInitialChecklistDecomposition };
 
 //# sourceMappingURL=post-turn.js.map

package/dist/src/agent/goals/post-turn.js.map

@@ -1 +1 @@
-{"version":3,"file":"post-turn.js","names":[],"sources":["../../../../src/agent/goals/post-turn.ts"],"sourcesContent":["import type { AgentMessage } from '@earendil-works/pi-agent-core';\n\nimport type { Config } from '../../config/schema.js';\nimport { getAgentDefaultModelRef } from '../../config/schema.js';\nimport { resolveEffectiveAgentProfileForSession } from '../../config/agent-profile.js';\nimport { GoalService } from '../../goals/index.js';\nimport { createLogger } from '../../utils/logger.js';\n\nimport { evaluateAfterTurnHermesLike } from './evaluate-turn.js';\nimport { resolveGoalUiLocale } from './goal-locale.js';\nimport type { PersistentGoalApis } from './persistent-goal-apis.js';\nimport type { PersistentGoalState } from './state.js';\n\nconst log = createLogger('PersistentGoal');\nconst goalService = new GoalService();\n\nfunction buildHistoryExcerpt(messages: AgentMessage[], maxChars: number): string {\n  if (maxChars <= 0) return '';\n  try {\n    const raw = JSON.stringify(messages);\n    if (raw.length <= maxChars) return raw;\n    return raw.slice(-maxChars);\n  } catch {\n    return '';\n  }\n}\n\nfunction resolveJudgeModelRef(\n  config: Config | undefined,\n  sessionKey: string,\n  storedJudge?: string,\n  runtimeSessionModelRef?: string,\n): string | undefined {\n  const fromCfg = config?.goals?.judgeModelRef?.trim();\n  if (fromCfg) return fromCfg;\n  if (storedJudge?.trim()) return storedJudge.trim();\n  if (config) {\n    const profile = resolveEffectiveAgentProfileForSession(config, sessionKey);\n    if (profile.primaryModelRef?.trim()) return profile.primaryModelRef.trim();\n  }\n  if (runtimeSessionModelRef?.trim()) return runtimeSessionModelRef.trim();\n  if (config) return getAgentDefaultModelRef(config);\n  return undefined;\n}\n\nasync function appendAssistantLine(apis: PersistentGoalApis, sessionKey: string, text: string): Promise<void> {\n  await apis.appendAssistantReceipt(sessionKey, text);\n}\n\nexport async function handlePersistentGoalPostTurn(opts: {\n  apis: PersistentGoalApis;\n  sessionKey: string;\n  assistantPlainText: string;\n  aborted: boolean;\n  streamError?: string;\n  skipPersistentGoalPostTurn: boolean;\n  config: Config | undefined;\n  /** Active session model (overrides, per-agent defaults) when goals/global defaults are unset. */\n  runtimeSessionModelRef?: string;\n  signal?: AbortSignal;\n  /** Hermes-style: verdict line also sent as a normal outbound message (non-webchat). */\n  publishVerdictToChannel?: (text: string) => Promise<void>;\n  onGoalStatusUpdated?: (payload: {\n    goalId: string;\n    sessionKey: string;\n    previousStatus: string;\n    status: string;\n    goal: import('../../goals/types.js').GoalWithDetails;\n  }) => void;\n}): Promise<void> {\n  const {\n    apis,\n    sessionKey,\n    assistantPlainText,\n    aborted,\n    streamError,\n    skipPersistentGoalPostTurn,\n    config,\n    runtimeSessionModelRef,\n    signal,\n    publishVerdictToChannel,\n    onGoalStatusUpdated,\n  } = opts;\n\n  if (skipPersistentGoalPostTurn) return;\n  if (aborted || streamError) return;\n\n  const meta = await apis.getSessionMetadata(sessionKey);\n  if (!meta) return;\n\n  const goal = goalService.getActiveForSession(sessionKey);\n  if (!goal || goal.status !== 'active') return;\n\n  const state: PersistentGoalState = {\n    goal: goal.title,\n    status: 'active',\n    turnsUsed: goal.turnsUsed,\n    maxTurns: goal.maxTurns,\n    createdAt: goal.createdAt,\n    lastTurnAt: goal.updatedAt,\n    lastReason: goal.blockedReason,\n    judgeModelRef: goal.judgeModelRef,\n    decomposed: goal.checklist.length > 0 ? true : undefined,\n    uiLocale: goal.uiLocale,\n    checklist: goal.checklist.map((it) => ({\n      text: it.text,\n      status: it.status,\n      addedBy: it.addedBy,\n      addedAt: it.addedAt,\n      completedAt: it.completedAt,\n      evidence: it.evidenceSummary,\n    })),\n  };\n\n  const judgeRef = resolveJudgeModelRef(config, sessionKey, state.judgeModelRef, runtimeSessionModelRef);\n  if (!judgeRef) {\n    log.warn({ sessionKey }, 'Persistent goal: no judge model ref; skipping post-turn');\n    return;\n  }\n\n  const goalsCfg = config?.goals;\n  const historyCap = goalsCfg?.checklistHistoryChars ?? 24_000;\n  let historyExcerpt = '';\n  try {\n    const msgs = await apis.loadMessages(sessionKey);\n    historyExcerpt = buildHistoryExcerpt(msgs, historyCap);\n  } catch (err) {\n    log.debug(\n      { err, sessionKey },\n      `Persistent goal: could not load messages for judge context: ${err instanceof Error ? err.message : String(err)}`,\n    );\n  }\n\n  const decision = await evaluateAfterTurnHermesLike(state, assistantPlainText, judgeRef, signal, {\n    goalsSlice: goalsCfg,\n    historyExcerpt,\n    uiLocale: resolveGoalUiLocale(state),\n  });\n\n  if (decision.newState) {\n    const ns = decision.newState;\n    const completedChecklistItemIds = decision.completedChecklistItemIndexes\n      ?.map((index) => goal.checklist[index]?.id)\n      .filter((id): id is string => typeof id === 'string' && id.length > 0);\n    try {\n      const updatedGoal = goalService.syncPostTurnState({\n        goalId: goal.id,\n        sessionKey,\n        source: 'chat',\n        status: ns.status === 'done'\n          ? 'done'\n          : decision.verdict === 'blocked'\n            ? 'blocked'\n            : decision.verdict === 'needs_input'\n              ? 'needs_input'\n              : ns.status === 'paused'\n                ? 'paused'\n                : 'active',\n        turnsUsed: ns.turnsUsed,\n        maxTurns: ns.maxTurns,\n        reason: ns.pausedReason ?? decision.reason,\n        nextAction: decision.continuationPrompt ?? undefined,\n        assistantPreview: assistantPlainText,\n        verdict: decision.verdict === 'done'\n          ? 'done'\n          : decision.verdict === 'decompose'\n            ? 'decompose'\n            : decision.verdict === 'blocked'\n              ? 'blocked'\n              : decision.verdict === 'needs_input'\n                ? 'needs_input'\n                : 'continue',\n        confidence: decision.confidence,\n        missingEvidence: decision.missingEvidence,\n        userQuestion: decision.userQuestion,\n        completedChecklistItemIds,\n        checklist: ns.checklist?.map((it) => ({\n          text: it.text,\n          status: it.status,\n          addedBy: it.addedBy,\n          addedAt: it.addedAt,\n          completedAt: it.completedAt,\n          evidenceSummary: it.evidence,\n        })),\n      });\n      if (updatedGoal && updatedGoal.status !== goal.status) {\n        onGoalStatusUpdated?.({\n          goalId: goal.id,\n          sessionKey,\n          previousStatus: goal.status,\n          status: updatedGoal.status,\n          goal: updatedGoal,\n        });\n      }\n    } catch (err) {\n      log.warn(\n        { err, sessionKey, goalId: goal.id },\n        `Persistent goal: SQLite goal update failed: ${err instanceof Error ? err.message : String(err)}`,\n      );\n    }\n  }\n\n  if (decision.message) {\n    if (publishVerdictToChannel) {\n      try {\n        await publishVerdictToChannel(decision.message);\n      } catch (err) {\n        log.warn(\n          { err, sessionKey },\n          `Persistent goal: channel verdict send failed: ${err instanceof Error ? err.message : String(err)}`,\n        );\n      }\n    } else {\n      try {\n        await appendAssistantLine(apis, sessionKey, decision.message);\n      } catch (err) {\n        log.warn(\n          { err, sessionKey },\n          `Persistent goal: failed to append status message: ${err instanceof Error ? err.message : String(err)}`,\n        );\n      }\n    }\n  }\n\n  if (decision.shouldContinue && decision.continuationPrompt) {\n    apis.scheduleContinuation(sessionKey, decision.continuationPrompt);\n  }\n}\n"],"mappings":";;;;;;;;;aAGiE;aAGZ;AAOrD,MAAM,MAAM,aAAa,iBAAiB;AAC1C,MAAM,cAAc,IAAI,aAAa;AAErC,SAAS,oBAAoB,UAA0B,UAA0B;AAC/E,KAAI,YAAY,EAAG,QAAO;AAC1B,KAAI;EACF,MAAM,MAAM,KAAK,UAAU,SAAS;AACpC,MAAI,IAAI,UAAU,SAAU,QAAO;AACnC,SAAO,IAAI,MAAM,CAAC,SAAS;SACrB;AACN,SAAO;;;AAIX,SAAS,qBACP,QACA,YACA,aACA,wBACoB;CACpB,MAAM,UAAU,QAAQ,OAAO,eAAe,MAAM;AACpD,KAAI,QAAS,QAAO;AACpB,KAAI,aAAa,MAAM,CAAE,QAAO,YAAY,MAAM;AAClD,KAAI,QAAQ;EACV,MAAM,UAAU,uCAAuC,QAAQ,WAAW;AAC1E,MAAI,QAAQ,iBAAiB,MAAM,CAAE,QAAO,QAAQ,gBAAgB,MAAM;;AAE5E,KAAI,wBAAwB,MAAM,CAAE,QAAO,uBAAuB,MAAM;AACxE,KAAI,OAAQ,QAAO,wBAAwB,OAAO;;AAIpD,eAAe,oBAAoB,MAA0B,YAAoB,MAA6B;AAC5G,OAAM,KAAK,uBAAuB,YAAY,KAAK;;AAGrD,eAAsB,6BAA6B,MAoBjC;CAChB,MAAM,EACJ,MACA,YACA,oBACA,SACA,aACA,4BACA,QACA,wBACA,QACA,yBACA,wBACE;AAEJ,KAAI,2BAA4B;AAChC,KAAI,WAAW,YAAa;AAG5B,KAAI,CAAC,MADc,KAAK,mBAAmB,WAAW,CAC3C;CAEX,MAAM,OAAO,YAAY,oBAAoB,WAAW;AACxD,KAAI,CAAC,QAAQ,KAAK,WAAW,SAAU;CAEvC,MAAM,QAA6B;EACjC,MAAM,KAAK;EACX,QAAQ;EACR,WAAW,KAAK;EAChB,UAAU,KAAK;EACf,WAAW,KAAK;EAChB,YAAY,KAAK;EACjB,YAAY,KAAK;EACjB,eAAe,KAAK;EACpB,YAAY,KAAK,UAAU,SAAS,IAAI,OAAO,KAAA;EAC/C,UAAU,KAAK;EACf,WAAW,KAAK,UAAU,KAAK,QAAQ;GACrC,MAAM,GAAG;GACT,QAAQ,GAAG;GACX,SAAS,GAAG;GACZ,SAAS,GAAG;GACZ,aAAa,GAAG;GAChB,UAAU,GAAG;GACd,EAAE;EACJ;CAED,MAAM,WAAW,qBAAqB,QAAQ,YAAY,MAAM,eAAe,uBAAuB;AACtG,KAAI,CAAC,UAAU;AACb,MAAI,KAAK,EAAE,YAAY,EAAE,0DAA0D;AACnF;;CAGF,MAAM,WAAW,QAAQ;CACzB,MAAM,aAAa,UAAU,yBAAyB;CACtD,IAAI,iBAAiB;AACrB,KAAI;AAEF,mBAAiB,oBAAoB,MADlB,KAAK,aAAa,WAAW,EACL,WAAW;UAC/C,KAAK;AACZ,MAAI,MACF;GAAE;GAAK;GAAY,EACnB,+DAA+D,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GAChH;;CAGH,MAAM,WAAW,MAAM,4BAA4B,OAAO,oBAAoB,UAAU,QAAQ;EAC9F,YAAY;EACZ;EACA,UAAU,oBAAoB,MAAM;EACrC,CAAC;AAEF,KAAI,SAAS,UAAU;EACrB,MAAM,KAAK,SAAS;EACpB,MAAM,4BAA4B,SAAS,+BACvC,KAAK,UAAU,KAAK,UAAU,QAAQ,GAAG,CAC1C,QAAQ,OAAqB,OAAO,OAAO,YAAY,GAAG,SAAS,EAAE;AACxE,MAAI;GACF,MAAM,cAAc,YAAY,kBAAkB;IAChD,QAAQ,KAAK;IACb;IACA,QAAQ;IACR,QAAQ,GAAG,WAAW,SAClB,SACA,SAAS,YAAY,YACnB,YACA,SAAS,YAAY,gBACnB,gBACA,GAAG,WAAW,WACZ,WACA;IACV,WAAW,GAAG;IACd,UAAU,GAAG;IACb,QAAQ,GAAG,gBAAgB,SAAS;IACpC,YAAY,SAAS,sBAAsB,KAAA;IAC3C,kBAAkB;IAClB,SAAS,SAAS,YAAY,SAC1B,SACA,SAAS,YAAY,cACnB,cACA,SAAS,YAAY,YACnB,YACA,SAAS,YAAY,gBACnB,gBACA;IACV,YAAY,SAAS;IACrB,iBAAiB,SAAS;IAC1B,cAAc,SAAS;IACvB;IACA,WAAW,GAAG,WAAW,KAAK,QAAQ;KACpC,MAAM,GAAG;KACT,QAAQ,GAAG;KACX,SAAS,GAAG;KACZ,SAAS,GAAG;KACZ,aAAa,GAAG;KAChB,iBAAiB,GAAG;KACrB,EAAE;IACJ,CAAC;AACF,OAAI,eAAe,YAAY,WAAW,KAAK,OAC7C,uBAAsB;IACpB,QAAQ,KAAK;IACb;IACA,gBAAgB,KAAK;IACrB,QAAQ,YAAY;IACpB,MAAM;IACP,CAAC;WAEG,KAAK;AACZ,OAAI,KACF;IAAE;IAAK;IAAY,QAAQ,KAAK;IAAI,EACpC,+CAA+C,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GAChG;;;AAIL,KAAI,SAAS,QACX,KAAI,wBACF,KAAI;AACF,QAAM,wBAAwB,SAAS,QAAQ;UACxC,KAAK;AACZ,MAAI,KACF;GAAE;GAAK;GAAY,EACnB,iDAAiD,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GAClG;;KAGH,KAAI;AACF,QAAM,oBAAoB,MAAM,YAAY,SAAS,QAAQ;UACtD,KAAK;AACZ,MAAI,KACF;GAAE;GAAK;GAAY,EACnB,qDAAqD,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GACtG;;AAKP,KAAI,SAAS,kBAAkB,SAAS,mBACtC,MAAK,qBAAqB,YAAY,SAAS,mBAAmB"}
+{"version":3,"file":"post-turn.js","names":[],"sources":["../../../../src/agent/goals/post-turn.ts"],"sourcesContent":["import type { AgentMessage } from '@earendil-works/pi-agent-core';\n\nimport type { Config, GoalsConfig } from '../../config/schema.js';\nimport { getAgentDefaultModelRef } from '../../config/schema.js';\nimport { resolveEffectiveAgentProfileForSession } from '../../config/agent-profile.js';\nimport { GoalService } from '../../goals/index.js';\nimport { createLogger } from '../../utils/logger.js';\n\nimport { evaluateAfterTurnHermesLike } from './evaluate-turn.js';\nimport { resolveGoalUiLocale } from './goal-locale.js';\nimport type { PersistentGoalApis } from './persistent-goal-apis.js';\nimport type { PersistentGoalState } from './state.js';\n\nconst log = createLogger('PersistentGoal');\nconst goalService = new GoalService();\n\nfunction buildHistoryExcerpt(messages: AgentMessage[], maxChars: number): string {\n  if (maxChars <= 0) return '';\n  try {\n    const raw = JSON.stringify(messages);\n    if (raw.length <= maxChars) return raw;\n    return raw.slice(-maxChars);\n  } catch {\n    return '';\n  }\n}\n\nfunction resolveJudgeModelRef(\n  config: Config | undefined,\n  sessionKey: string,\n  storedJudge?: string,\n  runtimeSessionModelRef?: string,\n): string | undefined {\n  const fromCfg = config?.goals?.judgeModelRef?.trim();\n  if (fromCfg) return fromCfg;\n  if (storedJudge?.trim()) return storedJudge.trim();\n  if (config) {\n    const profile = resolveEffectiveAgentProfileForSession(config, sessionKey);\n    if (profile.primaryModelRef?.trim()) return profile.primaryModelRef.trim();\n  }\n  if (runtimeSessionModelRef?.trim()) return runtimeSessionModelRef.trim();\n  if (config) return getAgentDefaultModelRef(config);\n  return undefined;\n}\n\nexport function shouldRunInitialChecklistDecomposition(input: {\n  checklistLength: number;\n  turnsUsed: number;\n  checklistDecomposePolicy?: GoalsConfig['checklistDecomposePolicy'];\n}): boolean {\n  if (input.checklistLength === 0) return true;\n  return input.checklistDecomposePolicy === 'supplement_existing' && input.turnsUsed === 0;\n}\n\nasync function appendAssistantLine(apis: PersistentGoalApis, sessionKey: string, text: string): Promise<void> {\n  await apis.appendAssistantReceipt(sessionKey, text);\n}\n\nexport async function handlePersistentGoalPostTurn(opts: {\n  apis: PersistentGoalApis;\n  sessionKey: string;\n  assistantPlainText: string;\n  aborted: boolean;\n  streamError?: string;\n  skipPersistentGoalPostTurn: boolean;\n  config: Config | undefined;\n  /** Active session model (overrides, per-agent defaults) when goals/global defaults are unset. */\n  runtimeSessionModelRef?: string;\n  signal?: AbortSignal;\n  /** Hermes-style: verdict line also sent as a normal outbound message (non-webchat). */\n  publishVerdictToChannel?: (text: string) => Promise<void>;\n  onGoalStatusUpdated?: (payload: {\n    goalId: string;\n    sessionKey: string;\n    previousStatus: string;\n    status: string;\n    goal: import('../../goals/types.js').GoalWithDetails;\n  }) => void;\n}): Promise<void> {\n  const {\n    apis,\n    sessionKey,\n    assistantPlainText,\n    aborted,\n    streamError,\n    skipPersistentGoalPostTurn,\n    config,\n    runtimeSessionModelRef,\n    signal,\n    publishVerdictToChannel,\n    onGoalStatusUpdated,\n  } = opts;\n\n  if (skipPersistentGoalPostTurn) return;\n  if (aborted || streamError) return;\n\n  const meta = await apis.getSessionMetadata(sessionKey);\n  if (!meta) return;\n\n  const goal = goalService.getActiveForSession(sessionKey);\n  if (!goal || goal.status !== 'active') return;\n\n  const goalsCfg = config?.goals;\n  const shouldDecomposeChecklist = shouldRunInitialChecklistDecomposition({\n    checklistLength: goal.checklist.length,\n    turnsUsed: goal.turnsUsed,\n    checklistDecomposePolicy: goalsCfg?.checklistDecomposePolicy,\n  });\n  const state: PersistentGoalState = {\n    goal: goal.title,\n    status: 'active',\n    turnsUsed: goal.turnsUsed,\n    maxTurns: goal.maxTurns,\n    createdAt: goal.createdAt,\n    lastTurnAt: goal.updatedAt,\n    lastReason: goal.blockedReason,\n    judgeModelRef: goal.judgeModelRef,\n    decomposed: shouldDecomposeChecklist ? undefined : true,\n    uiLocale: goal.uiLocale,\n    checklist: goal.checklist.map((it) => ({\n      text: it.text,\n      status: it.status,\n      addedBy: it.addedBy,\n      addedAt: it.addedAt,\n      completedAt: it.completedAt,\n      evidence: it.evidenceSummary,\n    })),\n  };\n\n  const judgeRef = resolveJudgeModelRef(config, sessionKey, state.judgeModelRef, runtimeSessionModelRef);\n  if (!judgeRef) {\n    log.warn({ sessionKey }, 'Persistent goal: no judge model ref; skipping post-turn');\n    return;\n  }\n\n  const historyCap = goalsCfg?.checklistHistoryChars ?? 24_000;\n  let historyExcerpt = '';\n  try {\n    const msgs = await apis.loadMessages(sessionKey);\n    historyExcerpt = buildHistoryExcerpt(msgs, historyCap);\n  } catch (err) {\n    log.debug(\n      { err, sessionKey },\n      `Persistent goal: could not load messages for judge context: ${err instanceof Error ? err.message : String(err)}`,\n    );\n  }\n\n  const decision = await evaluateAfterTurnHermesLike(state, assistantPlainText, judgeRef, signal, {\n    goalsSlice: goalsCfg,\n    historyExcerpt,\n    uiLocale: resolveGoalUiLocale(state),\n  });\n\n  if (decision.newState) {\n    const ns = decision.newState;\n    const completedChecklistItemIds = decision.completedChecklistItemIndexes\n      ?.map((index) => goal.checklist[index]?.id)\n      .filter((id): id is string => typeof id === 'string' && id.length > 0);\n    try {\n      const updatedGoal = goalService.syncPostTurnState({\n        goalId: goal.id,\n        sessionKey,\n        source: 'chat',\n        status: ns.status === 'done'\n          ? 'done'\n          : decision.verdict === 'blocked'\n            ? 'blocked'\n            : decision.verdict === 'needs_input'\n              ? 'needs_input'\n              : ns.status === 'paused'\n                ? 'paused'\n                : 'active',\n        turnsUsed: ns.turnsUsed,\n        maxTurns: ns.maxTurns,\n        reason: ns.pausedReason ?? decision.reason,\n        nextAction: decision.continuationPrompt ?? undefined,\n        assistantPreview: assistantPlainText,\n        verdict: decision.verdict === 'done'\n          ? 'done'\n          : decision.verdict === 'decompose'\n            ? 'decompose'\n            : decision.verdict === 'blocked'\n              ? 'blocked'\n              : decision.verdict === 'needs_input'\n                ? 'needs_input'\n                : 'continue',\n        confidence: decision.confidence,\n        missingEvidence: decision.missingEvidence,\n        userQuestion: decision.userQuestion,\n        completedChecklistItemIds,\n        checklist: ns.checklist?.map((it) => ({\n          text: it.text,\n          status: it.status,\n          addedBy: it.addedBy,\n          addedAt: it.addedAt,\n          completedAt: it.completedAt,\n          evidenceSummary: it.evidence,\n        })),\n      });\n      if (updatedGoal && updatedGoal.status !== goal.status) {\n        onGoalStatusUpdated?.({\n          goalId: goal.id,\n          sessionKey,\n          previousStatus: goal.status,\n          status: updatedGoal.status,\n          goal: updatedGoal,\n        });\n      }\n    } catch (err) {\n      log.warn(\n        { err, sessionKey, goalId: goal.id },\n        `Persistent goal: SQLite goal update failed: ${err instanceof Error ? err.message : String(err)}`,\n      );\n    }\n  }\n\n  if (decision.message) {\n    if (publishVerdictToChannel) {\n      try {\n        await publishVerdictToChannel(decision.message);\n      } catch (err) {\n        log.warn(\n          { err, sessionKey },\n          `Persistent goal: channel verdict send failed: ${err instanceof Error ? err.message : String(err)}`,\n        );\n      }\n    } else {\n      try {\n        await appendAssistantLine(apis, sessionKey, decision.message);\n      } catch (err) {\n        log.warn(\n          { err, sessionKey },\n          `Persistent goal: failed to append status message: ${err instanceof Error ? err.message : String(err)}`,\n        );\n      }\n    }\n  }\n\n  if (decision.shouldContinue && decision.continuationPrompt) {\n    apis.scheduleContinuation(sessionKey, decision.continuationPrompt);\n  }\n}\n"],"mappings":";;;;;;;;;aAGiE;aAGZ;AAOrD,MAAM,MAAM,aAAa,iBAAiB;AAC1C,MAAM,cAAc,IAAI,aAAa;AAErC,SAAS,oBAAoB,UAA0B,UAA0B;AAC/E,KAAI,YAAY,EAAG,QAAO;AAC1B,KAAI;EACF,MAAM,MAAM,KAAK,UAAU,SAAS;AACpC,MAAI,IAAI,UAAU,SAAU,QAAO;AACnC,SAAO,IAAI,MAAM,CAAC,SAAS;SACrB;AACN,SAAO;;;AAIX,SAAS,qBACP,QACA,YACA,aACA,wBACoB;CACpB,MAAM,UAAU,QAAQ,OAAO,eAAe,MAAM;AACpD,KAAI,QAAS,QAAO;AACpB,KAAI,aAAa,MAAM,CAAE,QAAO,YAAY,MAAM;AAClD,KAAI,QAAQ;EACV,MAAM,UAAU,uCAAuC,QAAQ,WAAW;AAC1E,MAAI,QAAQ,iBAAiB,MAAM,CAAE,QAAO,QAAQ,gBAAgB,MAAM;;AAE5E,KAAI,wBAAwB,MAAM,CAAE,QAAO,uBAAuB,MAAM;AACxE,KAAI,OAAQ,QAAO,wBAAwB,OAAO;;AAIpD,SAAgB,uCAAuC,OAI3C;AACV,KAAI,MAAM,oBAAoB,EAAG,QAAO;AACxC,QAAO,MAAM,6BAA6B,yBAAyB,MAAM,cAAc;;AAGzF,eAAe,oBAAoB,MAA0B,YAAoB,MAA6B;AAC5G,OAAM,KAAK,uBAAuB,YAAY,KAAK;;AAGrD,eAAsB,6BAA6B,MAoBjC;CAChB,MAAM,EACJ,MACA,YACA,oBACA,SACA,aACA,4BACA,QACA,wBACA,QACA,yBACA,wBACE;AAEJ,KAAI,2BAA4B;AAChC,KAAI,WAAW,YAAa;AAG5B,KAAI,CAAC,MADc,KAAK,mBAAmB,WAAW,CAC3C;CAEX,MAAM,OAAO,YAAY,oBAAoB,WAAW;AACxD,KAAI,CAAC,QAAQ,KAAK,WAAW,SAAU;CAEvC,MAAM,WAAW,QAAQ;CACzB,MAAM,2BAA2B,uCAAuC;EACtE,iBAAiB,KAAK,UAAU;EAChC,WAAW,KAAK;EAChB,0BAA0B,UAAU;EACrC,CAAC;CACF,MAAM,QAA6B;EACjC,MAAM,KAAK;EACX,QAAQ;EACR,WAAW,KAAK;EAChB,UAAU,KAAK;EACf,WAAW,KAAK;EAChB,YAAY,KAAK;EACjB,YAAY,KAAK;EACjB,eAAe,KAAK;EACpB,YAAY,2BAA2B,KAAA,IAAY;EACnD,UAAU,KAAK;EACf,WAAW,KAAK,UAAU,KAAK,QAAQ;GACrC,MAAM,GAAG;GACT,QAAQ,GAAG;GACX,SAAS,GAAG;GACZ,SAAS,GAAG;GACZ,aAAa,GAAG;GAChB,UAAU,GAAG;GACd,EAAE;EACJ;CAED,MAAM,WAAW,qBAAqB,QAAQ,YAAY,MAAM,eAAe,uBAAuB;AACtG,KAAI,CAAC,UAAU;AACb,MAAI,KAAK,EAAE,YAAY,EAAE,0DAA0D;AACnF;;CAGF,MAAM,aAAa,UAAU,yBAAyB;CACtD,IAAI,iBAAiB;AACrB,KAAI;AAEF,mBAAiB,oBAAoB,MADlB,KAAK,aAAa,WAAW,EACL,WAAW;UAC/C,KAAK;AACZ,MAAI,MACF;GAAE;GAAK;GAAY,EACnB,+DAA+D,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GAChH;;CAGH,MAAM,WAAW,MAAM,4BAA4B,OAAO,oBAAoB,UAAU,QAAQ;EAC9F,YAAY;EACZ;EACA,UAAU,oBAAoB,MAAM;EACrC,CAAC;AAEF,KAAI,SAAS,UAAU;EACrB,MAAM,KAAK,SAAS;EACpB,MAAM,4BAA4B,SAAS,+BACvC,KAAK,UAAU,KAAK,UAAU,QAAQ,GAAG,CAC1C,QAAQ,OAAqB,OAAO,OAAO,YAAY,GAAG,SAAS,EAAE;AACxE,MAAI;GACF,MAAM,cAAc,YAAY,kBAAkB;IAChD,QAAQ,KAAK;IACb;IACA,QAAQ;IACR,QAAQ,GAAG,WAAW,SAClB,SACA,SAAS,YAAY,YACnB,YACA,SAAS,YAAY,gBACnB,gBACA,GAAG,WAAW,WACZ,WACA;IACV,WAAW,GAAG;IACd,UAAU,GAAG;IACb,QAAQ,GAAG,gBAAgB,SAAS;IACpC,YAAY,SAAS,sBAAsB,KAAA;IAC3C,kBAAkB;IAClB,SAAS,SAAS,YAAY,SAC1B,SACA,SAAS,YAAY,cACnB,cACA,SAAS,YAAY,YACnB,YACA,SAAS,YAAY,gBACnB,gBACA;IACV,YAAY,SAAS;IACrB,iBAAiB,SAAS;IAC1B,cAAc,SAAS;IACvB;IACA,WAAW,GAAG,WAAW,KAAK,QAAQ;KACpC,MAAM,GAAG;KACT,QAAQ,GAAG;KACX,SAAS,GAAG;KACZ,SAAS,GAAG;KACZ,aAAa,GAAG;KAChB,iBAAiB,GAAG;KACrB,EAAE;IACJ,CAAC;AACF,OAAI,eAAe,YAAY,WAAW,KAAK,OAC7C,uBAAsB;IACpB,QAAQ,KAAK;IACb;IACA,gBAAgB,KAAK;IACrB,QAAQ,YAAY;IACpB,MAAM;IACP,CAAC;WAEG,KAAK;AACZ,OAAI,KACF;IAAE;IAAK;IAAY,QAAQ,KAAK;IAAI,EACpC,+CAA+C,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GAChG;;;AAIL,KAAI,SAAS,QACX,KAAI,wBACF,KAAI;AACF,QAAM,wBAAwB,SAAS,QAAQ;UACxC,KAAK;AACZ,MAAI,KACF;GAAE;GAAK;GAAY,EACnB,iDAAiD,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GAClG;;KAGH,KAAI;AACF,QAAM,oBAAoB,MAAM,YAAY,SAAS,QAAQ;UACtD,KAAK;AACZ,MAAI,KACF;GAAE;GAAK;GAAY,EACnB,qDAAqD,eAAe,QAAQ,IAAI,UAAU,OAAO,IAAI,GACtG;;AAKP,KAAI,SAAS,kBAAkB,SAAS,mBACtC,MAAK,qBAAqB,YAAY,SAAS,mBAAmB"}

package/dist/src/agent/image/load-image-media.js

@@ -1,8 +1,8 @@
 import { checkFileSafety } from "../prompt/safety.js";
 import { decodeDataUrl } from "./image-helpers.js";
-import { promises } from "node:fs";
-import path from "node:path";
 import { homedir } from "node:os";
+import path from "node:path";
+import { promises } from "node:fs";
 //#region src/agent/image/load-image-media.ts
 function expandUser(p) {
 	if (p.startsWith("~/") || p === "~") return path.join(homedir(), p.slice(1));

package/dist/src/agent/ipc/inbox.js

@@ -1,9 +1,9 @@
-import { init_write_file_atomic, writeTextAtomic } from "../../infra/write-file-atomic.js";
-import { init_paths, resolveAgentDir } from "../../config/paths.js";
 import { createLogger } from "../../utils/logger/index.js";
 import { init_logger } from "../../utils/logger.js";
-import { existsSync, watch } from "fs";
+import { init_paths, resolveAgentDir } from "../../config/paths.js";
+import { init_write_file_atomic, writeTextAtomic } from "../../infra/write-file-atomic.js";
 import { join } from "path";
+import { existsSync, watch } from "fs";
 import { mkdir, readFile, readdir, rename } from "fs/promises";
 //#region src/agent/ipc/inbox.ts
 init_write_file_atomic();

package/dist/src/agent/ipc/socket.js

@@ -1,6 +1,6 @@
-import { init_paths, resolveSocketPath } from "../../config/paths.js";
 import { createLogger } from "../../utils/logger/index.js";
 import { init_logger } from "../../utils/logger.js";
+import { init_paths, resolveSocketPath } from "../../config/paths.js";
 import { isValidIPCMessage } from "./types.js";
 import { dirname } from "path";
 import { mkdir } from "fs/promises";

package/dist/src/agent/mcp/bundle-mcp-materialize.js

@@ -1,6 +1,6 @@
+import { init_string_coerce, normalizeLowercaseStringOrEmpty, normalizeOptionalString } from "../../utils/string-coerce.js";
 import { createLogger } from "../../utils/logger/index.js";
 import { init_logger } from "../../utils/logger.js";
-import { init_string_coerce, normalizeLowercaseStringOrEmpty, normalizeOptionalString } from "../../utils/string-coerce.js";
 import { buildSafeToolName, normalizeReservedToolNames } from "./bundle-mcp-names.js";
 import { setPluginToolMeta } from "./mcp-tool-meta.js";
 import crypto from "node:crypto";

package/dist/src/agent/mcp/bundle-mcp-runtime.js

@@ -1,6 +1,6 @@
+import { init_string_coerce, normalizeOptionalString } from "../../utils/string-coerce.js";
 import { createLogger } from "../../utils/logger/index.js";
 import { init_logger } from "../../utils/logger.js";
-import { init_string_coerce, normalizeOptionalString } from "../../utils/string-coerce.js";
 import { resolveGlobalSingleton } from "../../utils/global-singleton.js";
 import { sanitizeServerName } from "./bundle-mcp-names.js";
 import { redactSensitiveUrlLikeString } from "../../utils/redact-sensitive-url.js";

package/dist/src/agent/mcp/mcp-transport-config.js

@@ -1,6 +1,6 @@
+import { init_string_coerce, normalizeLowercaseStringOrEmpty } from "../../utils/string-coerce.js";
 import { createLogger } from "../../utils/logger/index.js";
 import { init_logger } from "../../utils/logger.js";
-import { init_string_coerce, normalizeLowercaseStringOrEmpty } from "../../utils/string-coerce.js";
 import { resolveXopcMcpTransportAlias } from "../../config/mcp-config-normalize.js";
 import { sanitizeForLog } from "../../utils/sanitize-log.js";
 import { describeHttpMcpServerLaunchConfig, resolveHttpMcpServerLaunchConfig } from "./mcp-http.js";

package/dist/src/agent/mcp/mcp-transport.js

@@ -1,6 +1,6 @@
+import { init_string_coerce, normalizeOptionalString } from "../../utils/string-coerce.js";
 import { createLogger } from "../../utils/logger/index.js";
 import { init_logger } from "../../utils/logger.js";
-import { init_string_coerce, normalizeOptionalString } from "../../utils/string-coerce.js";
 import { XopcStdioClientTransport } from "./mcp-stdio-transport.js";
 import { resolveMcpTransportConfig } from "./mcp-transport-config.js";
 import { SSEClientTransport } from "@modelcontextprotocol/sdk/client/sse.js";

package/dist/src/agent/memory/builtin-memory-store.js

@@ -1,6 +1,6 @@
+import { join } from "node:path";
 import { existsSync, mkdirSync, readFileSync } from "node:fs";
 import { mkdir, readFile, rename, writeFile } from "node:fs/promises";
-import { join } from "node:path";
 import lockfile from "proper-lockfile";
 //#region src/agent/memory/builtin-memory-store.ts
 /**