bosun 0.41.2 → 0.41.4

package/agent/agent-supervisor.mjs

@@ -30,6 +30,12 @@
  */
 
 const TAG = "[agent-supervisor]";
+const API_ERROR_CONTINUE_COOLDOWNS_MS = Object.freeze([
+  3 * 60_000,
+  5 * 60_000,
+  5 * 60_000,
+]);
+const API_ERROR_RECOVERY_RESET_MS = 15 * 60_000;
 
 // ── Situation Types (30+ edge cases) ────────────────────────────────────────
 
@@ -140,7 +146,7 @@ const INTERVENTION_LADDER = {
   [SITUATION.PRE_PUSH_FAILURE]:   [INTERVENTION.INJECT_PROMPT, INTERVENTION.INJECT_PROMPT, INTERVENTION.FORCE_NEW_THREAD, INTERVENTION.BLOCK_AND_NOTIFY],
 
   [SITUATION.RATE_LIMITED]:       [INTERVENTION.COOLDOWN, INTERVENTION.COOLDOWN, INTERVENTION.PAUSE_EXECUTOR],
-  [SITUATION.API_ERROR]:          [INTERVENTION.COOLDOWN, INTERVENTION.COOLDOWN, INTERVENTION.BLOCK_AND_NOTIFY],
+  [SITUATION.API_ERROR]:          [INTERVENTION.FORCE_NEW_THREAD, INTERVENTION.REDISPATCH_TASK, INTERVENTION.BLOCK_AND_NOTIFY],
   [SITUATION.TOKEN_OVERFLOW]:     [INTERVENTION.FORCE_NEW_THREAD, INTERVENTION.FORCE_NEW_THREAD, INTERVENTION.BLOCK_AND_NOTIFY],
   [SITUATION.SESSION_EXPIRED]:    [INTERVENTION.FORCE_NEW_THREAD, INTERVENTION.FORCE_NEW_THREAD, INTERVENTION.BLOCK_AND_NOTIFY],
   [SITUATION.MODEL_ERROR]:        [INTERVENTION.BLOCK_AND_NOTIFY],       // Not retryable — wrong model name
@@ -263,9 +269,12 @@ const RECOVERY_PROMPTS = {
   [SITUATION.REBASE_SPIRAL]: () =>
     `You're stuck in a rebase loop. STOP rebasing and try:\n` +
     `1. git rebase --abort\n` +
-    `2. git checkout -B <your-branch> origin/main\n` +
-    `3. Re-apply your changes manually\n` +
-    `4. Commit and push`,
+    `2. git stash (save any local changes)\n` +
+    `3. git fetch origin && git merge origin/main --no-edit\n` +
+    `4. If merge conflicts, resolve them manually\n` +
+    `5. git stash pop (if you stashed changes)\n` +
+    `6. Commit and push\n` +
+    `WARNING: Do NOT use git checkout -B to reset your branch to origin/main — this destroys all PR changes.`,
 
   [SITUATION.THOUGHT_SPINNING]: () =>
     `You've been reasoning without taking action. STOP thinking and START doing:\n` +
@@ -440,13 +449,15 @@ export class AgentSupervisor {
     const signals = this._gatherSignals(taskId, context);
     const situation = this._diagnose(signals, context);
     const healthScore = this._computeHealthScore(signals);
+    const recoveryOverride = this._selectRecoveryIntervention(taskId, situation, context, state);
     const attemptIndex = Math.min(
       state.interventionCount,
       (INTERVENTION_LADDER[situation] || [INTERVENTION.NONE]).length - 1,
     );
-    const intervention = (INTERVENTION_LADDER[situation] || [INTERVENTION.NONE])[attemptIndex];
+    const intervention = recoveryOverride?.intervention
+      || (INTERVENTION_LADDER[situation] || [INTERVENTION.NONE])[attemptIndex];
     const prompt = this._buildPrompt(situation, taskId, context);
-    const reason = this._buildReason(situation, signals, context);
+    const reason = recoveryOverride?.reason || this._buildReason(situation, signals, context);
 
     // Record
     state.situationHistory.push({ situation, ts: Date.now() });
@@ -482,6 +493,9 @@ export class AgentSupervisor {
           break;
 
         case INTERVENTION.CONTINUE_SIGNAL:
+          if (situation === SITUATION.API_ERROR) {
+            this._recordApiErrorContinue(taskId);
+          }
           if (this._sendContinueSignal) {
             this._sendContinueSignal(taskId);
           }
@@ -722,6 +736,12 @@ export class AgentSupervisor {
       qualityScore: state.qualityScore,
       reviewVerdict: state.reviewVerdict,
       reviewIssueCount: state.reviewIssues?.length || 0,
+      apiErrorRecovery: state.apiErrorRecovery
+        ? {
+            ...state.apiErrorRecovery,
+            cooldownRemainingMs: Math.max(0, Number(state.apiErrorRecovery.cooldownUntil || 0) - Date.now()),
+          }
+        : null,
       recentSituations: state.situationHistory.slice(-10),
     };
   }
@@ -786,11 +806,104 @@ export class AgentSupervisor {
         qualityScore: null,
         reviewVerdict: null,
         reviewIssues: null,
+        apiErrorRecovery: null,
       });
     }
     return this._taskState.get(taskId);
   }
 
+  _normalizeApiErrorSignature(context) {
+    const raw = String(context?.error || context?.output || "").trim().toLowerCase();
+    if (!raw) return "api_error";
+    return raw
+      .replace(/\s+/g, " ")
+      .replace(/\b\d{2,}\b/g, "#")
+      .slice(0, 240);
+  }
+
+  _selectRecoveryIntervention(taskId, situation, context, state) {
+    if (situation !== SITUATION.API_ERROR) {
+      if (state?.apiErrorRecovery) state.apiErrorRecovery = null;
+      return null;
+    }
+
+    const now = Date.now();
+    const signature = this._normalizeApiErrorSignature(context);
+    const current = state.apiErrorRecovery || {
+      signature,
+      continueAttempts: 0,
+      lastErrorAt: 0,
+      cooldownUntil: 0,
+    };
+
+    const shouldReset =
+      current.signature !== signature ||
+      (current.lastErrorAt > 0 && now - current.lastErrorAt > API_ERROR_RECOVERY_RESET_MS);
+
+    const nextState = shouldReset
+      ? {
+          signature,
+          continueAttempts: 0,
+          lastErrorAt: now,
+          cooldownUntil: 0,
+        }
+      : {
+          ...current,
+          signature,
+          lastErrorAt: now,
+        };
+
+    state.apiErrorRecovery = nextState;
+
+    if (Number(nextState.cooldownUntil || 0) > now) {
+      const remainingMs = Math.max(0, nextState.cooldownUntil - now);
+      return {
+        intervention: INTERVENTION.COOLDOWN,
+        reason: `Transient API failure on cooldown for ${Math.ceil(remainingMs / 60000)} minute(s) before retrying the same thread.`,
+      };
+    }
+
+    if (nextState.continueAttempts < API_ERROR_CONTINUE_COOLDOWNS_MS.length) {
+      const cooldownMs = API_ERROR_CONTINUE_COOLDOWNS_MS[nextState.continueAttempts];
+      return {
+        intervention: INTERVENTION.CONTINUE_SIGNAL,
+        reason: `Transient API failure — continue the current thread and back off for ${Math.ceil(cooldownMs / 60000)} minute(s) if it repeats.`,
+      };
+    }
+
+    const ladder = INTERVENTION_LADDER[SITUATION.API_ERROR] || [INTERVENTION.BLOCK_AND_NOTIFY];
+    const escalationIndex = Math.min(
+      nextState.continueAttempts - API_ERROR_CONTINUE_COOLDOWNS_MS.length,
+      ladder.length - 1,
+    );
+    const escalation = ladder[escalationIndex];
+    const escalationReason = escalation === INTERVENTION.FORCE_NEW_THREAD
+      ? "Repeated API failures survived 3 continue attempts — forcing a fresh thread."
+      : escalation === INTERVENTION.REDISPATCH_TASK
+        ? "Repeated API failures survived continue attempts and a fresh thread — redispatching the task."
+        : "Repeated API failures survived all automated recovery attempts — blocking for human review.";
+    return {
+      intervention: escalation,
+      reason: escalationReason,
+    };
+  }
+
+  _recordApiErrorContinue(taskId) {
+    const state = this._getTaskState(taskId);
+    if (!state?.apiErrorRecovery) return;
+    const attemptIndex = Math.min(
+      state.apiErrorRecovery.continueAttempts,
+      API_ERROR_CONTINUE_COOLDOWNS_MS.length - 1,
+    );
+    const cooldownMs = API_ERROR_CONTINUE_COOLDOWNS_MS[attemptIndex] || 0;
+    state.apiErrorRecovery = {
+      ...state.apiErrorRecovery,
+      continueAttempts: Number(state.apiErrorRecovery.continueAttempts || 0) + 1,
+      cooldownUntil: cooldownMs > 0 ? Date.now() + cooldownMs : 0,
+      lastErrorAt: Date.now(),
+    };
+  }
+
   _getTaskState(taskId) {
     return this._taskState.get(taskId) || null;
   }

package/agent/autofix-git.mjs

@@ -0,0 +1,33 @@
+import { execSync } from "node:child_process";
+
+export function detectChangedFiles(repoRoot) {
+  try {
+    const output = execSync("git diff --name-only", {
+      cwd: repoRoot,
+      encoding: "utf8",
+      timeout: 10_000,
+      stdio: ["pipe", "pipe", "pipe"],
+    });
+    return output
+      .split(/\r?\n/)
+      .map((filePath) => filePath.trim())
+      .filter(Boolean);
+  } catch {
+    return [];
+  }
+}
+
+export function getChangeSummary(repoRoot, files) {
+  if (!files.length) return "(no file changes detected)";
+  try {
+    const diff = execSync("git diff --stat", {
+      cwd: repoRoot,
+      encoding: "utf8",
+      timeout: 10_000,
+      stdio: ["pipe", "pipe", "pipe"],
+    });
+    return diff.trim() || files.join(", ");
+  } catch {
+    return files.join(", ");
+  }
+}

package/agent/autofix-prompts.mjs

@@ -0,0 +1,151 @@
+import { resolvePromptTemplate } from "./agent-prompts.mjs";
+
+function buildRecentMessagesContext(recentMessages) {
+  if (!recentMessages || !recentMessages.length) return "";
+  const msgs = recentMessages.slice(-15);
+  return `
+## Recent monitor notifications (for context — shows what led to this crash)
+${msgs.map((message, index) => `[${index + 1}] ${message}`).join("\n")}
+`;
+}
+
+export function buildFixPrompt(
+  error,
+  sourceContext,
+  reason,
+  recentMessages,
+  promptTemplate = "",
+) {
+  const messagesCtx = buildRecentMessagesContext(recentMessages);
+
+  const fallback = `You are a PowerShell expert fixing a crash in a running orchestrator script.
+
+## Error
+Type: ${error.errorType}
+File: ${error.file}
+Line: ${error.line}${error.column ? `\nColumn: ${error.column}` : ""}
+Message: ${error.message}${error.codeLine ? `\nFailing code: ${error.codeLine}` : ""}
+Crash reason: ${reason}
+
+## Source context around line ${error.line}
+\`\`\`powershell
+${sourceContext}
+\`\`\`
+${messagesCtx}
+## Instructions
+1. Read the file "${error.file}"
+2. Identify the root cause of the error at line ${error.line}
+3. Fix ONLY the bug — minimal change, don't refactor unrelated code
+4. Common PowerShell pitfalls:
+   - \`+=\` on arrays with single items fails — use [List[object]] or @() wrapping
+   - \`$a + $b\` on PSObjects fails — iterate and add individually
+   - Pipeline output can be a single object, not an array — always wrap with @()
+   - \`$null.Method()\` crashes — add null guards
+   - Named mutex with "Global\\" prefix fails on non-elevated Windows — use plain names
+   - \`$Var:\` is treated as a scope-qualified variable — use \`\${Var}:\` to embed colon in string
+   - ParserError: check for syntax issues like unclosed brackets, bad string interpolation
+5. Write the fix to the file. Do NOT create new files or refactor other functions.
+6. Keep all existing functionality intact.`;
+  return resolvePromptTemplate(
+    promptTemplate,
+    {
+      ERROR_TYPE: error.errorType,
+      ERROR_FILE: error.file,
+      ERROR_LINE: error.line,
+      ERROR_COLUMN_LINE: error.column ? `Column: ${error.column}` : "",
+      ERROR_MESSAGE: error.message,
+      ERROR_CODE_LINE: error.codeLine ? `Failing code: ${error.codeLine}` : "",
+      CRASH_REASON: reason,
+      SOURCE_CONTEXT: sourceContext,
+      RECENT_MESSAGES_CONTEXT: messagesCtx,
+    },
+    fallback,
+  );
+}
+
+export function buildFallbackPrompt(
+  fallback,
+  recentMessages,
+  promptTemplate = "",
+) {
+  const messagesCtx = buildRecentMessagesContext(recentMessages);
+
+  const defaultPrompt = `You are a PowerShell expert analyzing an orchestrator script crash.
+No structured error was extracted — the process terminated with: ${fallback.reason}
+
+## Error indicators from log tail
+${fallback.errorLines.length > 0 ? fallback.errorLines.join("\n") : "(no explicit error lines detected — possible SIGKILL, OOM, or silent crash)"}
+
+## Last ${Math.min(80, fallback.lineCount)} lines of crash log
+\`\`\`
+${fallback.tail}
+\`\`\`
+${messagesCtx}
+## Instructions
+1. Analyze the log for the root cause of the crash
+2. The main orchestrator script is: scripts/bosun/ve-orchestrator.ps1
+3. If you can identify a fixable bug, apply a minimal fix to the file
+4. Common crash causes:
+   - PowerShell syntax errors (\$Var: treated as scope, missing brackets)
+   - Array/object operation errors (+=, +, pipeline single-item issues)
+   - Null reference errors on optional API responses
+   - Infinite loops or stack overflow from recursive calls
+   - Exit code 4294967295 = unsigned overflow from uncaught exception
+5. If the crash is external (SIGKILL, OOM) with no code bug, do nothing
+6. Write any fix directly to the file. Keep existing functionality intact.`;
+  return resolvePromptTemplate(
+    promptTemplate,
+    {
+      FALLBACK_REASON: fallback.reason,
+      FALLBACK_ERROR_LINES:
+        fallback.errorLines.length > 0
+          ? fallback.errorLines.join("\n")
+          : "(no explicit error lines detected — possible SIGKILL, OOM, or silent crash)",
+      FALLBACK_LINE_COUNT: Math.min(80, fallback.lineCount),
+      FALLBACK_TAIL: fallback.tail,
+      RECENT_MESSAGES_CONTEXT: messagesCtx,
+    },
+    defaultPrompt,
+  );
+}
+
+export function buildLoopPrompt(
+  errorLine,
+  repeatCount,
+  recentMessages,
+  promptTemplate = "",
+) {
+  const messagesCtx = buildRecentMessagesContext(recentMessages);
+
+  const defaultPrompt = `You are a PowerShell expert fixing a loop bug in a running orchestrator script.
+
+## Problem
+The following error line is repeating ${repeatCount} times in the orchestrator output,
+indicating an infinite retry loop that needs to be fixed:
+
+"${errorLine}"
+
+${messagesCtx}
+
+## Instructions
+1. The main script is: scripts/bosun/ve-orchestrator.ps1
+2. Search for the code that produces this error message
+3. Identify why it loops (missing break/continue/return, no state change between iterations, etc.)
+4. Fix the loop by adding proper exit conditions, error handling, or state tracking
+5. Common loop-causing patterns in this codebase:
+   - PR lifecycle handoff repeatedly retried with no diff between branch and base
+   - API calls returning the same error repeatedly with no backoff or give-up logic
+   - Status not updated after failure → next cycle tries the same thing
+   - Missing \`continue\` or state change in foreach loops over tracked attempts
+6. Apply a minimal fix. Do NOT refactor unrelated code.
+7. Write the fix directly to the file.`;
+  return resolvePromptTemplate(
+    promptTemplate,
+    {
+      REPEAT_COUNT: repeatCount,
+      ERROR_LINE: errorLine,
+      RECENT_MESSAGES_CONTEXT: messagesCtx,
+    },
+    defaultPrompt,
+  );
+}

package/agent/autofix.mjs

@@ -27,14 +27,19 @@
  *  - Raw log fallback: when no structured errors found, feeds raw tail to Codex
  */
 
-import { spawn, execSync } from "node:child_process";
+import { spawn } from "node:child_process";
 import { existsSync, mkdirSync, createWriteStream, readFileSync } from "node:fs";
 import { readFile, writeFile } from "node:fs/promises";
 import { resolve, dirname } from "node:path";
 import { fileURLToPath } from "node:url";
 import { getConsoleLevel, LogLevel } from "../lib/logger.mjs";
 import { isBenignErrorMention } from "../utils.mjs";
-import { resolvePromptTemplate } from "./agent-prompts.mjs";
+import {
+  buildFallbackPrompt,
+  buildFixPrompt,
+  buildLoopPrompt,
+} from "./autofix-prompts.mjs";
+import { detectChangedFiles, getChangeSummary } from "./autofix-git.mjs";
 import { resolveCodexProfileRuntime } from "../shell/codex-model-profiles.mjs";
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
@@ -693,41 +698,6 @@ export function runCodexExec(
  * Detect which files were modified by comparing git status before/after.
  * Returns array of changed file paths.
  */
-function detectChangedFiles(repoRoot) {
-  try {
-    const output = execSync("git diff --name-only", {
-      cwd: repoRoot,
-      encoding: "utf8",
-      timeout: 10_000,
-      stdio: ["pipe", "pipe", "pipe"],
-    });
-    return output
-      .split(/\r?\n/)
-      .map((f) => f.trim())
-      .filter(Boolean);
-  } catch {
-    return [];
-  }
-}
-
-/**
- * Get git diff summary for changed files (short, for Telegram).
- */
-function getChangeSummary(repoRoot, files) {
-  if (!files.length) return "(no file changes detected)";
-  try {
-    const diff = execSync("git diff --stat", {
-      cwd: repoRoot,
-      encoding: "utf8",
-      timeout: 10_000,
-      stdio: ["pipe", "pipe", "pipe"],
-    });
-    return diff.trim() || files.join(", ");
-  } catch {
-    return files.join(", ");
-  }
-}
-
 /**
  * Attempt to auto-fix errors found in a crash log.
  *
@@ -1106,113 +1076,6 @@ export async function attemptAutoFix(opts) {
   };
 }
 
-// ── Prompt builders ─────────────────────────────────────────────────────────
-
-function buildRecentMessagesContext(recentMessages) {
-  if (!recentMessages || !recentMessages.length) return "";
-  const msgs = recentMessages.slice(-15);
-  return `
-## Recent monitor notifications (for context — shows what led to this crash)
-${msgs.map((m, i) => `[${i + 1}] ${m}`).join("\n")}
-`;
-}
-
-function buildFixPrompt(
-  error,
-  sourceContext,
-  reason,
-  recentMessages,
-  promptTemplate = "",
-) {
-  const messagesCtx = buildRecentMessagesContext(recentMessages);
-
-  const fallback = `You are a PowerShell expert fixing a crash in a running orchestrator script.
-
-## Error
-Type: ${error.errorType}
-File: ${error.file}
-Line: ${error.line}${error.column ? `\nColumn: ${error.column}` : ""}
-Message: ${error.message}${error.codeLine ? `\nFailing code: ${error.codeLine}` : ""}
-Crash reason: ${reason}
-
-## Source context around line ${error.line}
-\`\`\`powershell
-${sourceContext}
-\`\`\`
-${messagesCtx}
-## Instructions
-1. Read the file "${error.file}"
-2. Identify the root cause of the error at line ${error.line}
-3. Fix ONLY the bug — minimal change, don't refactor unrelated code
-4. Common PowerShell pitfalls:
-   - \`+=\` on arrays with single items fails — use [List[object]] or @() wrapping
-   - \`$a + $b\` on PSObjects fails — iterate and add individually
-   - Pipeline output can be a single object, not an array — always wrap with @()
-   - \`$null.Method()\` crashes — add null guards
-   - Named mutex with "Global\\\\" prefix fails on non-elevated Windows — use plain names
-   - \`$Var:\` is treated as a scope-qualified variable — use \`\${Var}:\` to embed colon in string
-   - ParserError: check for syntax issues like unclosed brackets, bad string interpolation
-5. Write the fix to the file. Do NOT create new files or refactor other functions.
-6. Keep all existing functionality intact.`;
-  return resolvePromptTemplate(
-    promptTemplate,
-    {
-      ERROR_TYPE: error.errorType,
-      ERROR_FILE: error.file,
-      ERROR_LINE: error.line,
-      ERROR_COLUMN_LINE: error.column ? `Column: ${error.column}` : "",
-      ERROR_MESSAGE: error.message,
-      ERROR_CODE_LINE: error.codeLine ? `Failing code: ${error.codeLine}` : "",
-      CRASH_REASON: reason,
-      SOURCE_CONTEXT: sourceContext,
-      RECENT_MESSAGES_CONTEXT: messagesCtx,
-    },
-    fallback,
-  );
-}
-
-function buildFallbackPrompt(fallback, recentMessages, promptTemplate = "") {
-  const messagesCtx = buildRecentMessagesContext(recentMessages);
-
-  const defaultPrompt = `You are a PowerShell expert analyzing an orchestrator script crash.
-No structured error was extracted — the process terminated with: ${fallback.reason}
-
-## Error indicators from log tail
-${fallback.errorLines.length > 0 ? fallback.errorLines.join("\n") : "(no explicit error lines detected — possible SIGKILL, OOM, or silent crash)"}
-
-## Last ${Math.min(80, fallback.lineCount)} lines of crash log
-\`\`\`
-${fallback.tail}
-\`\`\`
-${messagesCtx}
-## Instructions
-1. Analyze the log for the root cause of the crash
-2. The main orchestrator script is: scripts/bosun/ve-orchestrator.ps1
-3. If you can identify a fixable bug, apply a minimal fix to the file
-4. Common crash causes:
-   - PowerShell syntax errors (\$Var: treated as scope, missing brackets)
-   - Array/object operation errors (+=, +, pipeline single-item issues)
-   - Null reference errors on optional API responses
-   - Infinite loops or stack overflow from recursive calls
-   - Exit code 4294967295 = unsigned overflow from uncaught exception
-5. If the crash is external (SIGKILL, OOM) with no code bug, do nothing
-6. Write any fix directly to the file. Keep existing functionality intact.`;
-  return resolvePromptTemplate(
-    promptTemplate,
-    {
-      FALLBACK_REASON: fallback.reason,
-      FALLBACK_ERROR_LINES:
-        fallback.errorLines.length > 0
-          ? fallback.errorLines.join("\n")
-          : "(no explicit error lines detected — possible SIGKILL, OOM, or silent crash)",
-      FALLBACK_LINE_COUNT: Math.min(80, fallback.lineCount),
-      FALLBACK_TAIL: fallback.tail,
-      RECENT_MESSAGES_CONTEXT: messagesCtx,
-    },
-    defaultPrompt,
-  );
-}
-
 // ── Repeating error (loop) fixer ────────────────────────────────────────────
 
 /**
@@ -1265,38 +1128,11 @@ export async function fixLoopingError(opts) {
     );
   }
 
-  const messagesCtx = buildRecentMessagesContext(recentMessages);
-
-  const defaultPrompt = `You are a PowerShell expert fixing a loop bug in a running orchestrator script.
-
-## Problem
-The following error line is repeating ${repeatCount} times in the orchestrator output,
-indicating an infinite retry loop that needs to be fixed:
-
-"${errorLine}"
-
-${messagesCtx}
-
-## Instructions
-1. The main script is: scripts/bosun/ve-orchestrator.ps1
-2. Search for the code that produces this error message
-3. Identify why it loops (missing break/continue/return, no state change between iterations, etc.)
-4. Fix the loop by adding proper exit conditions, error handling, or state tracking
-5. Common loop-causing patterns in this codebase:
-   - PR lifecycle handoff repeatedly retried with no diff between branch and base
-   - API calls returning the same error repeatedly with no backoff or give-up logic
-   - Status not updated after failure → next cycle tries the same thing
-   - Missing \`continue\` or state change in foreach loops over tracked attempts
-6. Apply a minimal fix. Do NOT refactor unrelated code.
-7. Write the fix directly to the file.`;
-  const prompt = resolvePromptTemplate(
+  const prompt = buildLoopPrompt(
+    errorLine,
+    repeatCount,
+    recentMessages,
     promptTemplate,
-    {
-      REPEAT_COUNT: repeatCount,
-      ERROR_LINE: errorLine,
-      RECENT_MESSAGES_CONTEXT: messagesCtx,
-    },
-    defaultPrompt,
   );
 
   // Audit log

package/agent/bosun-skills.mjs

@@ -41,9 +41,9 @@ const _SKILL_STREAM_PATH = resolve(
  *
  * @param {string} skillName
  * @param {string} [skillTitle]
- * @param {{ taskId?: string, executor?: string }} [opts]
+ * @param {{ taskId?: string, executor?: string, source?: string }} [opts]
  */
-function emitSkillInvokeEvent(skillName, skillTitle, opts = {}) {
+export function emitSkillInvokeEvent(skillName, skillTitle, opts = {}) {
   try {
     const event = {
       timestamp: new Date().toISOString(),
@@ -51,6 +51,7 @@ function emitSkillInvokeEvent(skillName, skillTitle, opts = {}) {
       data: { skill_name: skillName, skill_title: skillTitle || skillName },
       ...(opts.taskId ? { task_id: String(opts.taskId) } : {}),
       ...(opts.executor ? { executor: String(opts.executor) } : {}),
+      ...(opts.source ? { source: String(opts.source) } : {}),
     };
     mkdirSync(dirname(_SKILL_STREAM_PATH), { recursive: true });
     appendFileSync(_SKILL_STREAM_PATH, JSON.stringify(event) + "\n", "utf8");

package/bosun.config.example.json

@@ -5,6 +5,23 @@
 
   "_comment_workflowFirst": "Set to true to run everything as workflows (task lifecycle, PR management, etc.)",
   "workflowFirst": false,
+  "_comment_worktreeRecoveryCooldown": "Minutes to keep a task blocked after a non-retryable worktree failure before Bosun automatically returns it to todo.",
+  "workflowWorktreeRecoveryCooldownMin": 15,
+  "_comment_worktreeBootstrap": "Optional managed-worktree bootstrap policy. Commands are per detected stack and can be overridden per repo.",
+  "worktreeBootstrap": {
+    "enabled": true,
+    "linkSharedPaths": true,
+    "commandTimeoutMs": 600000,
+    "commandsByStack": {
+      "node": ["npm install"],
+      "python": ["poetry install --no-interaction"],
+      "dotnet": ["dotnet restore"]
+    },
+    "sharedPathsByStack": {
+      "node": ["node_modules"],
+      "php": ["vendor"]
+    }
+  },
   "workflows": [
     {
       "type": "continuation-loop",