opencastle 0.27.0 → 0.27.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (242) hide show
  1. package/bin/cli.mjs +6 -0
  2. package/dist/cli/agents.d.ts +3 -0
  3. package/dist/cli/agents.d.ts.map +1 -0
  4. package/dist/cli/agents.js +161 -0
  5. package/dist/cli/agents.js.map +1 -0
  6. package/dist/cli/baselines.d.ts +3 -0
  7. package/dist/cli/baselines.d.ts.map +1 -0
  8. package/dist/cli/baselines.js +128 -0
  9. package/dist/cli/baselines.js.map +1 -0
  10. package/dist/cli/convoy/dashboard-types.d.ts +146 -0
  11. package/dist/cli/convoy/dashboard-types.d.ts.map +1 -0
  12. package/dist/cli/convoy/dashboard-types.js +2 -0
  13. package/dist/cli/convoy/dashboard-types.js.map +1 -0
  14. package/dist/cli/convoy/engine.d.ts +67 -2
  15. package/dist/cli/convoy/engine.d.ts.map +1 -1
  16. package/dist/cli/convoy/engine.js +2036 -28
  17. package/dist/cli/convoy/engine.js.map +1 -1
  18. package/dist/cli/convoy/engine.test.js +1659 -70
  19. package/dist/cli/convoy/engine.test.js.map +1 -1
  20. package/dist/cli/convoy/event-schemas.d.ts +9 -0
  21. package/dist/cli/convoy/event-schemas.d.ts.map +1 -0
  22. package/dist/cli/convoy/event-schemas.js +185 -0
  23. package/dist/cli/convoy/event-schemas.js.map +1 -0
  24. package/dist/cli/convoy/events.d.ts +12 -1
  25. package/dist/cli/convoy/events.d.ts.map +1 -1
  26. package/dist/cli/convoy/events.js +186 -13
  27. package/dist/cli/convoy/events.js.map +1 -1
  28. package/dist/cli/convoy/events.test.js +325 -28
  29. package/dist/cli/convoy/events.test.js.map +1 -1
  30. package/dist/cli/convoy/expertise.d.ts +16 -0
  31. package/dist/cli/convoy/expertise.d.ts.map +1 -0
  32. package/dist/cli/convoy/expertise.js +121 -0
  33. package/dist/cli/convoy/expertise.js.map +1 -0
  34. package/dist/cli/convoy/expertise.test.d.ts +2 -0
  35. package/dist/cli/convoy/expertise.test.d.ts.map +1 -0
  36. package/dist/cli/convoy/expertise.test.js +96 -0
  37. package/dist/cli/convoy/expertise.test.js.map +1 -0
  38. package/dist/cli/convoy/export.test.js +1 -0
  39. package/dist/cli/convoy/export.test.js.map +1 -1
  40. package/dist/cli/convoy/formula.d.ts +19 -0
  41. package/dist/cli/convoy/formula.d.ts.map +1 -0
  42. package/dist/cli/convoy/formula.js +142 -0
  43. package/dist/cli/convoy/formula.js.map +1 -0
  44. package/dist/cli/convoy/formula.test.d.ts +2 -0
  45. package/dist/cli/convoy/formula.test.d.ts.map +1 -0
  46. package/dist/cli/convoy/formula.test.js +342 -0
  47. package/dist/cli/convoy/formula.test.js.map +1 -0
  48. package/dist/cli/convoy/gates.d.ts +128 -0
  49. package/dist/cli/convoy/gates.d.ts.map +1 -0
  50. package/dist/cli/convoy/gates.js +606 -0
  51. package/dist/cli/convoy/gates.js.map +1 -0
  52. package/dist/cli/convoy/gates.test.d.ts +2 -0
  53. package/dist/cli/convoy/gates.test.d.ts.map +1 -0
  54. package/dist/cli/convoy/gates.test.js +976 -0
  55. package/dist/cli/convoy/gates.test.js.map +1 -0
  56. package/dist/cli/convoy/health.d.ts +11 -0
  57. package/dist/cli/convoy/health.d.ts.map +1 -1
  58. package/dist/cli/convoy/health.js +54 -0
  59. package/dist/cli/convoy/health.js.map +1 -1
  60. package/dist/cli/convoy/health.test.js +56 -1
  61. package/dist/cli/convoy/health.test.js.map +1 -1
  62. package/dist/cli/convoy/issues.d.ts +8 -0
  63. package/dist/cli/convoy/issues.d.ts.map +1 -0
  64. package/dist/cli/convoy/issues.js +98 -0
  65. package/dist/cli/convoy/issues.js.map +1 -0
  66. package/dist/cli/convoy/issues.test.d.ts +2 -0
  67. package/dist/cli/convoy/issues.test.d.ts.map +1 -0
  68. package/dist/cli/convoy/issues.test.js +107 -0
  69. package/dist/cli/convoy/issues.test.js.map +1 -0
  70. package/dist/cli/convoy/knowledge.d.ts +5 -0
  71. package/dist/cli/convoy/knowledge.d.ts.map +1 -0
  72. package/dist/cli/convoy/knowledge.js +116 -0
  73. package/dist/cli/convoy/knowledge.js.map +1 -0
  74. package/dist/cli/convoy/knowledge.test.d.ts +2 -0
  75. package/dist/cli/convoy/knowledge.test.d.ts.map +1 -0
  76. package/dist/cli/convoy/knowledge.test.js +87 -0
  77. package/dist/cli/convoy/knowledge.test.js.map +1 -0
  78. package/dist/cli/convoy/lessons.d.ts +17 -0
  79. package/dist/cli/convoy/lessons.d.ts.map +1 -0
  80. package/dist/cli/convoy/lessons.js +149 -0
  81. package/dist/cli/convoy/lessons.js.map +1 -0
  82. package/dist/cli/convoy/lessons.test.d.ts +2 -0
  83. package/dist/cli/convoy/lessons.test.d.ts.map +1 -0
  84. package/dist/cli/convoy/lessons.test.js +135 -0
  85. package/dist/cli/convoy/lessons.test.js.map +1 -0
  86. package/dist/cli/convoy/lock.d.ts +13 -0
  87. package/dist/cli/convoy/lock.d.ts.map +1 -0
  88. package/dist/cli/convoy/lock.js +88 -0
  89. package/dist/cli/convoy/lock.js.map +1 -0
  90. package/dist/cli/convoy/lock.test.d.ts +2 -0
  91. package/dist/cli/convoy/lock.test.d.ts.map +1 -0
  92. package/dist/cli/convoy/lock.test.js +136 -0
  93. package/dist/cli/convoy/lock.test.js.map +1 -0
  94. package/dist/cli/convoy/log-merge.test.d.ts +2 -0
  95. package/dist/cli/convoy/log-merge.test.d.ts.map +1 -0
  96. package/dist/cli/convoy/log-merge.test.js +147 -0
  97. package/dist/cli/convoy/log-merge.test.js.map +1 -0
  98. package/dist/cli/convoy/merge.d.ts +4 -0
  99. package/dist/cli/convoy/merge.d.ts.map +1 -1
  100. package/dist/cli/convoy/merge.js +18 -1
  101. package/dist/cli/convoy/merge.js.map +1 -1
  102. package/dist/cli/convoy/merge.test.js +6 -7
  103. package/dist/cli/convoy/merge.test.js.map +1 -1
  104. package/dist/cli/convoy/partition.d.ts +51 -0
  105. package/dist/cli/convoy/partition.d.ts.map +1 -0
  106. package/dist/cli/convoy/partition.js +186 -0
  107. package/dist/cli/convoy/partition.js.map +1 -0
  108. package/dist/cli/convoy/partition.test.d.ts +2 -0
  109. package/dist/cli/convoy/partition.test.d.ts.map +1 -0
  110. package/dist/cli/convoy/partition.test.js +315 -0
  111. package/dist/cli/convoy/partition.test.js.map +1 -0
  112. package/dist/cli/convoy/pipeline.test.js +6 -0
  113. package/dist/cli/convoy/pipeline.test.js.map +1 -1
  114. package/dist/cli/convoy/store.d.ts +99 -7
  115. package/dist/cli/convoy/store.d.ts.map +1 -1
  116. package/dist/cli/convoy/store.js +764 -31
  117. package/dist/cli/convoy/store.js.map +1 -1
  118. package/dist/cli/convoy/store.test.js +1810 -18
  119. package/dist/cli/convoy/store.test.js.map +1 -1
  120. package/dist/cli/convoy/types.d.ts +427 -5
  121. package/dist/cli/convoy/types.d.ts.map +1 -1
  122. package/dist/cli/convoy/types.js +42 -1
  123. package/dist/cli/convoy/types.js.map +1 -1
  124. package/dist/cli/log.d.ts +11 -0
  125. package/dist/cli/log.d.ts.map +1 -1
  126. package/dist/cli/log.js +114 -2
  127. package/dist/cli/log.js.map +1 -1
  128. package/dist/cli/run/adapters/claude.d.ts +2 -0
  129. package/dist/cli/run/adapters/claude.d.ts.map +1 -1
  130. package/dist/cli/run/adapters/claude.js +89 -49
  131. package/dist/cli/run/adapters/claude.js.map +1 -1
  132. package/dist/cli/run/adapters/claude.test.d.ts +2 -0
  133. package/dist/cli/run/adapters/claude.test.d.ts.map +1 -0
  134. package/dist/cli/run/adapters/claude.test.js +205 -0
  135. package/dist/cli/run/adapters/claude.test.js.map +1 -0
  136. package/dist/cli/run/adapters/copilot.d.ts +1 -0
  137. package/dist/cli/run/adapters/copilot.d.ts.map +1 -1
  138. package/dist/cli/run/adapters/copilot.js +84 -46
  139. package/dist/cli/run/adapters/copilot.js.map +1 -1
  140. package/dist/cli/run/adapters/copilot.test.d.ts +2 -0
  141. package/dist/cli/run/adapters/copilot.test.d.ts.map +1 -0
  142. package/dist/cli/run/adapters/copilot.test.js +195 -0
  143. package/dist/cli/run/adapters/copilot.test.js.map +1 -0
  144. package/dist/cli/run/adapters/cursor.d.ts +1 -0
  145. package/dist/cli/run/adapters/cursor.d.ts.map +1 -1
  146. package/dist/cli/run/adapters/cursor.js +83 -47
  147. package/dist/cli/run/adapters/cursor.js.map +1 -1
  148. package/dist/cli/run/adapters/cursor.test.d.ts +2 -0
  149. package/dist/cli/run/adapters/cursor.test.d.ts.map +1 -0
  150. package/dist/cli/run/adapters/cursor.test.js +129 -0
  151. package/dist/cli/run/adapters/cursor.test.js.map +1 -0
  152. package/dist/cli/run/adapters/opencode.d.ts +1 -0
  153. package/dist/cli/run/adapters/opencode.d.ts.map +1 -1
  154. package/dist/cli/run/adapters/opencode.js +81 -47
  155. package/dist/cli/run/adapters/opencode.js.map +1 -1
  156. package/dist/cli/run/adapters/opencode.test.d.ts +2 -0
  157. package/dist/cli/run/adapters/opencode.test.d.ts.map +1 -0
  158. package/dist/cli/run/adapters/opencode.test.js +119 -0
  159. package/dist/cli/run/adapters/opencode.test.js.map +1 -0
  160. package/dist/cli/run/executor.js +1 -1
  161. package/dist/cli/run/executor.js.map +1 -1
  162. package/dist/cli/run/schema.d.ts.map +1 -1
  163. package/dist/cli/run/schema.js +245 -4
  164. package/dist/cli/run/schema.js.map +1 -1
  165. package/dist/cli/run/schema.test.js +669 -0
  166. package/dist/cli/run/schema.test.js.map +1 -1
  167. package/dist/cli/run.d.ts.map +1 -1
  168. package/dist/cli/run.js +362 -22
  169. package/dist/cli/run.js.map +1 -1
  170. package/dist/cli/types.d.ts +85 -2
  171. package/dist/cli/types.d.ts.map +1 -1
  172. package/dist/cli/types.js.map +1 -1
  173. package/dist/cli/watch.d.ts +15 -0
  174. package/dist/cli/watch.d.ts.map +1 -0
  175. package/dist/cli/watch.js +279 -0
  176. package/dist/cli/watch.js.map +1 -0
  177. package/package.json +5 -1
  178. package/src/cli/agents.ts +177 -0
  179. package/src/cli/baselines.ts +143 -0
  180. package/src/cli/convoy/TELEMETRY.md +203 -0
  181. package/src/cli/convoy/dashboard-types.ts +141 -0
  182. package/src/cli/convoy/engine.test.ts +1937 -70
  183. package/src/cli/convoy/engine.ts +2350 -40
  184. package/src/cli/convoy/event-schemas.ts +195 -0
  185. package/src/cli/convoy/events.test.ts +384 -39
  186. package/src/cli/convoy/events.ts +202 -16
  187. package/src/cli/convoy/expertise.test.ts +128 -0
  188. package/src/cli/convoy/expertise.ts +163 -0
  189. package/src/cli/convoy/export.test.ts +1 -0
  190. package/src/cli/convoy/formula.test.ts +405 -0
  191. package/src/cli/convoy/formula.ts +174 -0
  192. package/src/cli/convoy/gates.test.ts +1169 -0
  193. package/src/cli/convoy/gates.ts +774 -0
  194. package/src/cli/convoy/health.test.ts +64 -2
  195. package/src/cli/convoy/health.ts +80 -2
  196. package/src/cli/convoy/issues.test.ts +143 -0
  197. package/src/cli/convoy/issues.ts +136 -0
  198. package/src/cli/convoy/knowledge.test.ts +101 -0
  199. package/src/cli/convoy/knowledge.ts +132 -0
  200. package/src/cli/convoy/lessons.test.ts +188 -0
  201. package/src/cli/convoy/lessons.ts +164 -0
  202. package/src/cli/convoy/lock.test.ts +181 -0
  203. package/src/cli/convoy/lock.ts +103 -0
  204. package/src/cli/convoy/log-merge.test.ts +179 -0
  205. package/src/cli/convoy/merge.test.ts +6 -7
  206. package/src/cli/convoy/merge.ts +19 -1
  207. package/src/cli/convoy/partition.test.ts +423 -0
  208. package/src/cli/convoy/partition.ts +232 -0
  209. package/src/cli/convoy/pipeline.test.ts +6 -0
  210. package/src/cli/convoy/store.test.ts +2041 -20
  211. package/src/cli/convoy/store.ts +945 -46
  212. package/src/cli/convoy/types.ts +278 -4
  213. package/src/cli/log.ts +120 -2
  214. package/src/cli/run/adapters/claude.test.ts +234 -0
  215. package/src/cli/run/adapters/claude.ts +45 -5
  216. package/src/cli/run/adapters/copilot.test.ts +224 -0
  217. package/src/cli/run/adapters/copilot.ts +34 -4
  218. package/src/cli/run/adapters/cursor.test.ts +144 -0
  219. package/src/cli/run/adapters/cursor.ts +33 -2
  220. package/src/cli/run/adapters/opencode.test.ts +135 -0
  221. package/src/cli/run/adapters/opencode.ts +30 -2
  222. package/src/cli/run/executor.ts +1 -1
  223. package/src/cli/run/schema.test.ts +758 -0
  224. package/src/cli/run/schema.ts +300 -25
  225. package/src/cli/run.ts +341 -21
  226. package/src/cli/types.ts +86 -1
  227. package/src/cli/watch.ts +298 -0
  228. package/src/dashboard/dist/_astro/{index.DtnyD8a5.css → index.6L3_HsPT.css} +1 -1
  229. package/src/dashboard/dist/data/.gitkeep +0 -0
  230. package/src/dashboard/dist/data/convoy-list.json +1 -0
  231. package/src/dashboard/dist/data/overall-stats.json +24 -0
  232. package/src/dashboard/dist/index.html +701 -3
  233. package/src/dashboard/node_modules/.vite/deps/_metadata.json +6 -6
  234. package/src/dashboard/public/data/.gitkeep +0 -0
  235. package/src/dashboard/public/data/convoy-list.json +1 -0
  236. package/src/dashboard/public/data/overall-stats.json +24 -0
  237. package/src/dashboard/scripts/etl.test.ts +210 -0
  238. package/src/dashboard/scripts/etl.ts +108 -0
  239. package/src/dashboard/scripts/integration-test.ts +504 -0
  240. package/src/dashboard/src/pages/index.astro +854 -15
  241. package/src/dashboard/src/styles/dashboard.css +557 -1
  242. package/src/orchestrator/prompts/generate-convoy.prompt.md +212 -13
@@ -1,20 +1,248 @@
1
1
  import { execFile as execFileCb } from 'node:child_process';
2
2
  import { createHash } from 'node:crypto';
3
- import { mkdirSync } from 'node:fs';
3
+ import { appendFileSync, existsSync, mkdirSync, readFileSync, renameSync, unlinkSync, } from 'node:fs';
4
4
  import { dirname, join, resolve } from 'node:path';
5
+ import { DatabaseSync } from 'node:sqlite';
5
6
  import { promisify } from 'node:util';
6
- import { createConvoyStore } from './store.js';
7
- import { createEventEmitter } from './events.js';
7
+ import { createConvoyStore, ConvoyArtifactLimitError } from './store.js';
8
+ import { acquireEngineLock } from './lock.js';
9
+ import { createEventEmitter, ndjsonPathForConvoy, recoverNdjson } from './events.js';
8
10
  import { createWorktreeManager } from './worktree.js';
9
- import { createMergeQueue } from './merge.js';
10
- import { createHealthMonitor } from './health.js';
11
+ import { createMergeQueue, MergeConflictError } from './merge.js';
12
+ import { createHealthMonitor, detectDrift } from './health.js';
11
13
  import { exportConvoyToNdjson } from './export.js';
12
14
  import { buildPhases, formatDuration } from '../run/executor.js';
13
- import { parseTimeout } from '../run/schema.js';
15
+ import { parseTimeout, parseYaml } from '../run/schema.js';
14
16
  import { getAdapter, detectAdapter } from '../run/adapters/index.js';
15
17
  import { c } from '../prompt.js';
18
+ import { validateFilePartitions, scanSymlinks, scanNewSymlinks, normalizePath, pathsOverlap } from './partition.js';
19
+ import { scanForSecrets, runSecretScanGate, runBlastRadiusGate, browserTestGate } from './gates.js';
20
+ import { readLessons, captureLessons, consolidateLessons } from './lessons.js';
21
+ import { updateExpertise, feedCircuitBreaker } from './expertise.js';
22
+ import { buildKnowledgeGraph } from './knowledge.js';
23
+ import { injectDiscoveredIssuesInstruction, checkDiscoveredIssues, consolidateIssues } from './issues.js';
16
24
  const execFile = promisify(execFileCb);
17
- // ── Internal helpers ──────────────────────────────────────────────────────────
25
+ export class CircuitBreakerManager {
26
+ states = new Map();
27
+ threshold;
28
+ cooldownMs;
29
+ fallbackAgent;
30
+ constructor(config, initialState) {
31
+ this.threshold = config?.threshold ?? 3;
32
+ this.cooldownMs = config?.cooldown_ms ?? 300_000;
33
+ this.fallbackAgent = config?.fallback_agent ?? null;
34
+ if (initialState) {
35
+ for (const [agent, state] of Object.entries(initialState)) {
36
+ this.states.set(agent, state);
37
+ }
38
+ }
39
+ }
40
+ getState(agent) {
41
+ return this.states.get(agent) ?? { status: 'closed', failures: 0, last_failure_at: null, opened_at: null };
42
+ }
43
+ recordFailure(agent) {
44
+ const state = this.getState(agent);
45
+ const now = new Date().toISOString();
46
+ if (state.status === 'half-open') {
47
+ // Probe failed — back to open, reset cooldown
48
+ state.status = 'open';
49
+ state.opened_at = now;
50
+ state.last_failure_at = now;
51
+ this.states.set(agent, state);
52
+ return { tripped: true, state };
53
+ }
54
+ state.failures += 1;
55
+ state.last_failure_at = now;
56
+ if (state.failures >= this.threshold) {
57
+ state.status = 'open';
58
+ state.opened_at = now;
59
+ this.states.set(agent, state);
60
+ return { tripped: true, state };
61
+ }
62
+ this.states.set(agent, state);
63
+ return { tripped: false, state };
64
+ }
65
+ recordSuccess(agent) {
66
+ const state = this.getState(agent);
67
+ if (state.status === 'half-open') {
68
+ // Probe succeeded — close circuit
69
+ state.status = 'closed';
70
+ state.failures = 0;
71
+ state.opened_at = null;
72
+ }
73
+ else if (state.status === 'closed') {
74
+ state.failures = 0;
75
+ }
76
+ this.states.set(agent, state);
77
+ return state;
78
+ }
79
+ canAssign(agent) {
80
+ const state = this.getState(agent);
81
+ if (state.status === 'closed')
82
+ return true;
83
+ if (state.status === 'half-open')
84
+ return true; // allow 1 probe
85
+ // Open — check cooldown
86
+ if (state.opened_at) {
87
+ const elapsed = Date.now() - new Date(state.opened_at).getTime();
88
+ if (elapsed >= this.cooldownMs) {
89
+ state.status = 'half-open';
90
+ this.states.set(agent, state);
91
+ return true;
92
+ }
93
+ }
94
+ return false;
95
+ }
96
+ get fallback() {
97
+ return this.fallbackAgent;
98
+ }
99
+ serialize() {
100
+ return JSON.stringify(Object.fromEntries(this.states));
101
+ }
102
+ }
103
+ // ── Branch management ───────────────────────────────────────────────────────
104
+ /**
105
+ * Ensure the given branch exists and is checked out.
106
+ * Creates the branch from HEAD if it does not yet exist.
107
+ * Fails fast if there are uncommitted changes.
108
+ */
109
+ export async function ensureBranch(branchName, basePath) {
110
+ // Validate refspec — reject shell metacharacters
111
+ if (!/^[a-zA-Z0-9\-/_\.]+$/.test(branchName)) {
112
+ throw new Error(`Invalid branch name "${branchName}": only alphanumeric, -, /, _, and . are allowed`);
113
+ }
114
+ // Refuse to switch branches with uncommitted changes
115
+ const { stdout: statusOut } = await execFile('git', ['status', '--porcelain'], {
116
+ cwd: basePath,
117
+ });
118
+ if (statusOut.trim()) {
119
+ throw new Error(`Uncommitted changes detected in "${basePath}". Commit or stash before switching branches.`);
120
+ }
121
+ // Check if branch already exists
122
+ try {
123
+ await execFile('git', ['rev-parse', '--verify', branchName], { cwd: basePath });
124
+ // Branch exists — check it out
125
+ await execFile('git', ['checkout', branchName], { cwd: basePath });
126
+ }
127
+ catch {
128
+ // Branch does not exist — create from current HEAD
129
+ await execFile('git', ['checkout', '-b', branchName], { cwd: basePath });
130
+ }
131
+ }
132
+ export function runConvoyGuard(store, convoyId, _wtManager, ndjsonPath, guardConfig) {
133
+ // If guard is explicitly disabled, skip all checks
134
+ if (guardConfig?.enabled === false) {
135
+ return { passed: true, warnings: [] };
136
+ }
137
+ const warnings = [];
138
+ const tasks = store.getTasksByConvoy(convoyId);
139
+ // Check 1: All task statuses are terminal
140
+ const terminalStatuses = new Set(['done', 'failed', 'skipped', 'timed-out', 'gate-failed', 'review-blocked', 'hook-failed', 'disputed']);
141
+ const nonTerminal = tasks.filter(t => !terminalStatuses.has(t.status));
142
+ if (nonTerminal.length > 0) {
143
+ warnings.push(`Non-terminal tasks: ${nonTerminal.map(t => `${t.id}(${t.status})`).join(', ')}`);
144
+ }
145
+ // Check 2: NDJSON file exists and record count >= completed task count
146
+ const completedTasks = tasks.filter(t => t.status === 'done');
147
+ try {
148
+ const content = readFileSync(ndjsonPath, 'utf8');
149
+ const lines = content.split('\n').filter(l => l.trim());
150
+ // Per-convoy file — all records belong to this convoy, no need to filter by convoy_id
151
+ if (lines.length < completedTasks.length) {
152
+ warnings.push(`NDJSON record count (${lines.length}) < completed tasks (${completedTasks.length})`);
153
+ }
154
+ }
155
+ catch {
156
+ if (completedTasks.length > 0) {
157
+ warnings.push(`NDJSON file not found at ${ndjsonPath} but ${completedTasks.length} tasks completed`);
158
+ }
159
+ }
160
+ // Check 3: Every retried task has events for each attempt
161
+ const retriedTasks = tasks.filter(t => t.retries > 0);
162
+ const events = store.getEvents(convoyId);
163
+ for (const task of retriedTasks) {
164
+ const taskEvents = events.filter(e => e.task_id === task.id && e.type === 'task_started');
165
+ if (taskEvents.length < task.retries) {
166
+ warnings.push(`Task ${task.id} has ${task.retries} retries but only ${taskEvents.length} task_started events`);
167
+ }
168
+ }
169
+ // Check 4: Gate results recorded for all gates that ran
170
+ const gateEvents = events.filter(e => {
171
+ if (e.type === 'built_in_gate_result')
172
+ return true;
173
+ if (e.data == null)
174
+ return false;
175
+ try {
176
+ const parsed = JSON.parse(e.data);
177
+ return 'gate' in parsed;
178
+ }
179
+ catch {
180
+ return false;
181
+ }
182
+ });
183
+ const tasksWithGates = tasks.filter(t => t.gates);
184
+ if (tasksWithGates.length > 0 && gateEvents.length === 0) {
185
+ warnings.push('Tasks have gates configured but no gate result events found');
186
+ }
187
+ // Check 5: Token/cost totals computed
188
+ const convoy = store.getConvoy(convoyId);
189
+ if (convoy && convoy.total_tokens == null) {
190
+ const totalTokens = tasks.reduce((sum, t) => sum + (t.total_tokens ?? 0), 0);
191
+ if (totalTokens > 0) {
192
+ warnings.push('Convoy total_tokens not persisted despite tasks having token data');
193
+ }
194
+ }
195
+ // Check 6: No orphaned worktrees — engine already calls removeAll() during cleanup.
196
+ // Synchronous check is not possible; the engine handles this.
197
+ return { passed: warnings.length === 0, warnings };
198
+ }
199
+ export function evaluateReviewLevel(task, diff, heuristics, allGatesPassed) {
200
+ const panelPaths = heuristics?.panel_paths ?? ['auth/', 'security/', 'migrations/', 'rls/'];
201
+ const panelAgents = heuristics?.panel_agents ?? ['security-expert', 'database-engineer'];
202
+ const autoPassAgents = heuristics?.auto_pass_agents ?? ['documentation-writer', 'copywriter'];
203
+ const autoPassMaxLines = heuristics?.auto_pass_max_lines ?? 10;
204
+ const autoPassMaxFiles = heuristics?.auto_pass_max_files ?? 2;
205
+ // Panel: sensitive paths or agents
206
+ if (panelPaths.some(p => diff.filePaths.some(fp => fp.startsWith(p) || fp.includes('/' + p))))
207
+ return 'panel';
208
+ if (panelAgents.includes(task.agent))
209
+ return 'panel';
210
+ // Auto-pass: documentation/copy agents
211
+ if (autoPassAgents.includes(task.agent))
212
+ return 'auto-pass';
213
+ // Auto-pass: small diffs with all gates passing
214
+ if (diff.linesChanged <= autoPassMaxLines && diff.filesChanged <= autoPassMaxFiles && allGatesPassed !== false)
215
+ return 'auto-pass';
216
+ // Large diffs → fast review
217
+ if (diff.linesChanged > 200 || diff.filesChanged > 5)
218
+ return 'fast';
219
+ // Default → fast review
220
+ return 'fast';
221
+ }
222
+ class ReviewSemaphore {
223
+ max;
224
+ current = 0;
225
+ queue = [];
226
+ constructor(max) {
227
+ this.max = max;
228
+ }
229
+ async acquire() {
230
+ if (this.current < this.max) {
231
+ this.current++;
232
+ return;
233
+ }
234
+ return new Promise(resolve => {
235
+ this.queue.push(() => { this.current++; resolve(); });
236
+ });
237
+ }
238
+ release() {
239
+ this.current--;
240
+ if (this.queue.length > 0) {
241
+ const next = this.queue.shift();
242
+ next();
243
+ }
244
+ }
245
+ }
18
246
  function msToTimeout(ms) {
19
247
  if (ms >= 3_600_000 && ms % 3_600_000 === 0)
20
248
  return `${ms / 3_600_000}h`;
@@ -22,6 +250,58 @@ function msToTimeout(ms) {
22
250
  return `${ms / 60_000}m`;
23
251
  return `${ms / 1_000}s`;
24
252
  }
253
+ // ── DLQ markdown dual-write ───────────────────────────────────────────────────
254
+ // Builds the DLQ markdown entry text (no I/O, no scanning).
255
+ function buildDlqMarkdownEntry(dlqId, task, failureType, errorOutput) {
256
+ const marker = `<!-- dlq:${dlqId} -->`;
257
+ const entry = `\n${marker}\n### ${dlqId}\n\n| Field | Value |\n|-------|-------|\n| Task | ${task.id} |\n| Agent | ${task.agent} |\n| Type | ${failureType} |\n| Attempts | ${task.retries + 1} |\n| Date | ${new Date().toISOString()} |\n\n**Error:**\n\`\`\`\n${(errorOutput ?? '(no output)').slice(0, 2000)}\n\`\`\`\n`;
258
+ return { marker, entry };
259
+ }
260
+ // Appends a pre-scanned DLQ entry to AGENT-FAILURES.md. The caller must have
261
+ // already verified the entry is clean via scanForSecrets — no re-scan here.
262
+ function appendDlqMarkdownClean(marker, entry) {
263
+ const mdPath = join(resolve(process.cwd()), '.opencastle', 'AGENT-FAILURES.md');
264
+ try {
265
+ const existing = readFileSync(mdPath, 'utf8');
266
+ if (existing.includes(marker))
267
+ return;
268
+ }
269
+ catch {
270
+ // File doesn't exist yet — will create
271
+ }
272
+ mkdirSync(dirname(mdPath), { recursive: true });
273
+ appendFileSync(mdPath, entry);
274
+ }
275
+ function writeDisputeToMarkdown(disputeId, convoyId, task, panelResults, events) {
276
+ const mdPath = join(resolve(process.cwd()), 'DISPUTES.md');
277
+ const marker = `<!-- dispute:${disputeId} -->`;
278
+ try {
279
+ const existing = readFileSync(mdPath, 'utf8');
280
+ if (existing.includes(marker))
281
+ return;
282
+ }
283
+ catch {
284
+ // File doesn't exist yet
285
+ }
286
+ const blockingReasons = panelResults
287
+ .filter(r => r.verdict === 'block')
288
+ .map(r => r.feedback)
289
+ .join('\n\n');
290
+ const entry = `\n${marker}\n## Dispute: ${task.id}\n\n| Field | Value |\n|-------|-------|\n| Convoy | ${convoyId} |\n| Task | ${task.id} |\n| Date | ${new Date().toISOString()} |\n| Panel attempts | ${task.panel_attempts + 1} |\n| Agent | ${task.agent} |\n| Status | Open |\n\n**Blocking reasons:**\n\n${blockingReasons}\n`;
291
+ const scanResult = scanForSecrets(entry, 'DISPUTES.md');
292
+ if (!scanResult.clean) {
293
+ if (events) {
294
+ events.emit('secret_leak_prevented', {
295
+ task_id: task.id,
296
+ findings_count: scanResult.findings.length,
297
+ patterns: scanResult.findings.map((f) => f.pattern),
298
+ context: 'dispute_markdown_write',
299
+ }, { convoy_id: convoyId, task_id: task.id });
300
+ }
301
+ return;
302
+ }
303
+ appendFileSync(mdPath, entry);
304
+ }
25
305
  function taskRecordToTask(record) {
26
306
  return {
27
307
  id: record.id,
@@ -34,6 +314,7 @@ function taskRecordToTask(record) {
34
314
  model: record.model ?? undefined,
35
315
  max_retries: record.max_retries,
36
316
  adapter: record.adapter ?? undefined,
317
+ gates: record.gates ? JSON.parse(record.gates) : undefined,
37
318
  };
38
319
  }
39
320
  function makeTimeoutPromise(ms) {
@@ -44,11 +325,322 @@ function makeTimeoutPromise(ms) {
44
325
  return { promise, clear: () => { if (timerId !== undefined)
45
326
  clearTimeout(timerId); } };
46
327
  }
328
+ // ── Step condition evaluation ─────────────────────────────────────────────────
329
+ function evaluateStepCondition(condition, stepResults, worktreePath, basePath) {
330
+ if (!condition)
331
+ return true;
332
+ if (condition.exitCode) {
333
+ const prevResult = stepResults.get(condition.step);
334
+ if (!prevResult)
335
+ return false;
336
+ const code = prevResult.exitCode;
337
+ const ec = condition.exitCode;
338
+ if (ec.eq !== undefined && code !== ec.eq)
339
+ return false;
340
+ if (ec.ne !== undefined && code === ec.ne)
341
+ return false;
342
+ if (ec.gt !== undefined && !(code > ec.gt))
343
+ return false;
344
+ if (ec.lt !== undefined && !(code < ec.lt))
345
+ return false;
346
+ }
347
+ if (condition.fileExists) {
348
+ const base = worktreePath ?? basePath;
349
+ if (condition.fileExists.path.startsWith('/')) {
350
+ return false; // Absolute paths not allowed in step conditions
351
+ }
352
+ const filePath = join(base, condition.fileExists.path);
353
+ const resolved = resolve(filePath);
354
+ const resolvedBase = resolve(base);
355
+ if (!resolved.startsWith(resolvedBase + '/') && resolved !== resolvedBase) {
356
+ return false; // path escapes the worktree — treat as "file doesn't exist"
357
+ }
358
+ if (!existsSync(filePath))
359
+ return false;
360
+ }
361
+ return true;
362
+ }
363
+ async function executeSteps(taskRecord, steps, adapter, worktreePath, basePath, store, convoyId, verbose) {
364
+ const now = () => new Date().toISOString();
365
+ const stepResults = new Map();
366
+ let combinedOutput = '';
367
+ let lastExitCode = 0;
368
+ // Track total_steps in DB
369
+ store.updateTaskStatus(taskRecord.id, convoyId, 'running', {});
370
+ for (let i = 0; i < steps.length; i++) {
371
+ const step = steps[i];
372
+ // Evaluate condition — skip step if condition is not met
373
+ if (step.if) {
374
+ const condMet = evaluateStepCondition(step.if, stepResults, worktreePath, basePath);
375
+ if (!condMet) {
376
+ const stepId = store.insertTaskStep({
377
+ task_id: taskRecord.id,
378
+ step_index: i,
379
+ prompt: step.prompt,
380
+ gates: step.gates ? JSON.stringify(step.gates) : null,
381
+ status: 'skipped',
382
+ exit_code: null,
383
+ output: 'Skipped: condition not met',
384
+ started_at: now(),
385
+ finished_at: now(),
386
+ });
387
+ if (step.id) {
388
+ stepResults.set(step.id, { exitCode: 0 });
389
+ }
390
+ combinedOutput += `\n[Step ${i + 1} skipped: condition not met]`;
391
+ continue;
392
+ }
393
+ }
394
+ // Insert step record as running
395
+ const stepDbId = store.insertTaskStep({
396
+ task_id: taskRecord.id,
397
+ step_index: i,
398
+ prompt: step.prompt,
399
+ gates: step.gates ? JSON.stringify(step.gates) : null,
400
+ status: 'running',
401
+ exit_code: null,
402
+ output: null,
403
+ started_at: now(),
404
+ finished_at: null,
405
+ });
406
+ // Update current_step on the task record
407
+ store.updateTaskStatus(taskRecord.id, convoyId, 'running', {});
408
+ const stepMaxRetries = step.max_retries ?? taskRecord.max_retries;
409
+ let stepResult = { success: false, output: '', exitCode: -1 };
410
+ let stepAttempt = 0;
411
+ while (stepAttempt <= stepMaxRetries) {
412
+ // Prepend prior failure context on retries
413
+ let stepPrompt = step.prompt;
414
+ if (stepAttempt > 0 && stepResult) {
415
+ const failedOutput = stepResult.output || '(no output)';
416
+ stepPrompt = `Previous attempt failed.\nExit code: ${stepResult.exitCode}\nError output:\n${failedOutput}\n\nFix the issues and try again.\n\n` + step.prompt;
417
+ }
418
+ const stepTask = {
419
+ id: taskRecord.id,
420
+ prompt: stepPrompt,
421
+ agent: taskRecord.agent,
422
+ timeout: `${taskRecord.timeout_ms}ms`,
423
+ depends_on: [],
424
+ files: taskRecord.files ? JSON.parse(taskRecord.files) : [],
425
+ description: `step ${i + 1}`,
426
+ max_retries: stepMaxRetries,
427
+ };
428
+ try {
429
+ stepResult = await adapter.execute(stepTask, { verbose, cwd: worktreePath ?? basePath });
430
+ }
431
+ catch (err) {
432
+ stepResult = { success: false, output: err.message, exitCode: -1 };
433
+ }
434
+ if (stepResult.success)
435
+ break;
436
+ stepAttempt++;
437
+ if (stepAttempt <= stepMaxRetries) {
438
+ process.stdout.write(` ↺ step ${i + 1}/${steps.length} failed, retry ${stepAttempt}/${stepMaxRetries}\n`);
439
+ }
440
+ }
441
+ lastExitCode = stepResult.exitCode;
442
+ combinedOutput += `\n[Step ${i + 1}]\n${stepResult.output}`;
443
+ if (step.id) {
444
+ stepResults.set(step.id, { exitCode: stepResult.exitCode });
445
+ }
446
+ // Run step-level gates if present
447
+ if (step.gates && step.gates.length > 0 && stepResult.success) {
448
+ let gateFailure = null;
449
+ const execFileCb = (await import('node:child_process')).execFile;
450
+ const execFileP = (await import('node:util')).promisify(execFileCb);
451
+ for (const command of step.gates) {
452
+ try {
453
+ // SECURITY: Gate/hook commands come from the .convoy.yml spec file, which is operator-controlled.
454
+ // They are NOT user-supplied and are part of the trusted build configuration.
455
+ await execFileP('sh', ['-c', command], { cwd: worktreePath ?? basePath });
456
+ }
457
+ catch (gateErr) {
458
+ const ge = gateErr;
459
+ const code = typeof ge.code === 'number' ? ge.code : 1;
460
+ const output = ge.stderr || ge.stdout || ge.message || '';
461
+ gateFailure = { command, exitCode: code, output };
462
+ break;
463
+ }
464
+ }
465
+ if (gateFailure !== null) {
466
+ stepResult = { success: false, output: `Gate failed: ${gateFailure.command}\nExit code: ${gateFailure.exitCode}\n${gateFailure.output}`, exitCode: gateFailure.exitCode };
467
+ lastExitCode = gateFailure.exitCode;
468
+ combinedOutput += `\n[Step ${i + 1} gate failed: ${gateFailure.command}]`;
469
+ }
470
+ }
471
+ // Update step record
472
+ store.updateTaskStep(stepDbId, {
473
+ status: stepResult.success ? 'done' : 'failed',
474
+ exit_code: stepResult.exitCode,
475
+ output: stepResult.output,
476
+ finished_at: now(),
477
+ });
478
+ if (!stepResult.success) {
479
+ return {
480
+ success: false,
481
+ output: combinedOutput.trim(),
482
+ exitCode: lastExitCode,
483
+ };
484
+ }
485
+ }
486
+ return {
487
+ success: true,
488
+ output: combinedOutput.trim(),
489
+ exitCode: lastExitCode,
490
+ };
491
+ }
492
+ // ── File-based injection ──────────────────────────────────────────────────────
493
+ const INJECT_DIR = '.opencastle/convoy-inject';
494
+ const CONVOY_ID_RE = /^[a-zA-Z0-9-]+$/;
495
+ const MAX_FILE_INJECTED_TASKS = 10;
496
+ function pollInjectFile(convoyId, store, events, basePath) {
497
+ // Path traversal guard: convoy_id must be alphanumeric + hyphens only
498
+ if (!CONVOY_ID_RE.test(convoyId))
499
+ return 0;
500
+ const injectDir = join(basePath, INJECT_DIR, convoyId);
501
+ const injectPath = join(injectDir, 'inject.yml');
502
+ if (!existsSync(injectPath))
503
+ return 0;
504
+ // Atomic rename to prevent double-read
505
+ const processingPath = injectPath + '.processing';
506
+ try {
507
+ renameSync(injectPath, processingPath);
508
+ }
509
+ catch {
510
+ return 0; // Another process may have grabbed it
511
+ }
512
+ let raw;
513
+ try {
514
+ raw = readFileSync(processingPath, 'utf8');
515
+ }
516
+ catch {
517
+ return 0;
518
+ }
519
+ finally {
520
+ try {
521
+ unlinkSync(processingPath);
522
+ }
523
+ catch { /* ignore */ }
524
+ }
525
+ let parsed;
526
+ try {
527
+ parsed = parseYaml(raw);
528
+ if (!parsed || typeof parsed !== 'object' || !Array.isArray(parsed.tasks)) {
529
+ process.stderr.write(`Warning: inject file has invalid format (expected { tasks: [...] })\n`);
530
+ return 0;
531
+ }
532
+ }
533
+ catch (err) {
534
+ process.stderr.write(`Warning: failed to parse inject file: ${err.message}\n`);
535
+ return 0;
536
+ }
537
+ const tasks = parsed.tasks;
538
+ const allExisting = store.getTasksByConvoy(convoyId);
539
+ const existingFileInjected = allExisting.filter(t => t.provenance === 'file-injection').length;
540
+ const remaining = MAX_FILE_INJECTED_TASKS - existingFileInjected;
541
+ let injectedCount = 0;
542
+ for (const rawTask of tasks) {
543
+ if (injectedCount >= remaining) {
544
+ process.stderr.write(`Warning: file injection limit reached (${MAX_FILE_INJECTED_TASKS}), skipping remaining tasks\n`);
545
+ break;
546
+ }
547
+ // Validate required fields
548
+ if (!rawTask.id || typeof rawTask.id !== 'string') {
549
+ process.stderr.write(`Warning: skipping injected task with missing/invalid id\n`);
550
+ continue;
551
+ }
552
+ if (!rawTask.prompt || typeof rawTask.prompt !== 'string') {
553
+ process.stderr.write(`Warning: skipping injected task "${rawTask.id}": missing prompt\n`);
554
+ continue;
555
+ }
556
+ if (!rawTask.agent || typeof rawTask.agent !== 'string') {
557
+ process.stderr.write(`Warning: skipping injected task "${rawTask.id}": missing agent\n`);
558
+ continue;
559
+ }
560
+ // Check ID uniqueness
561
+ if (allExisting.some(t => t.id === rawTask.id)) {
562
+ process.stderr.write(`Warning: skipping injected task "${rawTask.id}": ID already exists\n`);
563
+ continue;
564
+ }
565
+ // Determine phase — inject into last scheduled phase
566
+ const maxPhase = allExisting.reduce((max, t) => Math.max(max, t.phase), 0);
567
+ // Validate file paths before building the record
568
+ let validatedFiles = null;
569
+ if (rawTask.files && Array.isArray(rawTask.files)) {
570
+ try {
571
+ validatedFiles = JSON.stringify(rawTask.files.map(f => normalizePath(f)));
572
+ }
573
+ catch (err) {
574
+ process.stderr.write(`Warning: skipping injected task "${rawTask.id}": invalid file path: ${err.message}\n`);
575
+ continue;
576
+ }
577
+ }
578
+ const record = {
579
+ id: rawTask.id,
580
+ convoy_id: convoyId,
581
+ phase: maxPhase,
582
+ prompt: rawTask.prompt,
583
+ agent: rawTask.agent,
584
+ adapter: null,
585
+ model: null,
586
+ timeout_ms: typeof rawTask.timeout_ms === 'number' ? rawTask.timeout_ms : 1_800_000,
587
+ status: 'pending',
588
+ worker_id: null,
589
+ worktree: null,
590
+ output: null,
591
+ exit_code: null,
592
+ started_at: null,
593
+ finished_at: null,
594
+ retries: 0,
595
+ max_retries: typeof rawTask.max_retries === 'number' ? rawTask.max_retries : 1,
596
+ files: validatedFiles,
597
+ depends_on: null,
598
+ prompt_tokens: null,
599
+ completion_tokens: null,
600
+ total_tokens: null,
601
+ cost_usd: null,
602
+ gates: null,
603
+ on_exhausted: 'dlq',
604
+ injected: 1,
605
+ provenance: 'file-injection',
606
+ idempotency_key: null,
607
+ current_step: null,
608
+ total_steps: null,
609
+ review_level: null,
610
+ review_verdict: null,
611
+ review_tokens: null,
612
+ review_model: null,
613
+ panel_attempts: 0,
614
+ dispute_id: null,
615
+ drift_score: null,
616
+ drift_retried: 0,
617
+ outputs: null,
618
+ inputs: null,
619
+ discovered_issues: null,
620
+ };
621
+ try {
622
+ store.insertInjectedTask(record);
623
+ injectedCount++;
624
+ }
625
+ catch (err) {
626
+ process.stderr.write(`Warning: failed to inject task "${rawTask.id}": ${err.message}\n`);
627
+ }
628
+ }
629
+ if (injectedCount > 0) {
630
+ events.emit('file_injection_received', {
631
+ task_count: injectedCount,
632
+ source: injectPath,
633
+ }, { convoy_id: convoyId });
634
+ }
635
+ return injectedCount;
636
+ }
47
637
  // ── Core convoy execution ─────────────────────────────────────────────────────
48
- async function runConvoy(convoyId, spec, adapter, store, events, wtManager, mergeQueue, basePath, baseBranch, verbose, startTime) {
638
+ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, mergeQueue, basePath, baseBranch, verbose, startTime, ndjsonPath, reviewRunner) {
49
639
  const totalTasks = spec.tasks?.length ?? 0;
50
640
  let completedCount = 0;
51
641
  const activeTaskMap = new Map();
642
+ const reviewSemaphore = new ReviewSemaphore(spec.defaults?.max_concurrent_reviews ?? 3);
643
+ let reviewTokensTotal = 0;
52
644
  const taskAdapterMap = new Map();
53
645
  const healthMonitor = createHealthMonitor({
54
646
  store,
@@ -65,6 +657,17 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
65
657
  },
66
658
  });
67
659
  healthMonitor.start();
660
+ // ── Circuit breaker ────────────────────────────────────────────────────────
661
+ const circuitBreakerConfig = spec.defaults?.circuit_breaker;
662
+ const convoyRecord = store.getConvoy(convoyId);
663
+ const initialCircuitState = convoyRecord?.circuit_state ? JSON.parse(convoyRecord.circuit_state) : undefined;
664
+ const circuitBreaker = new CircuitBreakerManager(circuitBreakerConfig, initialCircuitState);
665
+ // ── Trust model ────────────────────────────────────────────────────────────
666
+ // Gate commands, hook commands, and step commands in .convoy.yml are treated
667
+ // as operator-controlled build configuration (analogous to Makefiles, CI
668
+ // configs, or package.json scripts). They are executed via sh -c and must
669
+ // NOT contain user-supplied input. The spec file itself is the trust boundary.
670
+ // ──────────────────────────────────────────────────────────────────────────
68
671
  // ── Task skipping ─────────────────────────────────────────────────────────
69
672
  function skipTask(taskId, reason, visited = new Set()) {
70
673
  if (visited.has(taskId))
@@ -101,6 +704,134 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
101
704
  }
102
705
  }
103
706
  }
707
+ function handleExhaustion(taskRecord, failureType, errorOutput) {
708
+ const exhausted = taskRecord.on_exhausted ?? 'dlq';
709
+ if (exhausted === 'dlq' || exhausted === 'stop') {
710
+ const dlqId = `dlq-${taskRecord.id}-${Date.now()}`;
711
+ // Pre-scan: build the markdown entry and check for secrets BEFORE any
712
+ // writes. This keeps the SQLite DLQ row and the Markdown file in sync —
713
+ // either both are written or neither is (MF-2 dual-write atomicity).
714
+ const { marker: dlqMarker, entry: dlqMdEntry } = buildDlqMarkdownEntry(dlqId, taskRecord, failureType, errorOutput);
715
+ const dlqScanResult = scanForSecrets(dlqMdEntry, 'AGENT-FAILURES.md');
716
+ if (!dlqScanResult.clean) {
717
+ // Block BOTH writes to maintain consistent state
718
+ events.emit('secret_leak_prevented', {
719
+ task_id: taskRecord.id,
720
+ findings_count: dlqScanResult.findings.length,
721
+ patterns: dlqScanResult.findings.map((f) => f.pattern),
722
+ context: 'dlq_dual_write',
723
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
724
+ }
725
+ else {
726
+ // Clean — proceed with both writes atomically
727
+ store.insertDlqEntry({
728
+ id: dlqId,
729
+ convoy_id: convoyId,
730
+ task_id: taskRecord.id,
731
+ agent: taskRecord.agent,
732
+ failure_type: failureType,
733
+ error_output: errorOutput,
734
+ attempts: taskRecord.retries + 1,
735
+ tokens_spent: taskRecord.total_tokens,
736
+ escalation_task_id: null,
737
+ resolved: 0,
738
+ resolution: null,
739
+ created_at: new Date().toISOString(),
740
+ resolved_at: null,
741
+ });
742
+ appendDlqMarkdownClean(dlqMarker, dlqMdEntry);
743
+ events.emit('dlq_entry_created', {
744
+ dlq_id: dlqId,
745
+ task_id: taskRecord.id,
746
+ agent: taskRecord.agent,
747
+ failure_type: failureType,
748
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
749
+ }
750
+ }
751
+ if (exhausted === 'stop') {
752
+ // Skip all remaining pending tasks + set convoy to failed
753
+ const allPending = store.getTasksByConvoy(convoyId).filter(t => t.status === 'pending');
754
+ for (const t of allPending) {
755
+ skipTask(t.id, `on_exhausted: stop — task "${taskRecord.id}" exhausted retries`);
756
+ }
757
+ store.updateConvoyStatus(convoyId, 'failed');
758
+ events.emit('convoy_failed', { status: 'failed', reason: `on_exhausted: stop — task "${taskRecord.id}" exhausted retries` }, { convoy_id: convoyId });
759
+ }
760
+ else if (exhausted === 'dlq' || exhausted === 'skip') {
761
+ // Default behavior: cascade failure to dependents only
762
+ cascadeFailure(taskRecord.id);
763
+ }
764
+ // ── Circuit breaker: record exhaustion failure ──────────────────────────
765
+ if (circuitBreakerConfig) {
766
+ const { tripped } = circuitBreaker.recordFailure(taskRecord.agent);
767
+ try {
768
+ store.updateConvoyCircuitState(convoyId, circuitBreaker.serialize());
769
+ }
770
+ catch { /* non-critical */ }
771
+ if (tripped) {
772
+ events.emit('circuit_breaker_tripped', {
773
+ agent: taskRecord.agent,
774
+ state: circuitBreaker.getState(taskRecord.agent),
775
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
776
+ }
777
+ }
778
+ }
779
+ // ── Hook execution ────────────────────────────────────────────────────────
780
+ async function runHooks(hooks, lifecycle, context) {
781
+ const filtered = hooks.filter(h => (h.on ?? 'post_task') === lifecycle);
782
+ for (const hook of filtered) {
783
+ if (hook.type === 'command' || hook.type === 'guard' || hook.type === 'validate') {
784
+ const cmd = hook.command;
785
+ if (!cmd)
786
+ continue;
787
+ try {
788
+ // SECURITY: Gate/hook commands come from the .convoy.yml spec file, which is operator-controlled.
789
+ // They are NOT user-supplied and are part of the trusted build configuration.
790
+ await execFile('sh', ['-c', cmd], { cwd: context.cwd });
791
+ }
792
+ catch (err) {
793
+ const execErr = err;
794
+ const errorMsg = execErr.stderr || execErr.stdout || execErr.message || '';
795
+ return { passed: false, failedHook: hook, error: errorMsg };
796
+ }
797
+ }
798
+ else if (hook.type === 'agent') {
799
+ if (!hook.prompt)
800
+ continue;
801
+ const hookTask = {
802
+ id: `hook-${lifecycle}-${context.taskId ?? 'convoy'}-${Date.now()}`,
803
+ prompt: hook.prompt,
804
+ agent: hook.name ?? 'developer',
805
+ timeout: '10m',
806
+ depends_on: [],
807
+ files: [],
808
+ description: `Hook: ${hook.name ?? hook.type}`,
809
+ max_retries: 0,
810
+ };
811
+ try {
812
+ const hookResult = await adapter.execute(hookTask, { verbose, cwd: context.cwd });
813
+ if (!hookResult.success) {
814
+ return { passed: false, failedHook: hook, error: hookResult.output };
815
+ }
816
+ }
817
+ catch (err) {
818
+ return { passed: false, failedHook: hook, error: err.message };
819
+ }
820
+ }
821
+ else if (hook.type === 'review') {
822
+ if (!context.taskId || !reviewRunner)
823
+ continue;
824
+ const reviewTaskRecord = store.getTask(context.taskId, context.convoyId);
825
+ if (reviewTaskRecord) {
826
+ const reviewResult = await reviewRunner(reviewTaskRecord, 'fast', spec.defaults?.reviewer_model ?? 'default');
827
+ if (reviewResult.verdict !== 'pass') {
828
+ return { passed: false, failedHook: hook, error: reviewResult.feedback };
829
+ }
830
+ }
831
+ }
832
+ }
833
+ return { passed: true };
834
+ }
104
835
  // ── Single-task executor ──────────────────────────────────────────────────
105
836
  async function executeOneTask(taskRecord) {
106
837
  const workerId = `worker-${taskRecord.id}-${Date.now()}`;
@@ -119,6 +850,65 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
119
850
  }
120
851
  }
121
852
  taskAdapterMap.set(taskRecord.id, taskAdapter);
853
+ // ── Check inputs availability ────────────────────────────────────────────
854
+ if (taskRecord.inputs) {
855
+ const inputs = JSON.parse(taskRecord.inputs);
856
+ for (const input of inputs) {
857
+ const artifact = store.getArtifact(convoyId, input.name);
858
+ if (!artifact) {
859
+ store.updateTaskStatus(taskRecord.id, convoyId, 'wait-for-input');
860
+ events.emit('task_waiting_input', {
861
+ task_id: taskRecord.id,
862
+ missing_artifact: input.name,
863
+ from_task: input.from,
864
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
865
+ taskAdapterMap.delete(taskRecord.id);
866
+ return;
867
+ }
868
+ }
869
+ }
870
+ // ── Circuit breaker check ──────────────────────────────────────────────
871
+ if (circuitBreakerConfig) {
872
+ if (!circuitBreaker.canAssign(taskRecord.agent)) {
873
+ const fallback = circuitBreaker.fallback;
874
+ if (fallback) {
875
+ events.emit('circuit_breaker_fallback', {
876
+ original_agent: taskRecord.agent,
877
+ fallback_agent: fallback,
878
+ state: circuitBreaker.getState(taskRecord.agent),
879
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
880
+ }
881
+ else {
882
+ events.emit('circuit_breaker_blocked', {
883
+ agent: taskRecord.agent,
884
+ state: circuitBreaker.getState(taskRecord.agent),
885
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
886
+ }
887
+ store.updateTaskStatus(taskRecord.id, convoyId, 'skipped', {
888
+ output: `Circuit breaker open for agent "${taskRecord.agent}". ${fallback ? `No fallback available.` : `No fallback configured.`}`,
889
+ });
890
+ completedCount++;
891
+ taskAdapterMap.delete(taskRecord.id);
892
+ cascadeFailure(taskRecord.id);
893
+ return;
894
+ }
895
+ }
896
+ // ── Intelligence: circuit breaker weak-area avoidance (Phase 18.2) ─────
897
+ if (spec.defaults?.avoid_weak_agents) {
898
+ try {
899
+ const weakAreas = feedCircuitBreaker(taskRecord.agent, basePath);
900
+ const taskFiles = taskRecord.files ? JSON.parse(taskRecord.files) : [];
901
+ const matchesWeakArea = weakAreas.some(area => taskFiles.some(f => f.toLowerCase().includes(area.toLowerCase())));
902
+ if (matchesWeakArea && taskRecord.retries === 0) {
903
+ events.emit('weak_area_skipped', { agent: taskRecord.agent, weak_areas: weakAreas, task_files: taskFiles }, { convoy_id: convoyId, task_id: taskRecord.id });
904
+ store.updateTaskStatus(taskRecord.id, convoyId, 'skipped', { output: `Agent "${taskRecord.agent}" has weak-area match for task files. Skipped by avoid_weak_agents policy.` });
905
+ completedCount++;
906
+ taskAdapterMap.delete(taskRecord.id);
907
+ return;
908
+ }
909
+ }
910
+ catch { /* non-critical */ }
911
+ }
122
912
  // Create worktree (skip for copilot adapter)
123
913
  let worktreePath = null;
124
914
  if (taskAdapter.name !== 'copilot') {
@@ -150,16 +940,149 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
150
940
  store.updateWorkerStatus(workerId, 'running');
151
941
  const task = taskRecordToTask(taskRecord);
152
942
  activeTaskMap.set(taskRecord.id, task);
943
+ // ── Inject inputs into prompt ────────────────────────────────────────────
944
+ if (taskRecord.inputs) {
945
+ const inputs = JSON.parse(taskRecord.inputs);
946
+ for (const input of inputs) {
947
+ const artifact = store.getArtifact(convoyId, input.name);
948
+ const templateVar = input.as ?? input.name;
949
+ task.prompt = task.prompt.replaceAll(`{{input.${templateVar}}}`, artifact.content);
950
+ }
951
+ }
952
+ // ── Scratchpad template substitution (Phase 17.1) ───────────────────────
953
+ const scratchpadRe = /\{\{scratchpad\.([a-zA-Z0-9_.-]+)\}\}/g;
954
+ let scratchpadMatch;
955
+ while ((scratchpadMatch = scratchpadRe.exec(task.prompt)) !== null) {
956
+ const spKey = scratchpadMatch[1];
957
+ const spVal = store.getScratchpadValue(spKey);
958
+ if (spVal !== null) {
959
+ task.prompt = task.prompt.replaceAll(`{{scratchpad.${spKey}}}`, spVal);
960
+ scratchpadRe.lastIndex = 0; // reset after replaceAll
961
+ }
962
+ }
153
963
  process.stdout.write(` ${c.cyan('▶')} ${c.bold(`[${taskRecord.id}]`)} ${taskRecord.agent}${worktreePath ? c.dim(' (worktree)') : ''}\n`);
154
964
  events.emit('task_started', { worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
155
965
  const taskStartTime = Date.now();
966
+ // ── Outbound prompt scan — NEVER send a prompt containing secrets ─────────
967
+ const promptScan = scanForSecrets(taskRecord.prompt, `task:${taskRecord.id}`);
968
+ if (!promptScan.clean) {
969
+ store.updateTaskStatus(taskRecord.id, convoyId, 'failed', {
970
+ finished_at: now(),
971
+ output: `Secret detected in prompt — task blocked before execution.\nFindings:\n${promptScan.findings
972
+ .map((f) => ` ${f.pattern} at line ${f.line}: ${f.snippet}`)
973
+ .join('\n')}`,
974
+ });
975
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: now() });
976
+ completedCount++;
977
+ events.emit('secret_leak_prevented', {
978
+ task_id: taskRecord.id,
979
+ findings_count: promptScan.findings.length,
980
+ patterns: promptScan.findings.map((f) => f.pattern),
981
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
982
+ cascadeFailure(taskRecord.id);
983
+ taskAdapterMap.delete(taskRecord.id);
984
+ return;
985
+ }
156
986
  const timeout = makeTimeoutPromise(taskRecord.timeout_ms);
157
987
  let result;
988
+ // Retrieve steps from spec if defined
989
+ const specTask = (spec.tasks ?? []).find(t => t.id === taskRecord.id);
990
+ const steps = specTask?.steps;
991
+ const taskHooks = specTask?.hooks ?? [];
992
+ // ── Intelligence: inject lessons (Phase 18.1) ─────────────────────────
993
+ if (spec.defaults?.inject_lessons !== false) {
994
+ try {
995
+ const taskFiles = taskRecord.files ? JSON.parse(taskRecord.files) : [];
996
+ const lessons = readLessons(taskRecord.agent, taskFiles, basePath);
997
+ if (lessons.length > 0) {
998
+ const lessonsBlock = '\n\n---\nRelevant lessons from previous sessions:\n'
999
+ + lessons.join('\n\n')
1000
+ + '\n---\n\n';
1001
+ task.prompt = lessonsBlock + task.prompt;
1002
+ }
1003
+ }
1004
+ catch { /* non-critical */ }
1005
+ }
1006
+ // ── Intelligence: inject persistent agent identity (Phase 17.2) ────────
1007
+ const specTaskForPersistent = (spec.tasks ?? []).find(t => t.id === taskRecord.id);
1008
+ if (specTaskForPersistent?.persistent) {
1009
+ try {
1010
+ const identities = store.getAgentIdentities(taskRecord.agent, 3);
1011
+ if (identities.length > 0) {
1012
+ const contextBlock = '\n\n[Previous work context]\n'
1013
+ + identities.map(id => id.summary).join('\n\n')
1014
+ + '\n[End previous context]\n\n';
1015
+ task.prompt = contextBlock + task.prompt;
1016
+ }
1017
+ }
1018
+ catch { /* non-critical */ }
1019
+ }
1020
+ // ── Intelligence: inject discovered issues instruction (Phase 18.4) ────
1021
+ if (spec.defaults?.track_discovered_issues) {
1022
+ task.prompt = injectDiscoveredIssuesInstruction(task.prompt);
1023
+ }
1024
+ // ── pre_task hooks ────────────────────────────────────────────────────────
1025
+ if (taskHooks.length > 0) {
1026
+ const preResult = await runHooks(taskHooks, 'pre_task', {
1027
+ taskId: taskRecord.id,
1028
+ convoyId,
1029
+ cwd: worktreePath ?? basePath,
1030
+ });
1031
+ if (!preResult.passed) {
1032
+ await removeWorktree();
1033
+ const hookLabel = preResult.failedHook?.name ?? preResult.failedHook?.type ?? 'unknown';
1034
+ store.withTransaction(() => {
1035
+ store.updateTaskStatus(taskRecord.id, convoyId, 'hook-failed', {
1036
+ finished_at: now(),
1037
+ output: `pre_task hook "${hookLabel}" failed: ${preResult.error ?? ''}`,
1038
+ exit_code: 1,
1039
+ });
1040
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: now() });
1041
+ });
1042
+ completedCount++;
1043
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} pre_task hook failed ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1044
+ events.emit('task_failed', { reason: 'hook-failed', hook: hookLabel, worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1045
+ cascadeFailure(taskRecord.id);
1046
+ taskAdapterMap.delete(taskRecord.id);
1047
+ return;
1048
+ }
1049
+ }
1050
+ // ── Symlink security scan (pre-execution) ────────────────────────────────
1051
+ const taskFiles = taskRecord.files ? JSON.parse(taskRecord.files) : [];
1052
+ if (taskFiles.length > 0 && worktreePath) {
1053
+ try {
1054
+ scanSymlinks(taskFiles, worktreePath);
1055
+ }
1056
+ catch (err) {
1057
+ await removeWorktree();
1058
+ store.withTransaction(() => {
1059
+ store.updateTaskStatus(taskRecord.id, convoyId, 'failed', {
1060
+ finished_at: now(),
1061
+ output: `Symlink security check failed: ${err.message}`,
1062
+ exit_code: 1,
1063
+ });
1064
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: now() });
1065
+ });
1066
+ completedCount++;
1067
+ events.emit('task_failed', { reason: 'symlink-escape', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1068
+ cascadeFailure(taskRecord.id);
1069
+ taskAdapterMap.delete(taskRecord.id);
1070
+ return;
1071
+ }
1072
+ }
158
1073
  try {
159
- result = await Promise.race([
160
- taskAdapter.execute(task, { verbose, cwd: worktreePath ?? basePath }),
161
- timeout.promise,
162
- ]);
1074
+ if (steps && steps.length > 0) {
1075
+ result = await Promise.race([
1076
+ executeSteps(taskRecord, steps, taskAdapter, worktreePath, basePath, store, convoyId, verbose),
1077
+ timeout.promise,
1078
+ ]);
1079
+ }
1080
+ else {
1081
+ result = await Promise.race([
1082
+ taskAdapter.execute(task, { verbose, cwd: worktreePath ?? basePath }),
1083
+ timeout.promise,
1084
+ ]);
1085
+ }
163
1086
  timeout.clear();
164
1087
  }
165
1088
  catch (err) {
@@ -184,12 +1107,14 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
184
1107
  await removeWorktree();
185
1108
  const freshRecord = store.getTask(taskRecord.id, convoyId);
186
1109
  if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1110
+ const contextPrefix = `Previous attempt timed out.\n\nFix the issues and try again.\n\n`;
187
1111
  store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
188
1112
  retries: freshRecord.retries + 1,
189
1113
  worker_id: null,
190
1114
  worktree: null,
191
1115
  started_at: null,
192
1116
  finished_at: null,
1117
+ prompt: contextPrefix + taskRecord.prompt,
193
1118
  });
194
1119
  store.updateWorkerStatus(workerId, 'killed', { finished_at: finishedAt });
195
1120
  process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} timed out, retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
@@ -226,23 +1151,654 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
226
1151
  phase: taskRecord.phase,
227
1152
  convoy_id: convoyId,
228
1153
  }, { convoy_id: convoyId, task_id: taskRecord.id });
229
- cascadeFailure(taskRecord.id);
1154
+ handleExhaustion(freshRecord, 'timeout', result.output || null);
230
1155
  }
231
1156
  taskAdapterMap.delete(taskRecord.id);
232
1157
  return;
233
1158
  }
234
1159
  // ── Success ─────────────────────────────────────────────────────────────
235
- if (result.success) {
236
- if (worktreePath) {
1160
+ if (result.success) { // ── Per-task gates ─────────────────────────────────────────────────────
1161
+ const taskGates = taskRecord.gates ? JSON.parse(taskRecord.gates) : [];
1162
+ if (taskGates.length > 0) {
1163
+ let gateFailure = null;
1164
+ for (const command of taskGates) {
1165
+ try {
1166
+ // SECURITY: Gate/hook commands come from the .convoy.yml spec file, which is operator-controlled.
1167
+ // They are NOT user-supplied and are part of the trusted build configuration.
1168
+ await execFile('sh', ['-c', command], { cwd: worktreePath ?? basePath });
1169
+ }
1170
+ catch (err) {
1171
+ const execErr = err;
1172
+ const code = typeof execErr.code === 'number' ? execErr.code : 1;
1173
+ const output = execErr.stderr || execErr.stdout || execErr.message || '';
1174
+ gateFailure = { command, exitCode: code, output };
1175
+ break;
1176
+ }
1177
+ }
1178
+ if (gateFailure !== null) {
1179
+ await removeWorktree();
1180
+ const freshRecord = store.getTask(taskRecord.id, convoyId);
1181
+ if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1182
+ const contextPrefix = `Previous attempt's gate check failed.\nGate: ${gateFailure.command}\nExit code: ${gateFailure.exitCode}\nOutput:\n${gateFailure.output || '(no output)'}\n\nFix the issues and try again.\n\n`;
1183
+ store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
1184
+ retries: freshRecord.retries + 1,
1185
+ worker_id: null,
1186
+ worktree: null,
1187
+ started_at: null,
1188
+ finished_at: null,
1189
+ prompt: contextPrefix + taskRecord.prompt,
1190
+ });
1191
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1192
+ process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} gate failed, retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
1193
+ }
1194
+ else {
1195
+ store.withTransaction(() => {
1196
+ store.updateTaskStatus(taskRecord.id, convoyId, 'gate-failed', {
1197
+ finished_at: finishedAt,
1198
+ output: `Gate failed: ${gateFailure.command}\nExit code: ${gateFailure.exitCode}\n${gateFailure.output}`,
1199
+ exit_code: gateFailure.exitCode,
1200
+ });
1201
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1202
+ });
1203
+ completedCount++;
1204
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} gate failed ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1205
+ events.emit('task_failed', { reason: 'gate-failed', gate: gateFailure.command, exit_code: gateFailure.exitCode, worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1206
+ events.emit('session', {
1207
+ agent: taskRecord.agent,
1208
+ model: taskRecord.model ?? taskAdapter.name,
1209
+ task: taskRecord.id,
1210
+ outcome: 'failed',
1211
+ duration_min: Math.round((Date.now() - taskStartTime) / 60_000),
1212
+ files_changed: 0,
1213
+ retries: freshRecord.retries,
1214
+ convoy_id: convoyId,
1215
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1216
+ events.emit('delegation', {
1217
+ session_id: convoyId,
1218
+ agent: taskRecord.agent,
1219
+ model: taskRecord.model ?? taskAdapter.name,
1220
+ tier: 'standard',
1221
+ mechanism: 'convoy',
1222
+ outcome: 'failed',
1223
+ retries: freshRecord.retries,
1224
+ phase: taskRecord.phase,
1225
+ convoy_id: convoyId,
1226
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1227
+ handleExhaustion(freshRecord, 'gate-failed', gateFailure.output || null);
1228
+ }
1229
+ taskAdapterMap.delete(taskRecord.id);
1230
+ return;
1231
+ }
1232
+ }
1233
+ // ── Built-in gates ────────────────────────────────────────────────────
1234
+ const builtInGates = spec.defaults?.built_in_gates;
1235
+ if (builtInGates && worktreePath) {
1236
+ if (builtInGates.browser_test) {
1237
+ const specTask = (spec.tasks ?? []).find(t => t.id === taskRecord.id);
1238
+ const taskBrowserConfig = specTask?.browser_test ?? spec.defaults?.browser_test;
1239
+ if (!taskBrowserConfig) {
1240
+ process.stderr.write(`Warning: browser_test gate enabled but no browser_test config (urls) found — skipping\n`);
1241
+ }
1242
+ else {
1243
+ const browserResult = await browserTestGate({
1244
+ mcpServers: spec.defaults?.mcp_servers ?? [],
1245
+ taskConfig: taskBrowserConfig,
1246
+ worktreePath,
1247
+ approvalTimeout: spec.defaults?.mcp_server_approval_timeout,
1248
+ });
1249
+ events.emit('built_in_gate_result', { gate: 'browser_test', passed: browserResult.passed, output: browserResult.output }, { convoy_id: convoyId, task_id: taskRecord.id });
1250
+ if (!browserResult.passed) {
1251
+ await removeWorktree();
1252
+ const freshRecord = store.getTask(taskRecord.id, convoyId);
1253
+ if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1254
+ store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
1255
+ retries: freshRecord.retries + 1,
1256
+ worker_id: null,
1257
+ worktree: null,
1258
+ started_at: null,
1259
+ finished_at: null,
1260
+ });
1261
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1262
+ process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} browser test gate failed, retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
1263
+ }
1264
+ else {
1265
+ store.withTransaction(() => {
1266
+ store.updateTaskStatus(taskRecord.id, convoyId, 'gate-failed', {
1267
+ finished_at: finishedAt,
1268
+ output: `Built-in gate (browser_test) failed:\n${browserResult.output}`,
1269
+ exit_code: 1,
1270
+ });
1271
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1272
+ });
1273
+ completedCount++;
1274
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} browser test gate failed ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1275
+ events.emit('task_failed', { reason: 'gate-failed', gate: 'browser_test', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1276
+ handleExhaustion(freshRecord, 'browser-test', browserResult.output);
1277
+ }
1278
+ taskAdapterMap.delete(taskRecord.id);
1279
+ return;
1280
+ }
1281
+ }
1282
+ }
1283
+ let changedFiles = [];
1284
+ let diff = '';
237
1285
  try {
238
- await mergeQueue.merge(worktreePath, `convoy-${workerId}`, baseBranch);
1286
+ const { stdout: filesOut } = await execFile('git', ['diff', '--name-only', `${baseBranch}..HEAD`], { cwd: worktreePath });
1287
+ changedFiles = filesOut.split('\n').filter(Boolean);
1288
+ const { stdout: diffOut } = await execFile('git', ['diff', `${baseBranch}..HEAD`], { cwd: worktreePath });
1289
+ diff = diffOut;
1290
+ }
1291
+ catch { /* no commits in worktree yet — skip */ }
1292
+ // Secret scan gate
1293
+ if (builtInGates.secret_scan && changedFiles.length > 0) {
1294
+ const scanResult = await runSecretScanGate(changedFiles, worktreePath);
1295
+ events.emit('built_in_gate_result', { gate: 'secret_scan', passed: scanResult.passed, output: scanResult.output }, { convoy_id: convoyId, task_id: taskRecord.id });
1296
+ if (!scanResult.passed) {
1297
+ await removeWorktree();
1298
+ const freshRecord = store.getTask(taskRecord.id, convoyId);
1299
+ if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1300
+ store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
1301
+ retries: freshRecord.retries + 1,
1302
+ worker_id: null,
1303
+ worktree: null,
1304
+ started_at: null,
1305
+ finished_at: null,
1306
+ prompt: `Secret scan gate failed.\n${scanResult.output}\n\nFix the issues and try again.\n\n${taskRecord.prompt}`,
1307
+ });
1308
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1309
+ process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} secret scan gate failed, retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
1310
+ }
1311
+ else {
1312
+ store.withTransaction(() => {
1313
+ store.updateTaskStatus(taskRecord.id, convoyId, 'gate-failed', {
1314
+ finished_at: finishedAt,
1315
+ output: `Built-in gate (secret_scan) failed:\n${scanResult.output}`,
1316
+ exit_code: 1,
1317
+ });
1318
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1319
+ });
1320
+ completedCount++;
1321
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} secret scan gate failed ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1322
+ events.emit('task_failed', { reason: 'gate-failed', gate: 'secret_scan', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1323
+ handleExhaustion(freshRecord, 'secret-scan', scanResult.output);
1324
+ }
1325
+ taskAdapterMap.delete(taskRecord.id);
1326
+ return;
1327
+ }
1328
+ }
1329
+ // Blast radius gate
1330
+ if (builtInGates.blast_radius && diff) {
1331
+ const blastResult = runBlastRadiusGate(diff);
1332
+ events.emit('built_in_gate_result', { gate: 'blast_radius', level: blastResult.level, passed: blastResult.passed, output: blastResult.output }, { convoy_id: convoyId, task_id: taskRecord.id });
1333
+ if (!blastResult.passed) {
1334
+ await removeWorktree();
1335
+ const freshRecord = store.getTask(taskRecord.id, convoyId);
1336
+ if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1337
+ store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
1338
+ retries: freshRecord.retries + 1,
1339
+ worker_id: null,
1340
+ worktree: null,
1341
+ started_at: null,
1342
+ finished_at: null,
1343
+ prompt: `Blast radius gate failed.\n${blastResult.output}\n\nFix the issues and try again.\n\n${taskRecord.prompt}`,
1344
+ });
1345
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1346
+ process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} blast radius gate failed, retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
1347
+ }
1348
+ else {
1349
+ store.withTransaction(() => {
1350
+ store.updateTaskStatus(taskRecord.id, convoyId, 'gate-failed', {
1351
+ finished_at: finishedAt,
1352
+ output: `Built-in gate (blast_radius) failed:\n${blastResult.output}`,
1353
+ exit_code: 1,
1354
+ });
1355
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1356
+ });
1357
+ completedCount++;
1358
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} blast radius gate failed ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1359
+ events.emit('task_failed', { reason: 'gate-failed', gate: 'blast_radius', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1360
+ handleExhaustion(freshRecord, 'gate-failed', blastResult.output);
1361
+ }
1362
+ taskAdapterMap.delete(taskRecord.id);
1363
+ return;
1364
+ }
1365
+ }
1366
+ }
1367
+ // ── Drift detection ──────────────────────────────────────────────────
1368
+ const specTaskForDrift = (spec.tasks ?? []).find(t => t.id === taskRecord.id);
1369
+ const isDriftEnabled = specTaskForDrift?.detect_drift ?? spec.defaults?.detect_drift ?? false;
1370
+ if (isDriftEnabled && taskRecord.drift_retried === 0) {
1371
+ const driftResult = await detectDrift(taskRecord, taskAdapter);
1372
+ events.emit('drift_check_result', {
1373
+ task_id: taskRecord.id,
1374
+ score: driftResult.score,
1375
+ threshold: driftResult.threshold,
1376
+ explanation: driftResult.explanation,
1377
+ drifted: driftResult.drifted,
1378
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1379
+ store.updateTaskDrift(taskRecord.id, convoyId, { drift_score: driftResult.score });
1380
+ if (driftResult.drifted) {
1381
+ events.emit('drift_detected', {
1382
+ task_id: taskRecord.id,
1383
+ score: driftResult.score,
1384
+ threshold: driftResult.threshold,
1385
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1386
+ await removeWorktree();
1387
+ store.updateTaskDrift(taskRecord.id, convoyId, { drift_retried: 1 });
1388
+ store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
1389
+ worker_id: null,
1390
+ worktree: null,
1391
+ started_at: null,
1392
+ finished_at: null,
1393
+ });
1394
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1395
+ process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} drift detected (score: ${driftResult.score.toFixed(2)}), retrying\n`);
1396
+ taskAdapterMap.delete(taskRecord.id);
1397
+ return;
1398
+ }
1399
+ }
1400
+ // ── Review pipeline ──────────────────────────────────────────────────
1401
+ const specTaskForReview = (spec.tasks ?? []).find(t => t.id === taskRecord.id);
1402
+ const taskReviewSetting = specTaskForReview?.review ?? spec.defaults?.review ?? 'auto';
1403
+ if (taskReviewSetting !== 'none') {
1404
+ // Compute diff stats from worktree
1405
+ let reviewChangedFiles = [];
1406
+ let reviewDiffLines = 0;
1407
+ if (worktreePath) {
1408
+ try {
1409
+ const { stdout: filesOut } = await execFile('git', ['diff', '--name-only', `${baseBranch}..HEAD`], { cwd: worktreePath });
1410
+ reviewChangedFiles = filesOut.split('\n').filter(Boolean);
1411
+ const { stdout: diffOut } = await execFile('git', ['diff', `${baseBranch}..HEAD`], { cwd: worktreePath });
1412
+ reviewDiffLines = diffOut.split('\n').filter(l => l.startsWith('+') || l.startsWith('-')).filter(l => !l.startsWith('+++') && !l.startsWith('---')).length;
1413
+ }
1414
+ catch { /* no commits yet */ }
1415
+ }
1416
+ const diffStats = {
1417
+ linesChanged: reviewDiffLines,
1418
+ filesChanged: reviewChangedFiles.length,
1419
+ filePaths: reviewChangedFiles,
1420
+ };
1421
+ // Determine review level
1422
+ let reviewLevel;
1423
+ if (taskReviewSetting === 'fast') {
1424
+ reviewLevel = 'fast';
1425
+ }
1426
+ else if (taskReviewSetting === 'panel') {
1427
+ reviewLevel = 'panel';
1428
+ }
1429
+ else {
1430
+ reviewLevel = evaluateReviewLevel(taskRecord, diffStats, spec.defaults?.review_heuristics, true);
1431
+ }
1432
+ const reviewerModel = spec.defaults?.reviewer_model ?? 'default';
1433
+ events.emit('review_started', { level: reviewLevel, task_id: taskRecord.id, model: reviewerModel }, { convoy_id: convoyId, task_id: taskRecord.id });
1434
+ if (reviewLevel === 'auto-pass') {
1435
+ store.updateTaskReview(taskRecord.id, convoyId, {
1436
+ review_level: 'auto-pass',
1437
+ review_verdict: 'pass',
1438
+ review_tokens: 0,
1439
+ review_model: reviewerModel,
1440
+ });
1441
+ events.emit('review_verdict', { level: 'auto-pass', verdict: 'pass', tokens: 0, model: reviewerModel, feedback_length: 0 }, { convoy_id: convoyId, task_id: taskRecord.id });
1442
+ }
1443
+ else if (reviewLevel === 'fast') {
1444
+ // Check review budget
1445
+ const reviewBudget = spec.defaults?.review_budget;
1446
+ const onBudgetExceeded = spec.defaults?.on_review_budget_exceeded ?? 'skip';
1447
+ if (reviewBudget != null && reviewTokensTotal >= reviewBudget) {
1448
+ if (onBudgetExceeded === 'stop') {
1449
+ const allPending = store.getTasksByConvoy(convoyId).filter(t => t.status === 'pending');
1450
+ for (const t of allPending)
1451
+ skipTask(t.id, 'review_budget exceeded with on_review_budget_exceeded: stop');
1452
+ store.withTransaction(() => {
1453
+ store.updateTaskStatus(taskRecord.id, convoyId, 'review-blocked', { finished_at: finishedAt, output: 'Review budget exceeded', exit_code: 1 });
1454
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1455
+ });
1456
+ completedCount++;
1457
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} review budget exceeded (stop) ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1458
+ events.emit('review_verdict', { level: 'fast', verdict: 'skip', tokens: 0, model: reviewerModel, feedback_length: 0, budget_exceeded: true }, { convoy_id: convoyId, task_id: taskRecord.id });
1459
+ taskAdapterMap.delete(taskRecord.id);
1460
+ return;
1461
+ }
1462
+ else if (onBudgetExceeded === 'downgrade') {
1463
+ store.updateTaskReview(taskRecord.id, convoyId, { review_level: 'fast', review_verdict: 'pass', review_tokens: 0, review_model: reviewerModel });
1464
+ events.emit('review_verdict', { level: 'fast', verdict: 'pass', tokens: 0, model: reviewerModel, feedback_length: 0, budget_downgrade: true }, { convoy_id: convoyId, task_id: taskRecord.id });
1465
+ }
1466
+ else {
1467
+ // 'skip': treat as passed
1468
+ events.emit('review_verdict', { level: 'fast', verdict: 'pass', tokens: 0, model: reviewerModel, feedback_length: 0, budget_skip: true }, { convoy_id: convoyId, task_id: taskRecord.id });
1469
+ }
1470
+ }
1471
+ else {
1472
+ await reviewSemaphore.acquire();
1473
+ let reviewResult;
1474
+ try {
1475
+ if (reviewRunner) {
1476
+ reviewResult = await reviewRunner(taskRecord, 'fast', reviewerModel);
1477
+ }
1478
+ else {
1479
+ reviewResult = { verdict: 'pass', feedback: '', tokens: 0, model: reviewerModel };
1480
+ }
1481
+ }
1482
+ finally {
1483
+ reviewSemaphore.release();
1484
+ }
1485
+ reviewTokensTotal += reviewResult.tokens;
1486
+ store.updateTaskReview(taskRecord.id, convoyId, {
1487
+ review_level: 'fast',
1488
+ review_verdict: reviewResult.verdict,
1489
+ review_tokens: reviewResult.tokens,
1490
+ review_model: reviewResult.model,
1491
+ });
1492
+ store.updateConvoyReviewTokens(convoyId, reviewTokensTotal);
1493
+ events.emit('review_verdict', { level: 'fast', verdict: reviewResult.verdict, tokens: reviewResult.tokens, model: reviewResult.model, feedback_length: reviewResult.feedback.length }, { convoy_id: convoyId, task_id: taskRecord.id });
1494
+ if (reviewResult.verdict === 'block') {
1495
+ await removeWorktree();
1496
+ const freshRecord = store.getTask(taskRecord.id, convoyId);
1497
+ if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1498
+ const contextPrefix = `Previous attempt was blocked by review.\nFeedback:\n${reviewResult.feedback}\n\nFix the issues and try again.\n\n`;
1499
+ store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
1500
+ retries: freshRecord.retries + 1,
1501
+ worker_id: null,
1502
+ worktree: null,
1503
+ started_at: null,
1504
+ finished_at: null,
1505
+ prompt: contextPrefix + taskRecord.prompt,
1506
+ });
1507
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1508
+ process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} review blocked, retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
1509
+ taskAdapterMap.delete(taskRecord.id);
1510
+ return;
1511
+ }
1512
+ else {
1513
+ store.withTransaction(() => {
1514
+ store.updateTaskStatus(taskRecord.id, convoyId, 'review-blocked', {
1515
+ finished_at: finishedAt,
1516
+ output: `Review blocked: ${reviewResult.feedback}`,
1517
+ exit_code: 1,
1518
+ });
1519
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1520
+ });
1521
+ completedCount++;
1522
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} review blocked ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1523
+ events.emit('task_failed', { reason: 'review-blocked', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1524
+ handleExhaustion(freshRecord, 'review-blocked', reviewResult.feedback || null);
1525
+ taskAdapterMap.delete(taskRecord.id);
1526
+ return;
1527
+ }
1528
+ }
1529
+ }
1530
+ }
1531
+ else {
1532
+ // panel: 3 concurrent reviewer calls, majority vote
1533
+ await reviewSemaphore.acquire();
1534
+ let panelResults;
1535
+ try {
1536
+ const noopRunner = (_t, _l, m) => Promise.resolve({ verdict: 'pass', feedback: '', tokens: 0, model: m });
1537
+ const runner = reviewRunner ?? noopRunner;
1538
+ panelResults = await Promise.all([
1539
+ runner(taskRecord, 'panel', reviewerModel),
1540
+ runner(taskRecord, 'panel', reviewerModel),
1541
+ runner(taskRecord, 'panel', reviewerModel),
1542
+ ]);
1543
+ }
1544
+ finally {
1545
+ reviewSemaphore.release();
1546
+ }
1547
+ const panelPasses = panelResults.filter(r => r.verdict === 'pass').length;
1548
+ const panelBlocks = panelResults.filter(r => r.verdict === 'block').length;
1549
+ const totalPanelTokens = panelResults.reduce((sum, r) => sum + r.tokens, 0);
1550
+ reviewTokensTotal += totalPanelTokens;
1551
+ const freshForPanel = store.getTask(taskRecord.id, convoyId);
1552
+ store.updateTaskReview(taskRecord.id, convoyId, {
1553
+ review_level: 'panel',
1554
+ review_verdict: panelPasses >= 2 ? 'pass' : 'block',
1555
+ review_tokens: totalPanelTokens,
1556
+ review_model: reviewerModel,
1557
+ panel_attempts: freshForPanel.panel_attempts + 1,
1558
+ });
1559
+ if (totalPanelTokens > 0)
1560
+ store.updateConvoyReviewTokens(convoyId, reviewTokensTotal);
1561
+ events.emit('review_verdict', { level: 'panel', verdict: panelPasses >= 2 ? 'pass' : 'block', tokens: totalPanelTokens, model: reviewerModel, feedback_length: panelResults.map(r => r.feedback).join('').length, passes: panelPasses, blocks: panelBlocks }, { convoy_id: convoyId, task_id: taskRecord.id });
1562
+ if (panelBlocks >= 2) {
1563
+ const blockFeedback = panelResults.filter(r => r.verdict === 'block').map(r => r.feedback).join('\n\n---\n\n');
1564
+ await removeWorktree();
1565
+ // Check for dispute trigger
1566
+ const updatedTask = store.getTask(taskRecord.id, convoyId);
1567
+ if (updatedTask.panel_attempts >= 3) {
1568
+ const disputeId = `dispute-${taskRecord.id}-${Date.now()}`;
1569
+ const onDispute = spec.defaults?.on_dispute ?? 'stop';
1570
+ store.updateTaskDisputeStatus(taskRecord.id, convoyId, 'disputed', disputeId);
1571
+ writeDisputeToMarkdown(disputeId, convoyId, taskRecord, panelResults, events);
1572
+ events.emit('dispute_opened', {
1573
+ dispute_id: disputeId,
1574
+ task_id: taskRecord.id,
1575
+ agent: taskRecord.agent,
1576
+ panel_attempts: updatedTask.panel_attempts,
1577
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1578
+ if (onDispute === 'stop') {
1579
+ const allPending = store.getTasksByConvoy(convoyId).filter(t => t.status === 'pending');
1580
+ for (const t of allPending) {
1581
+ skipTask(t.id, `on_dispute: stop — task "${taskRecord.id}" disputed`);
1582
+ }
1583
+ }
1584
+ completedCount++;
1585
+ process.stdout.write(` ${c.red('⚡')} ${c.bold(`[${taskRecord.id}]`)} disputed after ${updatedTask.panel_attempts} panel attempts\n`);
1586
+ taskAdapterMap.delete(taskRecord.id);
1587
+ return;
1588
+ }
1589
+ const freshRecord = store.getTask(taskRecord.id, convoyId);
1590
+ if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1591
+ const contextPrefix = `Previous attempt was blocked by panel review (${panelBlocks}/3 reviewers).\nMUST-FIX:\n${blockFeedback}\n\nFix the issues and try again.\n\n`;
1592
+ store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
1593
+ retries: freshRecord.retries + 1,
1594
+ worker_id: null,
1595
+ worktree: null,
1596
+ started_at: null,
1597
+ finished_at: null,
1598
+ prompt: contextPrefix + taskRecord.prompt,
1599
+ });
1600
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1601
+ process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} panel blocked (${panelBlocks}/3), retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
1602
+ taskAdapterMap.delete(taskRecord.id);
1603
+ return;
1604
+ }
1605
+ else {
1606
+ store.withTransaction(() => {
1607
+ store.updateTaskStatus(taskRecord.id, convoyId, 'review-blocked', {
1608
+ finished_at: finishedAt,
1609
+ output: `Panel review blocked (${panelBlocks}/3): ${blockFeedback}`,
1610
+ exit_code: 1,
1611
+ });
1612
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1613
+ });
1614
+ completedCount++;
1615
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} panel blocked ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1616
+ events.emit('task_failed', { reason: 'review-blocked', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1617
+ handleExhaustion(freshRecord, 'review-blocked', blockFeedback || null);
1618
+ taskAdapterMap.delete(taskRecord.id);
1619
+ return;
1620
+ }
1621
+ }
1622
+ }
1623
+ }
1624
+ // ── Intelligence: check discovered issues (Phase 18.4) ─────────────
1625
+ if (spec.defaults?.track_discovered_issues) {
1626
+ try {
1627
+ checkDiscoveredIssues(taskRecord.id, events, convoyId, worktreePath ?? basePath);
1628
+ }
1629
+ catch { /* non-critical */ }
1630
+ }
1631
+ // ── post_task hooks ───────────────────────────────────────────────────
1632
+ if (taskHooks.length > 0) {
1633
+ const postResult = await runHooks(taskHooks, 'post_task', {
1634
+ taskId: taskRecord.id,
1635
+ convoyId,
1636
+ cwd: worktreePath ?? basePath,
1637
+ });
1638
+ if (!postResult.passed) {
1639
+ await removeWorktree();
1640
+ const hookLabel = postResult.failedHook?.name ?? postResult.failedHook?.type ?? 'unknown';
1641
+ store.withTransaction(() => {
1642
+ store.updateTaskStatus(taskRecord.id, convoyId, 'hook-failed', {
1643
+ finished_at: finishedAt,
1644
+ output: `post_task hook "${hookLabel}" failed: ${postResult.error ?? ''}`,
1645
+ exit_code: 1,
1646
+ });
1647
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1648
+ });
1649
+ completedCount++;
1650
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} post_task hook failed ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1651
+ events.emit('task_failed', { reason: 'hook-failed', hook: hookLabel, worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1652
+ cascadeFailure(taskRecord.id);
1653
+ taskAdapterMap.delete(taskRecord.id);
1654
+ return;
1655
+ }
1656
+ }
1657
+ // ── Symlink security scan (post-execution) ───────────────────────────
1658
+ if (taskFiles.length > 0 && worktreePath) {
1659
+ try {
1660
+ scanNewSymlinks(worktreePath, taskFiles);
239
1661
  }
240
1662
  catch (err) {
241
- if (verbose) {
242
- process.stderr.write(`Warning: merge failed for ${taskRecord.id}: ${err.message}\n`);
1663
+ await removeWorktree();
1664
+ store.withTransaction(() => {
1665
+ store.updateTaskStatus(taskRecord.id, convoyId, 'failed', {
1666
+ finished_at: finishedAt,
1667
+ output: `Post-execution symlink security check failed: ${err.message}`,
1668
+ exit_code: 1,
1669
+ });
1670
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
1671
+ });
1672
+ completedCount++;
1673
+ events.emit('task_failed', { reason: 'symlink-escape-post', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1674
+ cascadeFailure(taskRecord.id);
1675
+ taskAdapterMap.delete(taskRecord.id);
1676
+ return;
1677
+ }
1678
+ }
1679
+ if (worktreePath) {
1680
+ let mergeAttempt = 0;
1681
+ const maxMergeAttempts = 2;
1682
+ let merged = false;
1683
+ while (mergeAttempt < maxMergeAttempts && !merged) {
1684
+ try {
1685
+ await mergeQueue.merge(worktreePath, `convoy-${workerId}`, baseBranch);
1686
+ merged = true;
1687
+ }
1688
+ catch (err) {
1689
+ if (err instanceof MergeConflictError) {
1690
+ mergeAttempt++;
1691
+ events.emit('merge_conflict_detected', {
1692
+ attempt: mergeAttempt,
1693
+ conflicting_files: err.conflictingFiles,
1694
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1695
+ if (mergeAttempt >= maxMergeAttempts) {
1696
+ events.emit('merge_conflict_failed', {
1697
+ attempts: mergeAttempt,
1698
+ conflicting_files: err.conflictingFiles,
1699
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1700
+ const freshRecord = store.getTask(taskRecord.id, convoyId);
1701
+ store.withTransaction(() => {
1702
+ store.updateTaskStatus(taskRecord.id, convoyId, 'failed', {
1703
+ finished_at: now(),
1704
+ output: `Merge conflict could not be resolved after ${mergeAttempt} attempts. Files: ${err.conflictingFiles.join(', ')}`,
1705
+ exit_code: 1,
1706
+ });
1707
+ store.updateWorkerStatus(workerId, 'failed', { finished_at: now() });
1708
+ });
1709
+ completedCount++;
1710
+ process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} merge conflict unresolved ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
1711
+ events.emit('task_failed', { reason: 'merge-conflict', worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
1712
+ cascadeFailure(taskRecord.id);
1713
+ handleExhaustion(freshRecord, 'merge-conflict', err.conflictingFiles.join(', '));
1714
+ break;
1715
+ }
1716
+ // Per spec: backoff on second attempt (unreachable with maxMergeAttempts=2 but follows spec)
1717
+ if (mergeAttempt === 2) {
1718
+ await new Promise(resolve => setTimeout(resolve, 30_000));
1719
+ }
1720
+ // Inject a resolution task
1721
+ const fileHash = createHash('sha256')
1722
+ .update(err.conflictingFiles.sort().join(','))
1723
+ .digest('hex')
1724
+ .slice(0, 12);
1725
+ const idempotencyKey = `merge-conflict:${taskRecord.phase}:${fileHash}`;
1726
+ const resolutionTaskId = `merge-fix-${taskRecord.id}-${mergeAttempt}`;
1727
+ const conflictPrompt = `Resolve merge conflicts in: ${err.conflictingFiles.join(', ')}. Ensure no conflict markers remain (<<<<<<<, =======, >>>>>>>), syntax is valid, no duplicate imports.`;
1728
+ const resolutionRecord = {
1729
+ id: resolutionTaskId,
1730
+ convoy_id: convoyId,
1731
+ phase: taskRecord.phase,
1732
+ prompt: conflictPrompt,
1733
+ agent: taskRecord.agent,
1734
+ adapter: null,
1735
+ model: null,
1736
+ timeout_ms: 600_000,
1737
+ status: 'pending',
1738
+ worker_id: null,
1739
+ worktree: null,
1740
+ output: null,
1741
+ exit_code: null,
1742
+ started_at: null,
1743
+ finished_at: null,
1744
+ retries: 0,
1745
+ max_retries: 1,
1746
+ files: JSON.stringify(err.conflictingFiles),
1747
+ depends_on: null,
1748
+ prompt_tokens: null,
1749
+ completion_tokens: null,
1750
+ total_tokens: null,
1751
+ cost_usd: null,
1752
+ gates: null,
1753
+ on_exhausted: 'dlq',
1754
+ injected: 1,
1755
+ provenance: 'merge-conflict',
1756
+ idempotency_key: idempotencyKey,
1757
+ current_step: null,
1758
+ total_steps: null,
1759
+ review_level: null,
1760
+ review_verdict: null,
1761
+ review_tokens: null,
1762
+ review_model: null,
1763
+ panel_attempts: 0,
1764
+ dispute_id: null,
1765
+ drift_score: null,
1766
+ drift_retried: 0,
1767
+ outputs: null,
1768
+ inputs: null,
1769
+ discovered_issues: null,
1770
+ };
1771
+ store.insertInjectedTask(resolutionRecord);
1772
+ const storedResolutionRecord = store.getTask(resolutionTaskId, convoyId);
1773
+ await executeOneTask(storedResolutionRecord);
1774
+ // Next loop iteration will retry the merge
1775
+ }
1776
+ else {
1777
+ // Non-conflict merge error — log warning and continue to done path
1778
+ if (verbose) {
1779
+ process.stderr.write(`Warning: merge failed for ${taskRecord.id}: ${err.message}\n`);
1780
+ }
1781
+ merged = true; // Preserve original behavior: continue despite error
1782
+ break;
1783
+ }
243
1784
  }
244
1785
  }
245
1786
  await removeWorktree();
1787
+ if (!merged) {
1788
+ taskAdapterMap.delete(taskRecord.id);
1789
+ return;
1790
+ }
1791
+ // ── Intelligence: update expertise post-merge (Phase 18.2) ─────────
1792
+ try {
1793
+ updateExpertise(taskRecord.agent, { taskId: taskRecord.id, success: true, retries: taskRecord.retries, files: taskRecord.files ? JSON.parse(taskRecord.files) : [] }, basePath);
1794
+ }
1795
+ catch { /* non-critical */ }
1796
+ // ── Intelligence: build knowledge graph post-merge (Phase 18.3) ────
1797
+ try {
1798
+ const { stdout: diffOut } = await execFile('git', ['diff', 'HEAD~1'], { cwd: basePath });
1799
+ buildKnowledgeGraph(diffOut, convoyId, basePath);
1800
+ }
1801
+ catch { /* non-critical */ }
246
1802
  }
247
1803
  const usageExtra = {};
248
1804
  if (result.usage) {
@@ -253,6 +1809,80 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
253
1809
  if (result.usage.total_tokens != null)
254
1810
  usageExtra.total_tokens = result.usage.total_tokens;
255
1811
  }
1812
+ // ── Capture outputs as artifacts ────────────────────────────────────────
1813
+ if (taskRecord.outputs) {
1814
+ const outputs = JSON.parse(taskRecord.outputs);
1815
+ for (const output of outputs) {
1816
+ let content;
1817
+ if (output.type === 'summary') {
1818
+ content = result.output.slice(-4096);
1819
+ }
1820
+ else if (output.type === 'json') {
1821
+ const jsonMatch = result.output.match(/```json\n([\s\S]*?)```/);
1822
+ content = jsonMatch ? jsonMatch[1].trim() : result.output;
1823
+ }
1824
+ else {
1825
+ content = result.output;
1826
+ }
1827
+ try {
1828
+ store.insertArtifact({
1829
+ id: `artifact-${taskRecord.id}-${output.name}-${Date.now()}`,
1830
+ convoy_id: convoyId,
1831
+ task_id: taskRecord.id,
1832
+ name: output.name,
1833
+ type: output.type,
1834
+ content,
1835
+ created_at: new Date().toISOString(),
1836
+ });
1837
+ }
1838
+ catch (err) {
1839
+ if (err instanceof ConvoyArtifactLimitError) {
1840
+ events.emit('artifact_limit_reached', {
1841
+ task_id: taskRecord.id,
1842
+ artifact_name: output.name,
1843
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1844
+ }
1845
+ else {
1846
+ throw err;
1847
+ }
1848
+ }
1849
+ }
1850
+ }
1851
+ // ── Intelligence: capture persistent agent identity (Phase 17.2) ─────
1852
+ const specTaskForCapture = (spec.tasks ?? []).find(t => t.id === taskRecord.id);
1853
+ if (specTaskForCapture?.persistent && result.output) {
1854
+ try {
1855
+ // Extract last 300 words, cap at 4KB
1856
+ const words = result.output.split(/\s+/);
1857
+ const lastWords = words.slice(-300).join(' ');
1858
+ let summary = lastWords.length > 4096 ? lastWords.slice(-4096) : lastWords;
1859
+ // Secret-scan the summary before storing
1860
+ const summaryScan = scanForSecrets(summary, `identity:${taskRecord.id}`);
1861
+ if (summaryScan.clean) {
1862
+ store.insertAgentIdentity({
1863
+ id: `identity-${taskRecord.id}-${Date.now()}`,
1864
+ agent: taskRecord.agent,
1865
+ convoy_id: convoyId,
1866
+ task_id: taskRecord.id,
1867
+ summary,
1868
+ created_at: new Date().toISOString(),
1869
+ retention_days: 90,
1870
+ });
1871
+ events.emit('agent_identity_captured', {
1872
+ agent: taskRecord.agent,
1873
+ summary_length: summary.length,
1874
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1875
+ }
1876
+ else {
1877
+ events.emit('agent_identity_rejected', {
1878
+ agent: taskRecord.agent,
1879
+ reason: 'secrets_detected',
1880
+ findings_count: summaryScan.findings.length,
1881
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1882
+ }
1883
+ }
1884
+ catch { /* non-critical */ }
1885
+ }
256
1886
  store.withTransaction(() => {
257
1887
  store.updateTaskStatus(taskRecord.id, convoyId, 'done', {
258
1888
  finished_at: finishedAt,
@@ -262,6 +1892,28 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
262
1892
  });
263
1893
  store.updateWorkerStatus(workerId, 'done', { finished_at: finishedAt });
264
1894
  });
1895
+ // ── Circuit breaker: record success ────────────────────────────────────
1896
+ if (circuitBreakerConfig) {
1897
+ circuitBreaker.recordSuccess(taskRecord.agent);
1898
+ try {
1899
+ store.updateConvoyCircuitState(convoyId, circuitBreaker.serialize());
1900
+ }
1901
+ catch { /* non-critical */ }
1902
+ }
1903
+ // ── Intelligence: capture retry lesson (Phase 18.1) ─────────────────
1904
+ if (taskRecord.retries > 0 && spec.defaults?.inject_lessons !== false) {
1905
+ try {
1906
+ captureLessons({
1907
+ title: `Retry success for ${taskRecord.agent} on ${taskRecord.id}`,
1908
+ category: 'convoy',
1909
+ agent: taskRecord.agent,
1910
+ problem: `Task ${taskRecord.id} required ${taskRecord.retries} retries`,
1911
+ solution: 'Succeeded after retry with adjusted approach',
1912
+ files: taskRecord.files ? JSON.parse(taskRecord.files) : undefined,
1913
+ }, basePath);
1914
+ }
1915
+ catch { /* non-critical */ }
1916
+ }
265
1917
  completedCount++;
266
1918
  process.stdout.write(` ${c.green('✓')} ${c.bold(`[${taskRecord.id}]`)} ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
267
1919
  events.emit('task_done', { exit_code: result.exitCode, worker_id: workerId }, { convoy_id: convoyId, task_id: taskRecord.id, worker_id: workerId });
@@ -295,12 +1947,15 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
295
1947
  await removeWorktree();
296
1948
  const freshRecord = store.getTask(taskRecord.id, convoyId);
297
1949
  if (freshRecord.retries < freshRecord.max_retries && spec.on_failure !== 'stop') {
1950
+ const failedOutput = result.output || '(no output)';
1951
+ const contextPrefix = `Previous attempt failed.\nExit code: ${result.exitCode}\nError output:\n${failedOutput}\n\nFix the issues and try again.\n\n`;
298
1952
  store.updateTaskStatus(taskRecord.id, convoyId, 'pending', {
299
1953
  retries: freshRecord.retries + 1,
300
1954
  worker_id: null,
301
1955
  worktree: null,
302
1956
  started_at: null,
303
1957
  finished_at: null,
1958
+ prompt: contextPrefix + taskRecord.prompt,
304
1959
  });
305
1960
  store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
306
1961
  process.stdout.write(` ${c.yellow('⟳')} ${c.bold(`[${taskRecord.id}]`)} retry ${freshRecord.retries + 1}/${freshRecord.max_retries}\n`);
@@ -314,6 +1969,25 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
314
1969
  });
315
1970
  store.updateWorkerStatus(workerId, 'failed', { finished_at: finishedAt });
316
1971
  });
1972
+ // ── Intelligence: record failure in expertise (Phase 18.2) ──────────
1973
+ try {
1974
+ updateExpertise(taskRecord.agent, { taskId: taskRecord.id, success: false, retries: freshRecord.retries, files: taskRecord.files ? JSON.parse(taskRecord.files) : [] }, basePath);
1975
+ }
1976
+ catch { /* non-critical */ }
1977
+ // ── Circuit breaker: record failure ────────────────────────────────────
1978
+ if (circuitBreakerConfig) {
1979
+ const { tripped } = circuitBreaker.recordFailure(taskRecord.agent);
1980
+ try {
1981
+ store.updateConvoyCircuitState(convoyId, circuitBreaker.serialize());
1982
+ }
1983
+ catch { /* non-critical */ }
1984
+ if (tripped) {
1985
+ events.emit('circuit_breaker_tripped', {
1986
+ agent: taskRecord.agent,
1987
+ state: circuitBreaker.getState(taskRecord.agent),
1988
+ }, { convoy_id: convoyId, task_id: taskRecord.id });
1989
+ }
1990
+ }
317
1991
  completedCount++;
318
1992
  process.stdout.write(` ${c.red('✗')} ${c.bold(`[${taskRecord.id}]`)} failed ${elapsed} ${c.dim(`[${completedCount}/${totalTasks}]`)}\n`);
319
1993
  if (verbose) {
@@ -342,26 +2016,52 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
342
2016
  phase: taskRecord.phase,
343
2017
  convoy_id: convoyId,
344
2018
  }, { convoy_id: convoyId, task_id: taskRecord.id });
345
- cascadeFailure(taskRecord.id);
2019
+ handleExhaustion(freshRecord, 'error', result.output || null);
346
2020
  }
347
2021
  taskAdapterMap.delete(taskRecord.id);
348
2022
  }
349
2023
  // ── Main execution loop ───────────────────────────────────────────────────
350
2024
  let lastPhase = -1;
2025
+ const isSwarmMode = spec.concurrency === 'auto';
2026
+ const maxSwarmConcurrency = spec.defaults?.max_swarm_concurrency ?? 8;
2027
+ let lastInjectPoll = 0;
2028
+ const INJECT_POLL_INTERVAL = 2000; // 2 seconds
351
2029
  try {
352
2030
  let ready = store.getReadyTasks(convoyId);
353
- const concurrency = spec.concurrency ?? 1;
354
2031
  while (ready.length > 0) {
2032
+ // Compute effective concurrency for this phase
2033
+ const effectiveConcurrency = isSwarmMode
2034
+ ? Math.min(ready.length, maxSwarmConcurrency)
2035
+ : (typeof spec.concurrency === 'number' ? spec.concurrency : 1);
355
2036
  for (const t of ready) {
356
2037
  if (t.phase !== lastPhase) {
357
2038
  lastPhase = t.phase;
358
2039
  const tasksInPhase = ready.filter(r => r.phase === t.phase);
359
2040
  const ids = tasksInPhase.map(r => r.id).join(', ');
360
2041
  process.stdout.write(`\n ${c.bold(`Phase ${t.phase + 1}:`)} ${c.dim(ids)}\n`);
2042
+ if (isSwarmMode) {
2043
+ events.emit('swarm_concurrency_update', {
2044
+ phase: t.phase,
2045
+ pending_count: ready.length,
2046
+ effective_concurrency: effectiveConcurrency,
2047
+ }, { convoy_id: convoyId });
2048
+ }
361
2049
  }
362
2050
  }
363
- for (let i = 0; i < ready.length; i += concurrency) {
364
- await Promise.all(ready.slice(i, i + concurrency).map(t => executeOneTask(t)));
2051
+ for (let i = 0; i < ready.length; i += effectiveConcurrency) {
2052
+ // Poll for file-based injection between batches
2053
+ const now = Date.now();
2054
+ if (now - lastInjectPoll >= INJECT_POLL_INTERVAL) {
2055
+ pollInjectFile(convoyId, store, events, basePath);
2056
+ lastInjectPoll = now;
2057
+ }
2058
+ await Promise.all(ready.slice(i, i + effectiveConcurrency).map(t => executeOneTask(t)));
2059
+ }
2060
+ // Reset wait-for-input tasks to pending so they are re-evaluated after
2061
+ // upstream artifacts may have been captured in this batch
2062
+ const waitingTasks = store.getTasksByConvoy(convoyId).filter(t => t.status === 'wait-for-input');
2063
+ for (const wt of waitingTasks) {
2064
+ store.updateTaskStatus(wt.id, convoyId, 'pending');
365
2065
  }
366
2066
  ready = store.getReadyTasks(convoyId);
367
2067
  }
@@ -380,6 +2080,8 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
380
2080
  process.stdout.write(`\n ${c.bold(gateAttempt === 0 ? 'Gates:' : `Gates (retry ${gateAttempt}/${maxGateRetries}):`)}\n`);
381
2081
  for (const command of spec.gates) {
382
2082
  try {
2083
+ // SECURITY: Gate/hook commands come from the .convoy.yml spec file, which is operator-controlled.
2084
+ // They are NOT user-supplied and are part of the trusted build configuration.
383
2085
  await execFile('sh', ['-c', command], { cwd: basePath });
384
2086
  gateResults.push({ command, exitCode: 0, passed: true });
385
2087
  process.stdout.write(` ${c.green('✓')} ${c.dim(command)}\n`);
@@ -425,12 +2127,41 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
425
2127
  break; // Don't retry if the fix task itself fails
426
2128
  }
427
2129
  }
2130
+ // ── post_convoy hooks ─────────────────────────────────────────────────────
2131
+ const specLevelHooks = spec.hooks ?? [];
2132
+ if (specLevelHooks.length > 0) {
2133
+ const postConvoyResult = await runHooks(specLevelHooks, 'post_convoy', {
2134
+ convoyId,
2135
+ cwd: basePath,
2136
+ });
2137
+ if (!postConvoyResult.passed) {
2138
+ const hookLabel = postConvoyResult.failedHook?.name ?? postConvoyResult.failedHook?.type ?? 'unknown';
2139
+ events.emit('post_convoy_hook_failed', {
2140
+ hook: hookLabel,
2141
+ error: postConvoyResult.error,
2142
+ }, { convoy_id: convoyId });
2143
+ process.stdout.write(` ${c.red('✗')} post_convoy hook "${hookLabel}" failed\n`);
2144
+ }
2145
+ }
2146
+ // ── Intelligence: post-convoy consolidation ──────────────────────────────
2147
+ if (spec.defaults?.inject_lessons !== false) {
2148
+ try {
2149
+ consolidateLessons(basePath);
2150
+ }
2151
+ catch { /* non-critical */ }
2152
+ }
2153
+ if (spec.defaults?.track_discovered_issues) {
2154
+ try {
2155
+ consolidateIssues(basePath);
2156
+ }
2157
+ catch { /* non-critical */ }
2158
+ }
428
2159
  // ── Final status & summary ────────────────────────────────────────────────
429
2160
  const allTasksFinal = store.getTasksByConvoy(convoyId);
430
2161
  const summary = {
431
2162
  total: allTasksFinal.length,
432
2163
  done: allTasksFinal.filter(t => t.status === 'done').length,
433
- failed: allTasksFinal.filter(t => t.status === 'failed').length,
2164
+ failed: allTasksFinal.filter(t => t.status === 'failed' || t.status === 'gate-failed' || t.status === 'review-blocked' || t.status === 'disputed').length,
434
2165
  skipped: allTasksFinal.filter(t => t.status === 'skipped').length,
435
2166
  timedOut: allTasksFinal.filter(t => t.status === 'timed-out').length,
436
2167
  };
@@ -451,6 +2182,24 @@ async function runConvoy(convoyId, spec, adapter, store, events, wtManager, merg
451
2182
  finished_at: new Date().toISOString(),
452
2183
  total_tokens: convoyTotalTokens,
453
2184
  });
2185
+ if (finalStatus === 'done') {
2186
+ events.emit('convoy_finished', { status: 'done' }, { convoy_id: convoyId });
2187
+ }
2188
+ else {
2189
+ events.emit('convoy_failed', { status: finalStatus, reason: finalStatus === 'gate-failed' ? 'Gate check failed' : 'One or more tasks failed' }, { convoy_id: convoyId });
2190
+ }
2191
+ // Run convoy guard checks
2192
+ const guardResult = runConvoyGuard(store, convoyId, wtManager, ndjsonPath, spec.guard);
2193
+ if (guardResult.warnings.length > 0) {
2194
+ process.stdout.write(`\n ${c.yellow('Guard warnings:')}\n`);
2195
+ for (const w of guardResult.warnings) {
2196
+ process.stdout.write(` ${c.dim('⚠')} ${w}\n`);
2197
+ }
2198
+ events.emit('convoy_guard', {
2199
+ passed: guardResult.passed,
2200
+ warnings: guardResult.warnings,
2201
+ }, { convoy_id: convoyId });
2202
+ }
454
2203
  return {
455
2204
  convoyId,
456
2205
  status: finalStatus,
@@ -481,9 +2230,44 @@ export function createConvoyEngine(options) {
481
2230
  const convoyId = `convoy-${startTime}`;
482
2231
  const specHash = createHash('sha256').update(specYaml).digest('hex');
483
2232
  const baseBranch = spec.branch ?? (await getCurrentBranch());
2233
+ // Ensure target branch exists before acquiring any locks.
2234
+ // Uses _ensureBranch injection so callers/tests can override.
2235
+ if (spec.branch !== undefined) {
2236
+ const branchFn = options._ensureBranch ?? ensureBranch;
2237
+ await branchFn(spec.branch, basePath);
2238
+ }
484
2239
  mkdirSync(dirname(dbPath), { recursive: true });
2240
+ const lockDb = new DatabaseSync(dbPath);
2241
+ lockDb.exec('PRAGMA journal_mode = WAL');
2242
+ lockDb.exec(`CREATE TABLE IF NOT EXISTS engine_lock (
2243
+ id INTEGER PRIMARY KEY CHECK (id = 1),
2244
+ pid INTEGER NOT NULL,
2245
+ hostname TEXT NOT NULL,
2246
+ started_at TEXT NOT NULL,
2247
+ last_heartbeat TEXT NOT NULL
2248
+ )`);
2249
+ const lock = (() => {
2250
+ try {
2251
+ return acquireEngineLock(lockDb, dbPath);
2252
+ }
2253
+ catch (err) {
2254
+ lockDb.close();
2255
+ throw err;
2256
+ }
2257
+ })();
2258
+ const versionRow = lockDb.prepare('SELECT sqlite_version() as v').get();
2259
+ const [major, minor] = versionRow.v.split('.').map(Number);
2260
+ if (major < 3 || (major === 3 && minor < 35)) {
2261
+ lock.release();
2262
+ lockDb.close();
2263
+ throw new Error(`SQLite version ${versionRow.v} is too old. Requires >= 3.35.0`);
2264
+ }
2265
+ lock.startHeartbeat();
485
2266
  const store = createConvoyStore(dbPath);
486
- const events = createEventEmitter(store, options.logsDir);
2267
+ const ndjsonPath = options.logsDir
2268
+ ? join(options.logsDir, 'convoys', `${convoyId}.ndjson`)
2269
+ : ndjsonPathForConvoy(convoyId, basePath);
2270
+ const events = createEventEmitter(store, { ndjsonPath });
487
2271
  const wtManager = options._worktreeManager ?? createWorktreeManager(basePath);
488
2272
  const mergeQueue = options._mergeQueue ?? createMergeQueue(basePath);
489
2273
  let result;
@@ -500,6 +2284,15 @@ export function createConvoyEngine(options) {
500
2284
  });
501
2285
  const tasks = spec.tasks ?? [];
502
2286
  const phases = buildPhases(tasks);
2287
+ // Validate file partitions before inserting tasks
2288
+ const partitionResult = validateFilePartitions(tasks, phases);
2289
+ if (!partitionResult.valid) {
2290
+ const conflictSummary = partitionResult.conflicts
2291
+ .map((cf) => `Phase ${cf.phase}: tasks "${cf.taskA}" and "${cf.taskB}" overlap on [${cf.overlapping.join(', ')}]`)
2292
+ .join('\n');
2293
+ events.emit('file_partition_conflict', { conflicts: partitionResult.conflicts }, { convoy_id: convoyId });
2294
+ throw new Error(`File partition conflicts detected:\n${conflictSummary}`);
2295
+ }
503
2296
  for (let phaseIdx = 0; phaseIdx < phases.length; phaseIdx++) {
504
2297
  for (const task of phases[phaseIdx]) {
505
2298
  store.insertTask({
@@ -516,27 +2309,62 @@ export function createConvoyEngine(options) {
516
2309
  max_retries: task.max_retries,
517
2310
  files: task.files.length > 0 ? JSON.stringify(task.files) : null,
518
2311
  depends_on: task.depends_on.length > 0 ? JSON.stringify(task.depends_on) : null,
2312
+ gates: task.gates && task.gates.length > 0 ? JSON.stringify(task.gates) : null,
2313
+ outputs: task.outputs && task.outputs.length > 0 ? JSON.stringify(task.outputs) : null,
2314
+ inputs: task.inputs && task.inputs.length > 0 ? JSON.stringify(task.inputs) : null,
519
2315
  });
520
2316
  }
521
2317
  }
522
2318
  store.updateConvoyStatus(convoyId, 'running', { started_at: new Date().toISOString() });
523
2319
  events.emit('convoy_started', { name: spec.name }, { convoy_id: convoyId });
524
- result = await runConvoy(convoyId, spec, adapter, store, events, wtManager, mergeQueue, basePath, baseBranch, verbose, startTime);
2320
+ result = await runConvoy(convoyId, spec, adapter, store, events, wtManager, mergeQueue, basePath, baseBranch, verbose, startTime, ndjsonPath, options._reviewRunner);
525
2321
  }
526
2322
  finally {
527
2323
  try {
528
2324
  await exportConvoyToNdjson(store, convoyId, options.logsDir);
529
2325
  }
530
2326
  catch { /* silent */ }
2327
+ events.close();
531
2328
  store.close();
2329
+ lock.release();
2330
+ lockDb.close();
532
2331
  }
533
2332
  return result;
534
2333
  }
535
2334
  async function resume(convoyId) {
536
2335
  const startTime = Date.now();
537
2336
  mkdirSync(dirname(dbPath), { recursive: true });
2337
+ const lockDb = new DatabaseSync(dbPath);
2338
+ lockDb.exec('PRAGMA journal_mode = WAL');
2339
+ lockDb.exec(`CREATE TABLE IF NOT EXISTS engine_lock (
2340
+ id INTEGER PRIMARY KEY CHECK (id = 1),
2341
+ pid INTEGER NOT NULL,
2342
+ hostname TEXT NOT NULL,
2343
+ started_at TEXT NOT NULL,
2344
+ last_heartbeat TEXT NOT NULL
2345
+ )`);
2346
+ const lock = (() => {
2347
+ try {
2348
+ return acquireEngineLock(lockDb, dbPath);
2349
+ }
2350
+ catch (err) {
2351
+ lockDb.close();
2352
+ throw err;
2353
+ }
2354
+ })();
2355
+ const versionRow = lockDb.prepare('SELECT sqlite_version() as v').get();
2356
+ const [major, minor] = versionRow.v.split('.').map(Number);
2357
+ if (major < 3 || (major === 3 && minor < 35)) {
2358
+ lock.release();
2359
+ lockDb.close();
2360
+ throw new Error(`SQLite version ${versionRow.v} is too old. Requires >= 3.35.0`);
2361
+ }
2362
+ lock.startHeartbeat();
538
2363
  const store = createConvoyStore(dbPath);
539
- const events = createEventEmitter(store, options.logsDir);
2364
+ const ndjsonPath = options.logsDir
2365
+ ? join(options.logsDir, 'convoys', `${convoyId}.ndjson`)
2366
+ : ndjsonPathForConvoy(convoyId, basePath);
2367
+ const events = createEventEmitter(store, { ndjsonPath });
540
2368
  const wtManager = options._worktreeManager ?? createWorktreeManager(basePath);
541
2369
  const mergeQueue = options._mergeQueue ?? createMergeQueue(basePath);
542
2370
  let result;
@@ -570,18 +2398,198 @@ export function createConvoyEngine(options) {
570
2398
  }
571
2399
  // Remove all orphaned worktrees from the crashed run
572
2400
  await wtManager.removeAll();
2401
+ // NDJSON recovery: truncate partial lines, replay missing events
2402
+ recoverNdjson(store, convoyId, ndjsonPath);
573
2403
  events.emit('convoy_resumed', { original_created_at: convoy.created_at }, { convoy_id: convoyId });
574
- result = await runConvoy(convoyId, spec, adapter, store, events, wtManager, mergeQueue, basePath, baseBranch, verbose, startTime);
2404
+ result = await runConvoy(convoyId, spec, adapter, store, events, wtManager, mergeQueue, basePath, baseBranch, verbose, startTime, ndjsonPath, options._reviewRunner);
575
2405
  }
576
2406
  finally {
577
2407
  try {
578
2408
  await exportConvoyToNdjson(store, convoyId, options.logsDir);
579
2409
  }
580
2410
  catch { /* silent */ }
2411
+ events.close();
581
2412
  store.close();
2413
+ lock.release();
2414
+ lockDb.close();
582
2415
  }
583
2416
  return result;
584
2417
  }
585
- return { run, resume };
2418
+ async function retryFailed(convoyId, taskIds) {
2419
+ mkdirSync(dirname(dbPath), { recursive: true });
2420
+ const store = createConvoyStore(dbPath);
2421
+ const ndjsonPath = options.logsDir
2422
+ ? join(options.logsDir, 'convoys', `${convoyId}.ndjson`)
2423
+ : ndjsonPathForConvoy(convoyId, basePath);
2424
+ const events = createEventEmitter(store, { ndjsonPath });
2425
+ try {
2426
+ const allTasks = store.getTasksByConvoy(convoyId);
2427
+ const retryableStatuses = ['failed', 'gate-failed', 'timed-out', 'review-blocked', 'disputed'];
2428
+ const tasksToRetry = allTasks.filter(t => {
2429
+ if (!retryableStatuses.includes(t.status))
2430
+ return false;
2431
+ if (taskIds && taskIds.length > 0)
2432
+ return taskIds.includes(t.id);
2433
+ return true;
2434
+ });
2435
+ for (const task of tasksToRetry) {
2436
+ store.updateTaskStatus(task.id, convoyId, 'pending', {
2437
+ worker_id: null,
2438
+ worktree: null,
2439
+ started_at: null,
2440
+ finished_at: null,
2441
+ });
2442
+ events.emit('task_retried', { previous_status: task.status }, { convoy_id: convoyId, task_id: task.id });
2443
+ }
2444
+ // Reset convoy status to running so resume can pick it up
2445
+ store.updateConvoyStatus(convoyId, 'running', {});
2446
+ }
2447
+ finally {
2448
+ events.close();
2449
+ store.close();
2450
+ }
2451
+ }
2452
+ function injectTask(convoyId, task) {
2453
+ mkdirSync(dirname(dbPath), { recursive: true });
2454
+ const store = createConvoyStore(dbPath);
2455
+ try {
2456
+ // Idempotency check
2457
+ if (task.idempotency_key) {
2458
+ const existing = store.getTaskByIdempotencyKey(convoyId, task.idempotency_key);
2459
+ if (existing)
2460
+ return existing;
2461
+ }
2462
+ const allTasks = store.getTasksByConvoy(convoyId);
2463
+ // Check max injectable tasks (10)
2464
+ const injectedCount = allTasks.filter(t => t.injected === 1).length;
2465
+ if (injectedCount >= 10) {
2466
+ throw new Error(`Max injectable tasks (10) reached for convoy ${convoyId}`);
2467
+ }
2468
+ // Validate ID uniqueness
2469
+ if (allTasks.some(t => t.id === task.id)) {
2470
+ throw new Error(`Task ID "${task.id}" already exists in convoy ${convoyId}`);
2471
+ }
2472
+ // Validate depends_on references exist
2473
+ const deps = task.depends_on ?? [];
2474
+ for (const dep of deps) {
2475
+ if (!allTasks.some(t => t.id === dep)) {
2476
+ throw new Error(`Dependency "${dep}" not found in convoy ${convoyId}`);
2477
+ }
2478
+ }
2479
+ // Validate no file partition overlap with pending/running tasks
2480
+ const taskFiles = task.files ?? [];
2481
+ if (taskFiles.length > 0) {
2482
+ // Normalize injected task file paths
2483
+ const normalizedTaskFiles = taskFiles.map(normalizePath);
2484
+ // Symlink pre-scan on injected files
2485
+ const basePath = options.basePath ?? process.cwd();
2486
+ try {
2487
+ scanSymlinks(normalizedTaskFiles, basePath);
2488
+ }
2489
+ catch (err) {
2490
+ throw new Error(`Injected task "${task.id}" failed symlink check: ${err.message}`);
2491
+ }
2492
+ // Full partition validation against active tasks
2493
+ const activeTasks = allTasks.filter(t => t.status === 'pending' || t.status === 'running' || t.status === 'assigned');
2494
+ for (const other of activeTasks) {
2495
+ const otherFiles = other.files ? JSON.parse(other.files) : [];
2496
+ if (otherFiles.length === 0)
2497
+ continue;
2498
+ const normalizedOther = otherFiles.map(normalizePath);
2499
+ const overlapping = [];
2500
+ for (const fileA of normalizedTaskFiles) {
2501
+ for (const fileB of normalizedOther) {
2502
+ if (pathsOverlap(fileA, fileB) && !overlapping.includes(fileA)) {
2503
+ overlapping.push(fileA);
2504
+ }
2505
+ }
2506
+ }
2507
+ if (overlapping.length > 0) {
2508
+ throw new Error(`File partition overlap with task "${other.id}": ${overlapping.join(', ')}`);
2509
+ }
2510
+ }
2511
+ }
2512
+ // Detect dependency cycles
2513
+ const depGraph = new Map();
2514
+ for (const t of allTasks) {
2515
+ depGraph.set(t.id, t.depends_on ? JSON.parse(t.depends_on) : []);
2516
+ }
2517
+ depGraph.set(task.id, deps);
2518
+ function hasCycle(nodeId, visited, stack) {
2519
+ visited.add(nodeId);
2520
+ stack.add(nodeId);
2521
+ for (const dep of depGraph.get(nodeId) ?? []) {
2522
+ if (!visited.has(dep)) {
2523
+ if (hasCycle(dep, visited, stack))
2524
+ return true;
2525
+ }
2526
+ else if (stack.has(dep)) {
2527
+ return true;
2528
+ }
2529
+ }
2530
+ stack.delete(nodeId);
2531
+ return false;
2532
+ }
2533
+ const visited = new Set();
2534
+ const stack = new Set();
2535
+ for (const nodeId of depGraph.keys()) {
2536
+ if (!visited.has(nodeId)) {
2537
+ if (hasCycle(nodeId, visited, stack)) {
2538
+ throw new Error(`Dependency cycle detected when injecting task "${task.id}"`);
2539
+ }
2540
+ }
2541
+ }
2542
+ // Insert the task
2543
+ const record = {
2544
+ id: task.id,
2545
+ convoy_id: convoyId,
2546
+ phase: task.phase,
2547
+ prompt: task.prompt,
2548
+ agent: task.agent,
2549
+ adapter: null,
2550
+ model: null,
2551
+ timeout_ms: task.timeout_ms ?? 1_800_000,
2552
+ status: 'pending',
2553
+ worker_id: null,
2554
+ worktree: null,
2555
+ output: null,
2556
+ exit_code: null,
2557
+ started_at: null,
2558
+ finished_at: null,
2559
+ retries: 0,
2560
+ max_retries: task.max_retries ?? 1,
2561
+ files: taskFiles.length > 0 ? JSON.stringify(taskFiles) : null,
2562
+ depends_on: deps.length > 0 ? JSON.stringify(deps) : null,
2563
+ prompt_tokens: null,
2564
+ completion_tokens: null,
2565
+ total_tokens: null,
2566
+ cost_usd: null,
2567
+ gates: null,
2568
+ on_exhausted: task.on_exhausted ?? 'dlq',
2569
+ injected: 1,
2570
+ provenance: task.provenance ?? null,
2571
+ idempotency_key: task.idempotency_key ?? null,
2572
+ current_step: null,
2573
+ total_steps: null,
2574
+ review_level: null,
2575
+ review_verdict: null,
2576
+ review_tokens: null,
2577
+ review_model: null,
2578
+ panel_attempts: 0,
2579
+ dispute_id: null,
2580
+ drift_score: null,
2581
+ drift_retried: 0,
2582
+ outputs: null,
2583
+ inputs: null,
2584
+ discovered_issues: null,
2585
+ };
2586
+ store.insertInjectedTask(record);
2587
+ return record;
2588
+ }
2589
+ finally {
2590
+ store.close();
2591
+ }
2592
+ }
2593
+ return { run, resume, retryFailed, injectTask };
586
2594
  }
587
2595
  //# sourceMappingURL=engine.js.map