@hasna/loops 0.3.60 → 0.4.0

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 (56) hide show
  1. package/CHANGELOG.md +247 -0
  2. package/LICENSE +197 -13
  3. package/README.md +24 -56
  4. package/dist/cli/index.js +7474 -6295
  5. package/dist/daemon/control.d.ts +21 -1
  6. package/dist/daemon/daemon.d.ts +5 -0
  7. package/dist/daemon/index.js +2808 -1209
  8. package/dist/daemon/install.d.ts +1 -1
  9. package/dist/index.d.ts +21 -8
  10. package/dist/index.js +5305 -4293
  11. package/dist/lib/accounts.d.ts +6 -1
  12. package/dist/lib/agent-adapter.d.ts +74 -0
  13. package/dist/lib/backup.d.ts +25 -0
  14. package/dist/lib/errors.d.ts +21 -0
  15. package/dist/lib/executor.d.ts +10 -1
  16. package/dist/lib/goal/metadata.d.ts +16 -0
  17. package/dist/lib/goal/model-factory.d.ts +1 -0
  18. package/dist/lib/goal/prompts.d.ts +3 -1
  19. package/dist/lib/goal/types.d.ts +3 -0
  20. package/dist/lib/health.d.ts +1 -1
  21. package/dist/lib/ids.d.ts +8 -1
  22. package/dist/lib/machines.d.ts +6 -0
  23. package/dist/lib/process-identity.d.ts +16 -0
  24. package/dist/lib/{schedule.d.ts → recurrence.d.ts} +1 -0
  25. package/dist/lib/redact.d.ts +21 -0
  26. package/dist/lib/route/cursors.d.ts +18 -0
  27. package/dist/lib/route/drain.d.ts +6 -0
  28. package/dist/lib/route/fields.d.ts +27 -0
  29. package/dist/lib/route/gates.d.ts +26 -0
  30. package/dist/lib/route/index.d.ts +18 -0
  31. package/dist/lib/route/options.d.ts +31 -0
  32. package/dist/lib/route/parse.d.ts +8 -0
  33. package/dist/lib/route/pr-review.d.ts +11 -0
  34. package/dist/lib/route/provider.d.ts +41 -0
  35. package/dist/lib/route/route-event.d.ts +23 -0
  36. package/dist/lib/route/route-tasks.d.ts +75 -0
  37. package/dist/lib/route/throttle.d.ts +47 -0
  38. package/dist/lib/route/todos-cli.d.ts +21 -0
  39. package/dist/lib/route/types.d.ts +88 -0
  40. package/dist/lib/run-artifacts.d.ts +17 -2
  41. package/dist/lib/run-envelope.d.ts +41 -0
  42. package/dist/lib/scheduler.d.ts +78 -2
  43. package/dist/lib/store.d.ts +63 -0
  44. package/dist/lib/store.js +1007 -287
  45. package/dist/lib/template-kit.d.ts +70 -0
  46. package/dist/lib/templates-custom.d.ts +34 -0
  47. package/dist/lib/templates.d.ts +13 -81
  48. package/dist/lib/workflow-runner.d.ts +2 -0
  49. package/dist/mcp/index.d.ts +3 -0
  50. package/dist/mcp/index.js +3216 -1382
  51. package/dist/sdk/index.d.ts +29 -3
  52. package/dist/sdk/index.js +2740 -1041
  53. package/dist/test-helpers.d.ts +37 -0
  54. package/dist/types.d.ts +2 -0
  55. package/docs/USAGE.md +36 -14
  56. package/package.json +6 -3
package/dist/sdk/index.js CHANGED
@@ -1,14 +1,50 @@
1
1
  // @bun
2
2
  // src/lib/store.ts
3
3
  import { Database } from "bun:sqlite";
4
- import { chmodSync, existsSync, mkdirSync as mkdirSync3, mkdtempSync, rmSync } from "fs";
4
+ import { chmodSync, existsSync, mkdirSync as mkdirSync3, mkdtempSync, rmSync as rmSync2 } from "fs";
5
5
  import { tmpdir } from "os";
6
- import { dirname, join as join3 } from "path";
6
+ import { basename as basename2, dirname as dirname2, join as join3 } from "path";
7
+
8
+ // src/lib/errors.ts
9
+ class CodedError extends Error {
10
+ code;
11
+ constructor(code, message) {
12
+ super(message);
13
+ this.name = new.target.name;
14
+ this.code = code;
15
+ }
16
+ }
17
+
18
+ class LoopNotFoundError extends CodedError {
19
+ constructor(idOrName) {
20
+ super("LOOP_NOT_FOUND", `loop not found: ${idOrName}`);
21
+ }
22
+ }
23
+
24
+ class LoopArchivedError extends CodedError {
25
+ constructor(idOrName) {
26
+ super("LOOP_ARCHIVED", `loop is archived: ${idOrName}; unarchive it before modifying`);
27
+ }
28
+ }
29
+
30
+ class AmbiguousNameError extends CodedError {
31
+ constructor(name) {
32
+ super("AMBIGUOUS_NAME", `ambiguous loop name: ${name}; use a loop id`);
33
+ }
34
+ }
35
+
36
+ class ValidationError extends CodedError {
37
+ constructor(message) {
38
+ super("VALIDATION_ERROR", message);
39
+ }
40
+ }
7
41
 
8
42
  // src/lib/ids.ts
9
43
  import { randomBytes } from "crypto";
10
- function genId(length = 12) {
11
- return randomBytes(Math.ceil(length / 2)).toString("hex").slice(0, length);
44
+ var TIME_HEX_LENGTH = 12;
45
+ function genId() {
46
+ const time = Date.now().toString(16).padStart(TIME_HEX_LENGTH, "0");
47
+ return `${time}${randomBytes(10).toString("hex")}`;
12
48
  }
13
49
  function nowIso() {
14
50
  return new Date().toISOString();
@@ -42,7 +78,166 @@ function launchdPlistPath() {
42
78
  return join(homedir(), "Library", "LaunchAgents", "com.hasna.loops.daemon.plist");
43
79
  }
44
80
 
45
- // src/lib/schedule.ts
81
+ // src/lib/process-identity.ts
82
+ import { readFileSync } from "fs";
83
+ import { spawnSync } from "child_process";
84
+ var START_TIME_TOLERANCE_MS = 5000;
85
+ var clockTicksCache;
86
+ function clockTicksPerSecond() {
87
+ if (clockTicksCache !== undefined)
88
+ return clockTicksCache;
89
+ try {
90
+ const run = spawnSync("getconf", ["CLK_TCK"], { encoding: "utf8" });
91
+ const value = Number(run.stdout.trim());
92
+ clockTicksCache = run.status === 0 && Number.isFinite(value) && value > 0 ? value : 100;
93
+ } catch {
94
+ clockTicksCache = 100;
95
+ }
96
+ return clockTicksCache;
97
+ }
98
+ var bootTimeCache;
99
+ var bootTimeResolved = false;
100
+ function bootTimeMs() {
101
+ if (bootTimeResolved)
102
+ return bootTimeCache;
103
+ bootTimeResolved = true;
104
+ try {
105
+ const match = readFileSync("/proc/stat", "utf8").match(/^btime (\d+)$/m);
106
+ bootTimeCache = match ? Number(match[1]) * 1000 : undefined;
107
+ } catch {
108
+ bootTimeCache = undefined;
109
+ }
110
+ return bootTimeCache;
111
+ }
112
+ function procStatFields(path) {
113
+ try {
114
+ const stat = readFileSync(path, "utf8");
115
+ const closeParen = stat.lastIndexOf(")");
116
+ if (closeParen < 0)
117
+ return;
118
+ return stat.slice(closeParen + 2).split(" ");
119
+ } catch {
120
+ return;
121
+ }
122
+ }
123
+ function processStartTimeMs(pid) {
124
+ if (!Number.isInteger(pid) || pid <= 0)
125
+ return;
126
+ if (process.platform === "linux") {
127
+ const fields = procStatFields(`/proc/${pid}/stat`);
128
+ const startTicks = fields ? Number(fields[19]) : Number.NaN;
129
+ const boot = bootTimeMs();
130
+ if (Number.isFinite(startTicks) && boot !== undefined) {
131
+ return boot + Math.round(startTicks / clockTicksPerSecond() * 1000);
132
+ }
133
+ }
134
+ try {
135
+ const run = spawnSync("ps", ["-o", "lstart=", "-p", String(pid)], { encoding: "utf8" });
136
+ if (run.status === 0) {
137
+ const parsed = Date.parse(run.stdout.trim());
138
+ if (Number.isFinite(parsed))
139
+ return parsed;
140
+ }
141
+ } catch {}
142
+ return;
143
+ }
144
+ function sameProcessStart(recorded, actualMs, toleranceMs = START_TIME_TOLERANCE_MS) {
145
+ if (recorded === undefined || actualMs === undefined)
146
+ return true;
147
+ const recordedMs = typeof recorded === "number" ? recorded : Date.parse(recorded);
148
+ if (!Number.isFinite(recordedMs))
149
+ return true;
150
+ return Math.abs(recordedMs - actualMs) <= toleranceMs;
151
+ }
152
+ function verifiedProcessStart(recorded, actualMs, toleranceMs = START_TIME_TOLERANCE_MS) {
153
+ if (recorded === undefined || actualMs === undefined)
154
+ return false;
155
+ const recordedMs = typeof recorded === "number" ? recorded : Date.parse(recorded);
156
+ if (!Number.isFinite(recordedMs))
157
+ return false;
158
+ return Math.abs(recordedMs - actualMs) <= toleranceMs;
159
+ }
160
+
161
+ // src/lib/redact.ts
162
+ var SCRUBBED = "[SCRUBBED]";
163
+ var TOKEN_PATTERNS = [
164
+ /\bsk-ant-[A-Za-z0-9_-]{8,}/g,
165
+ /\bsk-proj-[A-Za-z0-9_-]{8,}/g,
166
+ /\bAKIA[0-9A-Z]{16}\b/g,
167
+ /\bghp_[A-Za-z0-9]{16,}\b/g,
168
+ /\bgithub_pat_[A-Za-z0-9_]{16,}\b/g,
169
+ /\bxox[a-z]-[A-Za-z0-9-]{8,}/g,
170
+ /\beyJ[A-Za-z0-9_-]{10,}\.[A-Za-z0-9_-]{4,}\.[A-Za-z0-9_-]{10,}\b/g
171
+ ];
172
+ var AUTHORIZATION_PATTERN = /(\bAuthorization(?:\\?["'])?\s*[:=]\s*(?:\\?["'])?(?:Bearer|Basic)\s+)[A-Za-z0-9._~+/=-]{16,}/gi;
173
+ var PEM_PATTERN = /-----BEGIN [A-Z0-9 ]*PRIVATE KEY-----[\s\S]*?(?:-----END [A-Z0-9 ]*PRIVATE KEY-----|$)/g;
174
+ var ASSIGNMENT_PATTERN = /([A-Za-z0-9_.-]{0,64}(?:key|token|secret|passwd|password|passphrase|credentials?)(?:\\?["'])?\s*[:=]\s*)(\\"([^"\\\r\n]{12,})\\"|"((?:[^"\\\r\n]|\\[^\r\n]){12,})"|'([^'\\\r\n]{12,})'|([A-Za-z0-9+/_.=-]{16,}))/gi;
175
+ var BENIGN_KEY_NAME_PATTERN = /(?:^|[._-])(?:idempotency|dedupe|route)[_-]?key$/i;
176
+ var ASSIGNMENT_HINTS = ["key", "token", "secret", "pass", "credential"];
177
+ function mayContainAssignment(text) {
178
+ const lowered = text.toLowerCase();
179
+ return ASSIGNMENT_HINTS.some((hint) => lowered.includes(hint));
180
+ }
181
+ function shannonEntropy(value) {
182
+ const counts = new Map;
183
+ for (const char of value)
184
+ counts.set(char, (counts.get(char) ?? 0) + 1);
185
+ let entropy = 0;
186
+ for (const count of counts.values()) {
187
+ const probability = count / value.length;
188
+ entropy -= probability * Math.log2(probability);
189
+ }
190
+ return entropy;
191
+ }
192
+ function isLikelySecretValue(value) {
193
+ if (value.length < 12)
194
+ return false;
195
+ if (value.includes(SCRUBBED))
196
+ return false;
197
+ if (value.startsWith("/") || value.startsWith("~") || value.startsWith("./"))
198
+ return false;
199
+ return shannonEntropy(value) >= 3.2;
200
+ }
201
+ function scrubSecrets(text) {
202
+ if (!text)
203
+ return text;
204
+ let scrubbed = text;
205
+ for (const pattern of TOKEN_PATTERNS)
206
+ scrubbed = scrubbed.replace(pattern, SCRUBBED);
207
+ scrubbed = scrubbed.replace(AUTHORIZATION_PATTERN, `$1${SCRUBBED}`);
208
+ scrubbed = scrubbed.replace(PEM_PATTERN, SCRUBBED);
209
+ if (!mayContainAssignment(scrubbed))
210
+ return scrubbed;
211
+ scrubbed = scrubbed.replace(ASSIGNMENT_PATTERN, (match, prefix, quotedValue, escapedQuoted, doubleQuoted, singleQuoted, bare) => {
212
+ const value = escapedQuoted ?? doubleQuoted ?? singleQuoted ?? bare ?? "";
213
+ const keyName = /^[A-Za-z0-9_.-]*/.exec(prefix)?.[0] ?? "";
214
+ if (BENIGN_KEY_NAME_PATTERN.test(keyName))
215
+ return match;
216
+ if (!isLikelySecretValue(value))
217
+ return match;
218
+ const quote = quotedValue.startsWith("\\\"") ? "\\\"" : quotedValue.startsWith('"') ? '"' : quotedValue.startsWith("'") ? "'" : "";
219
+ return `${prefix}${quote}${SCRUBBED}${quote}`;
220
+ });
221
+ return scrubbed;
222
+ }
223
+ function scrubSecretsDeep(value) {
224
+ if (typeof value === "string")
225
+ return scrubSecrets(value);
226
+ if (Array.isArray(value))
227
+ return value.map((entry) => scrubSecretsDeep(entry));
228
+ if (value !== null && typeof value === "object") {
229
+ const toJSON = value.toJSON;
230
+ if (typeof toJSON === "function")
231
+ return scrubSecretsDeep(toJSON.call(value));
232
+ const scrubbed = {};
233
+ for (const [key, entry] of Object.entries(value))
234
+ scrubbed[key] = scrubSecretsDeep(entry);
235
+ return scrubbed;
236
+ }
237
+ return value;
238
+ }
239
+
240
+ // src/lib/recurrence.ts
46
241
  function assertDate(value, label) {
47
242
  const date = new Date(value);
48
243
  if (Number.isNaN(date.getTime()))
@@ -79,7 +274,20 @@ function parseField(expr, min, max) {
79
274
  }
80
275
  return out;
81
276
  }
277
+ var MINUTE_MS = 60000;
278
+ var PARSED_CRON_CACHE_LIMIT = 256;
279
+ var parsedCronCache = new Map;
280
+ var emittedScheduleWarnings = new Set;
281
+ function warnOnce(key, message) {
282
+ if (emittedScheduleWarnings.has(key))
283
+ return;
284
+ emittedScheduleWarnings.add(key);
285
+ console.warn(`[open-loops] WARN ${message}`);
286
+ }
82
287
  function parseCron(expr) {
288
+ const cached = parsedCronCache.get(expr);
289
+ if (cached)
290
+ return cached;
83
291
  const fields = expr.trim().split(/\s+/);
84
292
  if (fields.length !== 5)
85
293
  throw new Error(`cron must have 5 fields, got ${fields.length}: "${expr}"`);
@@ -89,7 +297,7 @@ function parseCron(expr) {
89
297
  dowSet.delete(7);
90
298
  dowSet.add(0);
91
299
  }
92
- return {
300
+ const parsed = {
93
301
  minute: parseField(minute, 0, 59),
94
302
  hour: parseField(hour, 0, 23),
95
303
  dom: parseField(dom, 1, 31),
@@ -98,6 +306,13 @@ function parseCron(expr) {
98
306
  domRestricted: dom !== "*",
99
307
  dowRestricted: dow !== "*"
100
308
  };
309
+ if (parsedCronCache.size >= PARSED_CRON_CACHE_LIMIT) {
310
+ const oldest = parsedCronCache.keys().next().value;
311
+ if (oldest !== undefined)
312
+ parsedCronCache.delete(oldest);
313
+ }
314
+ parsedCronCache.set(expr, parsed);
315
+ return parsed;
101
316
  }
102
317
  function cronMatches(cron, date) {
103
318
  if (!cron.minute.has(date.getMinutes()))
@@ -171,14 +386,59 @@ function latestIntervalSlot(first, now, everyMs) {
171
386
  const steps = Math.floor((now.getTime() - first.getTime()) / everyMs);
172
387
  return new Date(first.getTime() + steps * everyMs);
173
388
  }
389
+ function floorToMinute(date) {
390
+ const floored = new Date(date.getTime());
391
+ floored.setSeconds(0, 0);
392
+ return floored;
393
+ }
394
+ var LATEST_CRON_SCAN_LIMIT_MINUTES = 366 * 24 * 60;
395
+ function latestDailyCronSlot(cron, now) {
396
+ const hoursDesc = [...cron.hour].sort((a, b) => b - a);
397
+ const minutesDesc = [...cron.minute].sort((a, b) => b - a);
398
+ const candidate = floorToMinute(now);
399
+ for (let dayOffset = 0;dayOffset < 2; dayOffset += 1) {
400
+ const isToday = dayOffset === 0;
401
+ for (const hour of hoursDesc) {
402
+ if (isToday && hour > now.getHours())
403
+ continue;
404
+ const minuteCap = isToday && hour === now.getHours() ? now.getMinutes() : 59;
405
+ const minute = minutesDesc.find((value) => value <= minuteCap);
406
+ if (minute === undefined)
407
+ continue;
408
+ candidate.setHours(hour, minute, 0, 0);
409
+ return candidate;
410
+ }
411
+ candidate.setDate(candidate.getDate() - 1);
412
+ }
413
+ throw new Error("unreachable: daily cron pattern must match within two days");
414
+ }
174
415
  function latestCronSlot(first, now, expression) {
175
- let current = first;
176
- while (true) {
177
- const next = nextCronRun(expression, current);
178
- if (next.getTime() > now.getTime())
179
- return current;
180
- current = next;
416
+ if (first.getTime() > now.getTime())
417
+ return first;
418
+ const cron = parseCron(expression);
419
+ if (!cron.domRestricted && !cron.dowRestricted && cron.month.size === 12) {
420
+ const latest = latestDailyCronSlot(cron, now);
421
+ return latest.getTime() >= first.getTime() ? latest : first;
422
+ }
423
+ const scanFloorMs = Math.max(first.getTime(), now.getTime() - LATEST_CRON_SCAN_LIMIT_MINUTES * MINUTE_MS);
424
+ const cursor = floorToMinute(now);
425
+ while (cursor.getTime() >= scanFloorMs) {
426
+ if (cronMatches(cron, cursor))
427
+ return cursor;
428
+ cursor.setMinutes(cursor.getMinutes() - 1);
181
429
  }
430
+ if (first.getTime() < scanFloorMs) {
431
+ warnOnce(`latest-cron:${expression}`, `latestCronSlot: no match for "${expression}" within the ${LATEST_CRON_SCAN_LIMIT_MINUTES}-minute scan window; using the stored next run as the latest slot`);
432
+ }
433
+ return first;
434
+ }
435
+ var MAX_CATCH_UP_SLOTS = 1000;
436
+ function catchUpSlotLimit(loop) {
437
+ if (loop.catchUpLimit > MAX_CATCH_UP_SLOTS) {
438
+ warnOnce(`catch-up-limit:${loop.id}`, `dueSlots: loop ${loop.id} catchUpLimit ${loop.catchUpLimit} exceeds the per-plan cap; using ${MAX_CATCH_UP_SLOTS}`);
439
+ return MAX_CATCH_UP_SLOTS;
440
+ }
441
+ return loop.catchUpLimit;
182
442
  }
183
443
  function dueSlots(loop, now) {
184
444
  if (!loop.nextRunAt || loop.status !== "active")
@@ -197,9 +457,10 @@ function dueSlots(loop, now) {
197
457
  return { slots: [next.toISOString()] };
198
458
  case "interval": {
199
459
  if (catchUp === "all") {
460
+ const limit = catchUpSlotLimit(loop);
200
461
  const slots = [];
201
462
  let cursor = next;
202
- while (cursor.getTime() <= now.getTime() && slots.length < loop.catchUpLimit) {
463
+ while (cursor.getTime() <= now.getTime() && slots.length < limit) {
203
464
  slots.push(cursor.toISOString());
204
465
  cursor = new Date(cursor.getTime() + loop.schedule.everyMs);
205
466
  }
@@ -211,9 +472,10 @@ function dueSlots(loop, now) {
211
472
  }
212
473
  case "cron": {
213
474
  if (catchUp === "all") {
475
+ const limit = catchUpSlotLimit(loop);
214
476
  const slots = [];
215
477
  let cursor = next;
216
- while (cursor.getTime() <= now.getTime() && slots.length < loop.catchUpLimit) {
478
+ while (cursor.getTime() <= now.getTime() && slots.length < limit) {
217
479
  slots.push(cursor.toISOString());
218
480
  cursor = nextCronRun(loop.schedule.expression, cursor);
219
481
  }
@@ -312,8 +574,311 @@ function updateReadyFlags(nodes, planStatus) {
312
574
  }
313
575
 
314
576
  // src/lib/workflow-spec.ts
315
- import { readFileSync } from "fs";
577
+ import { readFileSync as readFileSync2 } from "fs";
316
578
  import { isAbsolute, resolve } from "path";
579
+
580
+ // src/lib/agent-adapter.ts
581
+ import { spawn } from "child_process";
582
+ var UNSAFE_CODEWITH_DURABLE_EXTRA_ARGS = new Set([
583
+ "e",
584
+ "exec",
585
+ "agent",
586
+ "start",
587
+ "--ephemeral",
588
+ "--ignore-rules",
589
+ "--skip-git-repo-check",
590
+ "--json",
591
+ "--output-last-message",
592
+ "-o",
593
+ "--output-schema",
594
+ "--dangerously-bypass-approvals-and-sandbox"
595
+ ]);
596
+ var CODEX_LIKE_SANDBOXES = ["read-only", "workspace-write", "danger-full-access"];
597
+ var CURSOR_SANDBOXES = ["enabled", "disabled"];
598
+ var PERMISSION_MODES = ["default", "plan", "auto", "bypass"];
599
+ function assertOptionalNonEmptyString(value, label) {
600
+ if (value === undefined)
601
+ return;
602
+ if (typeof value !== "string" || value.trim() === "")
603
+ throw new Error(`${label} must be a non-empty string`);
604
+ }
605
+ function validateAgentOptions(target, label, capabilities) {
606
+ const provider = target.provider;
607
+ if (typeof target.prompt !== "string" || target.prompt.trim() === "") {
608
+ throw new Error(`${label}.prompt must be a non-empty string`);
609
+ }
610
+ assertOptionalNonEmptyString(target.model, `${label}.model`);
611
+ assertOptionalNonEmptyString(target.variant, `${label}.variant`);
612
+ assertOptionalNonEmptyString(target.agent, `${label}.agent`);
613
+ assertOptionalNonEmptyString(target.authProfile, `${label}.authProfile`);
614
+ assertOptionalNonEmptyString(target.configIsolation, `${label}.configIsolation`);
615
+ if (target.configIsolation !== undefined && target.configIsolation !== "safe" && target.configIsolation !== "none") {
616
+ throw new Error(`${label}.configIsolation must be safe or none`);
617
+ }
618
+ if (target.authProfile !== undefined && provider !== "codewith") {
619
+ throw new Error(`${label}.authProfile is currently supported only for provider codewith`);
620
+ }
621
+ if (provider === "opencode" && (typeof target.model !== "string" || target.model.trim() === "")) {
622
+ throw new Error(`${label}.model is required for provider opencode; pass a provider/model id such as openrouter/google/gemini-2.5-flash`);
623
+ }
624
+ if (provider === "cursor" && target.variant !== undefined) {
625
+ throw new Error(`${label}.variant is not supported for provider cursor`);
626
+ }
627
+ if (provider === "codex" && target.agent !== undefined) {
628
+ throw new Error(`${label}.agent is not supported for provider codex`);
629
+ }
630
+ if (provider === "codewith" && target.agent !== undefined) {
631
+ throw new Error(`${label}.agent is not supported for provider codewith durable background-agent execution`);
632
+ }
633
+ if (provider === "codewith") {
634
+ const unsafe = target.extraArgs?.find((arg) => UNSAFE_CODEWITH_DURABLE_EXTRA_ARGS.has(arg));
635
+ if (unsafe) {
636
+ throw new Error(`${label}.extraArgs cannot include ${unsafe}; codewith agent steps use durable agent start, not exec/ephemeral flags`);
637
+ }
638
+ }
639
+ if (target.addDirs?.length && !["codewith", "codex"].includes(provider)) {
640
+ throw new Error(`${label}.addDirs is currently supported only for provider codewith or codex`);
641
+ }
642
+ if (target.permissionMode !== undefined) {
643
+ if (!PERMISSION_MODES.includes(target.permissionMode)) {
644
+ throw new Error(`${label}.permissionMode must be one of ${PERMISSION_MODES.join(", ")}`);
645
+ }
646
+ if (target.permissionMode === "plan" && !["claude", "cursor"].includes(provider)) {
647
+ throw new Error(`${label}.permissionMode plan is currently supported only for provider claude or cursor`);
648
+ }
649
+ if (target.permissionMode === "auto" && provider !== "claude") {
650
+ throw new Error(`${label}.permissionMode auto is currently supported only for provider claude`);
651
+ }
652
+ }
653
+ if (target.sandbox !== undefined) {
654
+ if (!capabilities.sandbox.length) {
655
+ throw new Error(`${label}.sandbox is currently supported only for provider codewith, codex, or cursor`);
656
+ }
657
+ if (!capabilities.sandbox.includes(target.sandbox)) {
658
+ throw new Error(`${label}.sandbox must be one of ${capabilities.sandbox.join(", ")}`);
659
+ }
660
+ }
661
+ }
662
+ function codewithLikeSandbox(target) {
663
+ return target.sandbox ?? (target.permissionMode === "bypass" ? "danger-full-access" : "workspace-write");
664
+ }
665
+ function configStringValue(value) {
666
+ return JSON.stringify(value);
667
+ }
668
+ function buildAgentInvocation(target) {
669
+ const isolation = target.configIsolation ?? "safe";
670
+ const permissionMode = target.permissionMode ?? "default";
671
+ const args = [];
672
+ switch (target.provider) {
673
+ case "claude": {
674
+ if (isolation === "safe")
675
+ args.push("--safe-mode", "--setting-sources", "local", "--no-session-persistence");
676
+ if (permissionMode !== "default") {
677
+ const mode = permissionMode === "bypass" ? "bypassPermissions" : permissionMode;
678
+ args.push("--permission-mode", mode);
679
+ }
680
+ args.push("-p", "--output-format", "json");
681
+ if (target.model)
682
+ args.push("--model", target.model);
683
+ if (target.variant)
684
+ args.push("--effort", target.variant);
685
+ if (target.agent)
686
+ args.push("--agent", target.agent);
687
+ args.push(...target.extraArgs ?? []);
688
+ return { command: "claude", args, stdin: target.prompt };
689
+ }
690
+ case "cursor": {
691
+ args.push("-c", [
692
+ "set -eu",
693
+ "if command -v agent >/dev/null 2>&1; then",
694
+ ' exec agent "$@"',
695
+ "else",
696
+ " echo 'Executable not found in PATH: agent' >&2",
697
+ " exit 127",
698
+ "fi"
699
+ ].join(`
700
+ `), "openloops-cursor", "-p");
701
+ if (permissionMode === "plan")
702
+ args.push("--mode", "plan");
703
+ if (permissionMode === "bypass")
704
+ args.push("--force");
705
+ const cursorSandbox = target.sandbox ?? (isolation === "safe" ? "enabled" : undefined);
706
+ if (cursorSandbox)
707
+ args.push("--sandbox", cursorSandbox);
708
+ if (target.model)
709
+ args.push("--model", target.model);
710
+ if (target.agent)
711
+ args.push("--agent", target.agent);
712
+ args.push(...target.extraArgs ?? []);
713
+ return { command: "sh", args, stdin: target.prompt, preflightAnyOf: ["agent"] };
714
+ }
715
+ case "codewith": {
716
+ args.push(...target.authProfile ? ["--auth-profile", target.authProfile] : []);
717
+ if (target.variant)
718
+ args.push("-c", `model_reasoning_effort=${configStringValue(target.variant)}`);
719
+ args.push("--ask-for-approval", "never", "--sandbox", codewithLikeSandbox(target));
720
+ if (target.cwd)
721
+ args.push("--cd", target.cwd);
722
+ for (const dir of target.addDirs ?? [])
723
+ args.push("--add-dir", dir);
724
+ if (target.model)
725
+ args.push("--model", target.model);
726
+ args.push(...target.extraArgs ?? []);
727
+ args.push("agent", "start");
728
+ if (target.cwd)
729
+ args.push("--cwd", target.cwd);
730
+ args.push(target.prompt);
731
+ return { command: "codewith", args };
732
+ }
733
+ case "codex": {
734
+ if (target.variant)
735
+ args.push("-c", `model_reasoning_effort=${configStringValue(target.variant)}`);
736
+ args.push("--ask-for-approval", "never", "exec", "--json", "--ephemeral", "--sandbox", codewithLikeSandbox(target), "--skip-git-repo-check");
737
+ if (isolation === "safe")
738
+ args.push("--ignore-rules");
739
+ if (target.cwd)
740
+ args.push("--cd", target.cwd);
741
+ for (const dir of target.addDirs ?? [])
742
+ args.push("--add-dir", dir);
743
+ if (target.model)
744
+ args.push("--model", target.model);
745
+ args.push(...target.extraArgs ?? []);
746
+ return { command: "codex", args, stdin: target.prompt };
747
+ }
748
+ case "aicopilot":
749
+ case "opencode": {
750
+ args.push("run", "--format", "json");
751
+ if (isolation === "safe")
752
+ args.push("--pure");
753
+ if (permissionMode === "bypass")
754
+ args.push("--dangerously-skip-permissions");
755
+ if (target.cwd)
756
+ args.push("--dir", target.cwd);
757
+ if (target.model)
758
+ args.push("--model", target.model);
759
+ if (target.variant)
760
+ args.push("--variant", target.variant);
761
+ if (target.agent)
762
+ args.push("--agent", target.agent);
763
+ args.push(...target.extraArgs ?? []);
764
+ return { command: target.provider, args, stdin: target.prompt };
765
+ }
766
+ }
767
+ }
768
+ function adapterFor(provider, capabilities) {
769
+ return {
770
+ provider,
771
+ capabilities,
772
+ validate(target, label = provider) {
773
+ validateAgentOptions(target, label, capabilities);
774
+ },
775
+ buildInvocation(target) {
776
+ validateAgentOptions(target, provider, capabilities);
777
+ return buildAgentInvocation(target);
778
+ }
779
+ };
780
+ }
781
+ var PROVIDER_ADAPTERS = {
782
+ claude: adapterFor("claude", { sandbox: [], durable: false, remote: true, promptChannel: "stdin" }),
783
+ cursor: adapterFor("cursor", { sandbox: CURSOR_SANDBOXES, durable: false, remote: true, promptChannel: "stdin" }),
784
+ codewith: adapterFor("codewith", { sandbox: CODEX_LIKE_SANDBOXES, durable: true, remote: false, promptChannel: "argv" }),
785
+ codex: adapterFor("codex", { sandbox: CODEX_LIKE_SANDBOXES, durable: false, remote: true, promptChannel: "stdin" }),
786
+ aicopilot: adapterFor("aicopilot", { sandbox: [], durable: false, remote: true, promptChannel: "stdin" }),
787
+ opencode: adapterFor("opencode", { sandbox: [], durable: false, remote: true, promptChannel: "stdin" })
788
+ };
789
+ var AGENT_PROVIDERS = Object.keys(PROVIDER_ADAPTERS);
790
+ function providerAdapter(provider) {
791
+ const adapter = PROVIDER_ADAPTERS[provider];
792
+ if (!adapter)
793
+ throw new Error(`unsupported agent provider: ${String(provider)}`);
794
+ return adapter;
795
+ }
796
+ var DEFAULT_CAPTURE_MAX_OUTPUT_BYTES = 256 * 1024;
797
+ function killProcessGroup(pgid) {
798
+ try {
799
+ process.kill(-pgid, "SIGTERM");
800
+ } catch {
801
+ try {
802
+ process.kill(pgid, "SIGTERM");
803
+ } catch {}
804
+ }
805
+ setTimeout(() => {
806
+ try {
807
+ process.kill(-pgid, "SIGKILL");
808
+ } catch {
809
+ try {
810
+ process.kill(pgid, "SIGKILL");
811
+ } catch {}
812
+ }
813
+ }, 2000).unref();
814
+ }
815
+
816
+ class BoundedOutputBuffer {
817
+ maxBytes;
818
+ text = "";
819
+ truncatedBytes = 0;
820
+ constructor(maxBytes) {
821
+ this.maxBytes = maxBytes;
822
+ }
823
+ append(chunk) {
824
+ if (!chunk)
825
+ return;
826
+ this.text += chunk;
827
+ if (Buffer.byteLength(this.text, "utf8") <= this.maxBytes)
828
+ return;
829
+ this.text = scrubSecrets(this.text);
830
+ const encoded = Buffer.from(this.text, "utf8");
831
+ if (encoded.length <= this.maxBytes)
832
+ return;
833
+ let start = encoded.length - this.maxBytes;
834
+ while (start < encoded.length && (encoded[start] & 192) === 128)
835
+ start++;
836
+ this.truncatedBytes += start;
837
+ this.text = encoded.subarray(start).toString("utf8");
838
+ }
839
+ value() {
840
+ if (this.truncatedBytes === 0)
841
+ return this.text;
842
+ return `[truncated ${this.truncatedBytes} bytes]
843
+ ${this.text}`;
844
+ }
845
+ }
846
+ async function spawnCapture(command, args, opts) {
847
+ const maxOutputBytes = opts.maxOutputBytes ?? DEFAULT_CAPTURE_MAX_OUTPUT_BYTES;
848
+ const stdout = new BoundedOutputBuffer(maxOutputBytes);
849
+ const stderr = new BoundedOutputBuffer(maxOutputBytes);
850
+ let timedOut = false;
851
+ let error;
852
+ const child = spawn(command, args, {
853
+ cwd: opts.cwd,
854
+ env: opts.env,
855
+ detached: true,
856
+ stdio: ["ignore", "pipe", "pipe"]
857
+ });
858
+ const timer = setTimeout(() => {
859
+ timedOut = true;
860
+ if (child.pid)
861
+ killProcessGroup(child.pid);
862
+ }, opts.timeoutMs);
863
+ timer.unref();
864
+ child.stdout?.setEncoding("utf8");
865
+ child.stderr?.setEncoding("utf8");
866
+ child.stdout?.on("data", (chunk) => stdout.append(chunk));
867
+ child.stderr?.on("data", (chunk) => stderr.append(chunk));
868
+ const [status, signal] = await new Promise((resolve) => {
869
+ child.once("error", (err) => {
870
+ error = err.message;
871
+ resolve([null, null]);
872
+ });
873
+ child.once("close", (code, sig) => resolve([code, sig]));
874
+ });
875
+ clearTimeout(timer);
876
+ if (timedOut && !error)
877
+ error = `timed out after ${opts.timeoutMs}ms`;
878
+ return { status, signal, stdout: stdout.value(), stderr: stderr.value(), error, timedOut };
879
+ }
880
+
881
+ // src/lib/workflow-spec.ts
317
882
  function assertObject(value, label) {
318
883
  if (!value || typeof value !== "object" || Array.isArray(value))
319
884
  throw new Error(`${label} must be an object`);
@@ -356,20 +921,6 @@ function optionalStringArray(value, label) {
356
921
  }).filter(Boolean);
357
922
  return values.length ? values : undefined;
358
923
  }
359
- var UNSAFE_CODEWITH_DURABLE_EXTRA_ARGS = new Set([
360
- "e",
361
- "exec",
362
- "agent",
363
- "start",
364
- "--ephemeral",
365
- "--ignore-rules",
366
- "--skip-git-repo-check",
367
- "--json",
368
- "--output-last-message",
369
- "-o",
370
- "--output-schema",
371
- "--dangerously-bypass-approvals-and-sandbox"
372
- ]);
373
924
  function optionalAccountRef(value, label) {
374
925
  if (value === undefined)
375
926
  return;
@@ -390,7 +941,7 @@ function readPromptFile(rawPath, label, opts) {
390
941
  const path = promptFilePath(rawPath.trim(), opts);
391
942
  let prompt;
392
943
  try {
393
- prompt = readFileSync(path, "utf8");
944
+ prompt = readFileSync2(path, "utf8");
394
945
  } catch (error) {
395
946
  const code = error && typeof error === "object" && "code" in error ? String(error.code) : undefined;
396
947
  throw new Error(`${label}.promptFile could not be read${code ? `: ${code}` : ""}`);
@@ -409,7 +960,7 @@ function matchingPromptSource(value, prompt, opts) {
409
960
  return;
410
961
  const path = promptFilePath(rawPath.trim(), opts);
411
962
  try {
412
- return readFileSync(path, "utf8") === prompt ? { type: "file", path } : undefined;
963
+ return readFileSync2(path, "utf8") === prompt ? { type: "file", path } : undefined;
413
964
  } catch {
414
965
  return;
415
966
  }
@@ -456,74 +1007,14 @@ function validateTarget(value, label, opts) {
456
1007
  if (!hasPrompt && !hasPromptFile)
457
1008
  throw new Error(`${label}.prompt must be a non-empty string or ${label}.promptFile must be set`);
458
1009
  const promptFields = hasPrompt ? { prompt: value.prompt, promptSource: matchingPromptSource(value.promptSource, value.prompt, opts) } : readPromptFile(value.promptFile, label, opts);
459
- const providers = ["claude", "cursor", "codewith", "aicopilot", "opencode", "codex"];
460
- if (!providers.includes(value.provider))
461
- throw new Error(`${label}.provider must be one of ${providers.join(", ")}`);
1010
+ if (!AGENT_PROVIDERS.includes(value.provider)) {
1011
+ throw new Error(`${label}.provider must be one of ${AGENT_PROVIDERS.join(", ")}`);
1012
+ }
462
1013
  optionalTimeoutMs(value.timeoutMs, `${label}.timeoutMs`);
463
1014
  optionalPositiveInteger(value.idleTimeoutMs, `${label}.idleTimeoutMs`);
464
- if (value.authProfile !== undefined) {
465
- assertString(value.authProfile, `${label}.authProfile`);
466
- if (value.provider !== "codewith")
467
- throw new Error(`${label}.authProfile is currently supported only for provider codewith`);
468
- }
469
- if (value.configIsolation !== undefined) {
470
- assertString(value.configIsolation, `${label}.configIsolation`);
471
- if (!["safe", "none"].includes(value.configIsolation))
472
- throw new Error(`${label}.configIsolation must be safe or none`);
473
- }
474
- if (value.model !== undefined)
475
- assertString(value.model, `${label}.model`);
476
- if (value.provider === "opencode" && (typeof value.model !== "string" || value.model.trim() === "")) {
477
- throw new Error(`${label}.model is required for provider opencode; pass a provider/model id such as openrouter/google/gemini-2.5-flash`);
478
- }
479
- if (value.variant !== undefined)
480
- assertString(value.variant, `${label}.variant`);
481
- if (value.agent !== undefined)
482
- assertString(value.agent, `${label}.agent`);
483
- if (value.provider === "cursor" && value.variant !== undefined)
484
- throw new Error(`${label}.variant is not supported for provider cursor`);
485
- if (value.provider === "codex" && value.agent !== undefined)
486
- throw new Error(`${label}.agent is not supported for provider codex`);
487
- if (value.provider === "codewith" && value.agent !== undefined) {
488
- throw new Error(`${label}.agent is not supported for provider codewith durable background-agent execution`);
489
- }
490
1015
  const extraArgs = optionalStringArray(value.extraArgs, `${label}.extraArgs`);
491
- if (value.provider === "codewith") {
492
- const unsafe = extraArgs?.find((arg) => UNSAFE_CODEWITH_DURABLE_EXTRA_ARGS.has(arg));
493
- if (unsafe)
494
- throw new Error(`${label}.extraArgs cannot include ${unsafe}; codewith agent steps use durable agent start, not exec/ephemeral flags`);
495
- }
496
1016
  optionalStringArray(value.addDirs, `${label}.addDirs`);
497
- if (Array.isArray(value.addDirs) && value.addDirs.length > 0 && !["codewith", "codex"].includes(value.provider)) {
498
- throw new Error(`${label}.addDirs is currently supported only for provider codewith or codex`);
499
- }
500
- if (value.permissionMode !== undefined) {
501
- assertString(value.permissionMode, `${label}.permissionMode`);
502
- const permissionModes = ["default", "plan", "auto", "bypass"];
503
- if (!permissionModes.includes(value.permissionMode)) {
504
- throw new Error(`${label}.permissionMode must be one of ${permissionModes.join(", ")}`);
505
- }
506
- if (value.permissionMode === "plan" && !["claude", "cursor"].includes(value.provider)) {
507
- throw new Error(`${label}.permissionMode plan is currently supported only for provider claude or cursor`);
508
- }
509
- if (value.permissionMode === "auto" && value.provider !== "claude") {
510
- throw new Error(`${label}.permissionMode auto is currently supported only for provider claude`);
511
- }
512
- }
513
- if (value.sandbox !== undefined) {
514
- assertString(value.sandbox, `${label}.sandbox`);
515
- const codexLike = ["read-only", "workspace-write", "danger-full-access"];
516
- const cursorLike = ["enabled", "disabled"];
517
- if (["codewith", "codex"].includes(value.provider)) {
518
- if (!codexLike.includes(value.sandbox))
519
- throw new Error(`${label}.sandbox must be one of ${codexLike.join(", ")}`);
520
- } else if (value.provider === "cursor") {
521
- if (!cursorLike.includes(value.sandbox))
522
- throw new Error(`${label}.sandbox must be one of ${cursorLike.join(", ")}`);
523
- } else {
524
- throw new Error(`${label}.sandbox is currently supported only for provider codewith, codex, or cursor`);
525
- }
526
- }
1017
+ providerAdapter(value.provider).validate({ ...value, extraArgs, ...promptFields }, label);
527
1018
  if (value.allowlist !== undefined) {
528
1019
  assertObject(value.allowlist, `${label}.allowlist`);
529
1020
  optionalStringArray(value.allowlist.tools, `${label}.allowlist.tools`);
@@ -653,8 +1144,8 @@ function workflowBodyFromJson(value, fallbackName, opts = {}) {
653
1144
 
654
1145
  // src/lib/run-artifacts.ts
655
1146
  import { createHash } from "crypto";
656
- import { mkdirSync as mkdirSync2, writeFileSync } from "fs";
657
- import { basename, join as join2 } from "path";
1147
+ import { mkdirSync as mkdirSync2, renameSync, rmdirSync, rmSync, writeFileSync } from "fs";
1148
+ import { basename, dirname, join as join2 } from "path";
658
1149
  function shortHash(value) {
659
1150
  return createHash("sha256").update(value).digest("hex").slice(0, 12);
660
1151
  }
@@ -675,13 +1166,14 @@ function workflowRunProjectSlug(projectKey) {
675
1166
  return "global";
676
1167
  return safeRunPathSlug(projectKey.startsWith("/") ? basename(projectKey) : projectKey, "project");
677
1168
  }
678
- function writeWorkflowRunManifest(args) {
1169
+ function stageWorkflowRunManifest(args) {
679
1170
  const projectSlug = workflowRunProjectSlug(args.projectKey);
680
1171
  const subjectKey = workflowRunSubjectKey(args.subjectKind, args.rawSubjectRef);
681
1172
  const dir = join2(args.loopsDataDir, "runs", projectSlug, subjectKey, args.workflowRunId);
682
1173
  mkdirSync2(dir, { recursive: true, mode: 448 });
683
1174
  const manifestPath = join2(dir, "manifest.json");
684
- writeFileSync(manifestPath, JSON.stringify({
1175
+ const tmpPath = `${manifestPath}.tmp`;
1176
+ writeFileSync(tmpPath, JSON.stringify({
685
1177
  version: 1,
686
1178
  workflowRunId: args.workflowRunId,
687
1179
  workflowId: args.workflowId,
@@ -694,13 +1186,26 @@ function writeWorkflowRunManifest(args) {
694
1186
  createdAt: new Date().toISOString(),
695
1187
  ...args.payload
696
1188
  }, null, 2), { mode: 384 });
697
- return manifestPath;
1189
+ return { manifestPath, tmpPath };
1190
+ }
1191
+ function commitWorkflowRunManifest(staged) {
1192
+ renameSync(staged.tmpPath, staged.manifestPath);
1193
+ return staged.manifestPath;
1194
+ }
1195
+ function discardWorkflowRunManifest(staged) {
1196
+ rmSync(staged.tmpPath, { force: true });
1197
+ try {
1198
+ rmdirSync(dirname(staged.manifestPath));
1199
+ } catch {}
698
1200
  }
699
1201
 
700
1202
  // src/lib/store.ts
701
1203
  var DEFAULT_RECOVERY_BATCH_LIMIT = 100;
702
1204
  var DEFAULT_RECOVERY_SCAN_MULTIPLIER = 5;
703
1205
  var LIVE_EXPIRED_RUN_GRACE_MS = 60000;
1206
+ var SCHEMA_USER_VERSION = 6;
1207
+ var TERMINAL_RUN_STATUSES = ["succeeded", "failed", "timed_out", "abandoned", "skipped"];
1208
+ var PRUNE_BATCH_SIZE = 400;
704
1209
  var GENERATED_ROUTE_TEMPLATE_IDS = new Set(["todos-task-worker-verifier", "task-lifecycle", "event-worker-verifier"]);
705
1210
  var GENERATED_ROUTE_KEYS = new Set(["todos-task", "generic-event"]);
706
1211
  function rowToLoop(row) {
@@ -741,6 +1246,8 @@ function rowToRun(row) {
741
1246
  claimedBy: row.claimed_by ?? undefined,
742
1247
  leaseExpiresAt: row.lease_expires_at ?? undefined,
743
1248
  pid: row.pid ?? undefined,
1249
+ pgid: row.pgid ?? undefined,
1250
+ processStartedAt: row.process_started_at ?? undefined,
744
1251
  exitCode: row.exit_code ?? undefined,
745
1252
  durationMs: row.duration_ms ?? undefined,
746
1253
  stdout: row.stdout ?? undefined,
@@ -926,6 +1433,24 @@ function isProcessAlive(pid) {
926
1433
  return false;
927
1434
  }
928
1435
  }
1436
+ function isRecordedProcessAlive(pid, processStartedAt) {
1437
+ if (!pid || !isProcessAlive(pid))
1438
+ return false;
1439
+ return sameProcessStart(processStartedAt ?? undefined, processStartTimeMs(pid));
1440
+ }
1441
+ function isoProcessStart(pid) {
1442
+ const startedMs = processStartTimeMs(pid);
1443
+ return startedMs === undefined ? undefined : new Date(startedMs).toISOString();
1444
+ }
1445
+ function isLiveStepProcess(pid, stepStartedAt) {
1446
+ if (!isProcessAlive(pid))
1447
+ return false;
1448
+ const actualMs = processStartTimeMs(pid);
1449
+ const stepStartMs = stepStartedAt ? Date.parse(stepStartedAt) : Number.NaN;
1450
+ if (actualMs === undefined || !Number.isFinite(stepStartMs))
1451
+ return true;
1452
+ return actualMs >= stepStartMs - START_TIME_TOLERANCE_MS;
1453
+ }
929
1454
  function rowToLease(row) {
930
1455
  return {
931
1456
  id: row.id,
@@ -945,6 +1470,9 @@ function workItemStatusForLoopRun(status, attempt, maxAttempts) {
945
1470
  }
946
1471
  return;
947
1472
  }
1473
+ function scrubbedOrNull(value) {
1474
+ return value == null ? null : scrubSecrets(value);
1475
+ }
948
1476
  function chmodIfExists(path, mode) {
949
1477
  try {
950
1478
  if (existsSync(path))
@@ -952,7 +1480,7 @@ function chmodIfExists(path, mode) {
952
1480
  } catch {}
953
1481
  }
954
1482
  function ensurePrivateStorePath(file) {
955
- const dir = dirname(file);
1483
+ const dir = dirname2(file);
956
1484
  mkdirSync3(dir, { recursive: true, mode: 448 });
957
1485
  chmodIfExists(dir, 448);
958
1486
  chmodIfExists(file, 384);
@@ -967,14 +1495,19 @@ class Store {
967
1495
  const file = path ?? dbPath();
968
1496
  if (file !== ":memory:")
969
1497
  ensurePrivateStorePath(file);
970
- this.rootDir = file === ":memory:" ? mkdtempSync(join3(tmpdir(), "open-loops-store-")) : dirname(file);
1498
+ this.rootDir = file === ":memory:" ? mkdtempSync(join3(tmpdir(), "open-loops-store-")) : dirname2(file);
971
1499
  this.db = new Database(file);
972
1500
  this.db.exec("PRAGMA foreign_keys = ON;");
973
1501
  this.db.exec("PRAGMA busy_timeout = 5000;");
974
1502
  this.db.exec("PRAGMA journal_mode = WAL;");
975
1503
  if (file !== ":memory:")
976
1504
  ensurePrivateStorePath(file);
977
- this.migrate();
1505
+ try {
1506
+ this.migrate();
1507
+ } catch (error) {
1508
+ this.db.close();
1509
+ throw error;
1510
+ }
978
1511
  }
979
1512
  migrate() {
980
1513
  this.db.exec(`
@@ -982,7 +1515,69 @@ class Store {
982
1515
  id TEXT PRIMARY KEY,
983
1516
  applied_at TEXT NOT NULL
984
1517
  );
985
-
1518
+ `);
1519
+ const versionRow = this.db.query("PRAGMA user_version").get();
1520
+ const userVersion = versionRow?.user_version ?? 0;
1521
+ if (userVersion > SCHEMA_USER_VERSION) {
1522
+ throw new Error(`loops database schema version ${userVersion} is newer than this binary supports (${SCHEMA_USER_VERSION}); upgrade open-loops before opening this database`);
1523
+ }
1524
+ const applied = new Set(this.db.query("SELECT id FROM schema_migrations").all().map((row) => row.id));
1525
+ for (const migration of this.migrations()) {
1526
+ if (!migration.baseline && applied.has(migration.id))
1527
+ continue;
1528
+ migration.apply();
1529
+ if (!applied.has(migration.id)) {
1530
+ this.db.query("INSERT OR IGNORE INTO schema_migrations (id, applied_at) VALUES (?, ?)").run(migration.id, nowIso());
1531
+ }
1532
+ }
1533
+ if (userVersion !== SCHEMA_USER_VERSION)
1534
+ this.db.exec(`PRAGMA user_version = ${SCHEMA_USER_VERSION}`);
1535
+ }
1536
+ migrations() {
1537
+ return [
1538
+ { id: "0001_initial_and_workflows", baseline: true, apply: () => this.createBaseSchema() },
1539
+ { id: "0002_loop_machines", baseline: true, apply: () => this.addColumnIfMissing("loops", "machine_json", "TEXT") },
1540
+ {
1541
+ id: "0003_goals",
1542
+ baseline: true,
1543
+ apply: () => {
1544
+ this.addColumnIfMissing("loops", "goal_json", "TEXT");
1545
+ this.addColumnIfMissing("loop_runs", "goal_run_id", "TEXT");
1546
+ this.addColumnIfMissing("workflow_specs", "goal_json", "TEXT");
1547
+ this.addColumnIfMissing("workflow_runs", "goal_run_id", "TEXT");
1548
+ this.addColumnIfMissing("workflow_step_runs", "goal_run_id", "TEXT");
1549
+ }
1550
+ },
1551
+ {
1552
+ id: "0004_loop_archive_metadata",
1553
+ baseline: true,
1554
+ apply: () => {
1555
+ this.addColumnIfMissing("loops", "archived_at", "TEXT");
1556
+ this.addColumnIfMissing("loops", "archived_from_status", "TEXT");
1557
+ }
1558
+ },
1559
+ {
1560
+ id: "0005_workflow_invocations_and_admission",
1561
+ baseline: true,
1562
+ apply: () => {
1563
+ this.addColumnIfMissing("workflow_runs", "invocation_id", "TEXT");
1564
+ this.addColumnIfMissing("workflow_runs", "work_item_id", "TEXT");
1565
+ this.addColumnIfMissing("workflow_runs", "manifest_path", "TEXT");
1566
+ this.addColumnIfMissing("workflow_step_runs", "pid", "INTEGER");
1567
+ this.createWorkflowRunBackfillIndexes();
1568
+ }
1569
+ },
1570
+ {
1571
+ id: "0006_run_process_tracking",
1572
+ apply: () => {
1573
+ this.addColumnIfMissing("loop_runs", "pgid", "INTEGER");
1574
+ this.addColumnIfMissing("loop_runs", "process_started_at", "TEXT");
1575
+ }
1576
+ }
1577
+ ];
1578
+ }
1579
+ createBaseSchema() {
1580
+ this.db.exec(`
986
1581
  CREATE TABLE IF NOT EXISTS loops (
987
1582
  id TEXT PRIMARY KEY,
988
1583
  name TEXT NOT NULL,
@@ -1021,6 +1616,8 @@ class Store {
1021
1616
  claimed_by TEXT,
1022
1617
  lease_expires_at TEXT,
1023
1618
  pid INTEGER,
1619
+ pgid INTEGER,
1620
+ process_started_at TEXT,
1024
1621
  exit_code INTEGER,
1025
1622
  duration_ms INTEGER,
1026
1623
  stdout TEXT,
@@ -1245,24 +1842,6 @@ class Store {
1245
1842
  CREATE INDEX IF NOT EXISTS idx_goal_runs_loop_run ON goal_runs(loop_run_id);
1246
1843
  CREATE INDEX IF NOT EXISTS idx_goal_runs_workflow_run ON goal_runs(workflow_run_id);
1247
1844
  `);
1248
- this.addColumnIfMissing("loops", "machine_json", "TEXT");
1249
- this.addColumnIfMissing("loops", "goal_json", "TEXT");
1250
- this.addColumnIfMissing("loops", "archived_at", "TEXT");
1251
- this.addColumnIfMissing("loops", "archived_from_status", "TEXT");
1252
- this.addColumnIfMissing("loop_runs", "goal_run_id", "TEXT");
1253
- this.addColumnIfMissing("workflow_specs", "goal_json", "TEXT");
1254
- this.addColumnIfMissing("workflow_runs", "goal_run_id", "TEXT");
1255
- this.addColumnIfMissing("workflow_runs", "invocation_id", "TEXT");
1256
- this.addColumnIfMissing("workflow_runs", "work_item_id", "TEXT");
1257
- this.addColumnIfMissing("workflow_runs", "manifest_path", "TEXT");
1258
- this.addColumnIfMissing("workflow_step_runs", "pid", "INTEGER");
1259
- this.addColumnIfMissing("workflow_step_runs", "goal_run_id", "TEXT");
1260
- this.createWorkflowRunBackfillIndexes();
1261
- this.db.query("INSERT OR IGNORE INTO schema_migrations (id, applied_at) VALUES (?, ?)").run("0001_initial_and_workflows", nowIso());
1262
- this.db.query("INSERT OR IGNORE INTO schema_migrations (id, applied_at) VALUES (?, ?)").run("0002_loop_machines", nowIso());
1263
- this.db.query("INSERT OR IGNORE INTO schema_migrations (id, applied_at) VALUES (?, ?)").run("0003_goals", nowIso());
1264
- this.db.query("INSERT OR IGNORE INTO schema_migrations (id, applied_at) VALUES (?, ?)").run("0004_loop_archive_metadata", nowIso());
1265
- this.db.query("INSERT OR IGNORE INTO schema_migrations (id, applied_at) VALUES (?, ?)").run("0005_workflow_invocations_and_admission", nowIso());
1266
1845
  }
1267
1846
  addColumnIfMissing(table, column, definition) {
1268
1847
  const columns = this.db.query(`PRAGMA table_info(${table})`).all();
@@ -1276,6 +1855,9 @@ class Store {
1276
1855
  CREATE INDEX IF NOT EXISTS idx_workflow_runs_work_item ON workflow_runs(work_item_id);
1277
1856
  `);
1278
1857
  }
1858
+ transact(fn) {
1859
+ return this.db.inTransaction ? fn() : this.writeTransaction(fn);
1860
+ }
1279
1861
  assertDaemonLeaseFence(opts = {}, now = nowIso()) {
1280
1862
  if (!opts.daemonLeaseId)
1281
1863
  return;
@@ -1357,14 +1939,14 @@ class Store {
1357
1939
  return byId;
1358
1940
  const rows = this.db.query("SELECT * FROM loops WHERE name = ? ORDER BY created_at DESC LIMIT 2").all(idOrName);
1359
1941
  if (rows.length === 0)
1360
- throw new Error(`loop not found: ${idOrName}`);
1942
+ throw new LoopNotFoundError(idOrName);
1361
1943
  if (rows.length > 1)
1362
- throw new Error(`ambiguous loop name: ${idOrName}; use a loop id`);
1944
+ throw new AmbiguousNameError(idOrName);
1363
1945
  return rowToLoop(rows[0]);
1364
1946
  }
1365
1947
  requireLoop(idOrName) {
1366
1948
  return this.getLoop(idOrName) ?? this.findLoopByName(idOrName) ?? (() => {
1367
- throw new Error(`loop not found: ${idOrName}`);
1949
+ throw new LoopNotFoundError(idOrName);
1368
1950
  })();
1369
1951
  }
1370
1952
  listLoops(opts = {}) {
@@ -1401,7 +1983,9 @@ class Store {
1401
1983
  try {
1402
1984
  const current = this.getLoop(id);
1403
1985
  if (!current)
1404
- throw new Error(`loop not found: ${id}`);
1986
+ throw new LoopNotFoundError(id);
1987
+ if (current.archivedAt)
1988
+ throw new LoopArchivedError(current.name || id);
1405
1989
  const merged = { ...current, ...patch, updatedAt: updated };
1406
1990
  const res = this.db.query(`UPDATE loops SET status=$status, next_run_at=$nextRun, retry_scheduled_for=$retrySlot,
1407
1991
  expires_at=$expiresAt, updated_at=$updated
@@ -1633,10 +2217,10 @@ class Store {
1633
2217
  renameLoop(id, name, opts = {}) {
1634
2218
  const current = this.getLoop(id);
1635
2219
  if (!current)
1636
- throw new Error(`loop not found: ${id}`);
2220
+ throw new LoopNotFoundError(id);
1637
2221
  const trimmed = name.trim();
1638
2222
  if (!trimmed)
1639
- throw new Error("loop name must not be empty");
2223
+ throw new ValidationError("loop name must not be empty");
1640
2224
  const updated = (opts.now ?? new Date).toISOString();
1641
2225
  this.db.query(`UPDATE loops SET name=$name, updated_at=$updated
1642
2226
  WHERE id=$id
@@ -1655,25 +2239,27 @@ class Store {
1655
2239
  return after;
1656
2240
  }
1657
2241
  archiveLoop(idOrName) {
1658
- const loop = this.requireLoop(idOrName);
1659
- if (loop.archivedAt)
1660
- return loop;
1661
- const updated = nowIso();
1662
- const archivedStatus = loop.status === "active" ? "paused" : loop.status;
1663
- this.db.query(`UPDATE loops
1664
- SET status=$status, archived_at=$archivedAt, archived_from_status=$archivedFromStatus, updated_at=$updated
1665
- WHERE id=$id`).run({
1666
- $id: loop.id,
1667
- $status: archivedStatus,
1668
- $archivedAt: updated,
1669
- $archivedFromStatus: loop.status,
1670
- $updated: updated
2242
+ return this.transact(() => {
2243
+ const loop = this.requireLoop(idOrName);
2244
+ if (loop.archivedAt)
2245
+ return loop;
2246
+ const updated = nowIso();
2247
+ const archivedStatus = loop.status === "active" ? "paused" : loop.status;
2248
+ this.db.query(`UPDATE loops
2249
+ SET status=$status, archived_at=$archivedAt, archived_from_status=$archivedFromStatus, updated_at=$updated
2250
+ WHERE id=$id`).run({
2251
+ $id: loop.id,
2252
+ $status: archivedStatus,
2253
+ $archivedAt: updated,
2254
+ $archivedFromStatus: loop.status,
2255
+ $updated: updated
2256
+ });
2257
+ this.setWorkflowWorkItemsForLoop(loop.id, "deferred", "loop archived", updated);
2258
+ const archived = this.getLoop(loop.id);
2259
+ if (!archived)
2260
+ throw new Error(`loop not found after archive: ${loop.id}`);
2261
+ return archived;
1671
2262
  });
1672
- this.setWorkflowWorkItemsForLoop(loop.id, "deferred", "loop archived", updated);
1673
- const archived = this.getLoop(loop.id);
1674
- if (!archived)
1675
- throw new Error(`loop not found after archive: ${loop.id}`);
1676
- return archived;
1677
2263
  }
1678
2264
  unarchiveLoop(idOrName) {
1679
2265
  const loop = this.requireLoop(idOrName);
@@ -1694,10 +2280,12 @@ class Store {
1694
2280
  return unarchived;
1695
2281
  }
1696
2282
  deleteLoop(idOrName) {
1697
- const loop = this.requireLoop(idOrName);
1698
- this.setWorkflowWorkItemsForLoop(loop.id, "cancelled", "loop deleted", nowIso());
1699
- const res = this.db.query("DELETE FROM loops WHERE id = ?").run(loop.id);
1700
- return res.changes > 0;
2283
+ return this.transact(() => {
2284
+ const loop = this.requireLoop(idOrName);
2285
+ this.setWorkflowWorkItemsForLoop(loop.id, "cancelled", "loop deleted", nowIso());
2286
+ const res = this.db.query("DELETE FROM loops WHERE id = ?").run(loop.id);
2287
+ return res.changes > 0;
2288
+ });
1701
2289
  }
1702
2290
  createWorkflow(input) {
1703
2291
  const normalized = normalizeCreateWorkflowInput(input);
@@ -2214,26 +2802,7 @@ class Store {
2214
2802
  try {
2215
2803
  this.assertDaemonLeaseFence(opts, now);
2216
2804
  for (const node of withReady) {
2217
- this.db.query(`INSERT OR IGNORE INTO goal_plan_nodes (id, goal_id, plan_id, key, sequence, priority, objective, status, ready,
2218
- token_budget, tokens_used, time_used_seconds, depends_on_json, created_at, updated_at)
2219
- VALUES ($id, $goalId, $planId, $key, $sequence, $priority, $objective, $status, $ready, $tokenBudget,
2220
- $tokensUsed, $timeUsedSeconds, $dependsOn, $created, $updated)`).run({
2221
- $id: node.nodeId,
2222
- $goalId: goal.goalId,
2223
- $planId: goal.planId,
2224
- $key: node.key,
2225
- $sequence: node.sequence,
2226
- $priority: node.priority,
2227
- $objective: node.objective,
2228
- $status: node.status,
2229
- $ready: node.ready ? 1 : 0,
2230
- $tokenBudget: node.tokenBudget ?? null,
2231
- $tokensUsed: node.tokensUsed,
2232
- $timeUsedSeconds: node.timeUsedSeconds,
2233
- $dependsOn: JSON.stringify(node.dependsOn),
2234
- $created: node.createdAt,
2235
- $updated: node.updatedAt
2236
- });
2805
+ this.insertGoalPlanNode(goal, node);
2237
2806
  }
2238
2807
  this.db.exec("COMMIT");
2239
2808
  } catch (error) {
@@ -2244,6 +2813,38 @@ class Store {
2244
2813
  }
2245
2814
  return this.listGoalPlanNodes(goalId);
2246
2815
  }
2816
+ insertGoalPlanNode(goal, node) {
2817
+ let id = node.nodeId;
2818
+ for (let attempt = 0;attempt < 3; attempt += 1) {
2819
+ const res = this.db.query(`INSERT OR IGNORE INTO goal_plan_nodes (id, goal_id, plan_id, key, sequence, priority, objective, status, ready,
2820
+ token_budget, tokens_used, time_used_seconds, depends_on_json, created_at, updated_at)
2821
+ VALUES ($id, $goalId, $planId, $key, $sequence, $priority, $objective, $status, $ready, $tokenBudget,
2822
+ $tokensUsed, $timeUsedSeconds, $dependsOn, $created, $updated)`).run({
2823
+ $id: id,
2824
+ $goalId: goal.goalId,
2825
+ $planId: goal.planId,
2826
+ $key: node.key,
2827
+ $sequence: node.sequence,
2828
+ $priority: node.priority,
2829
+ $objective: node.objective,
2830
+ $status: node.status,
2831
+ $ready: node.ready ? 1 : 0,
2832
+ $tokenBudget: node.tokenBudget ?? null,
2833
+ $tokensUsed: node.tokensUsed,
2834
+ $timeUsedSeconds: node.timeUsedSeconds,
2835
+ $dependsOn: JSON.stringify(node.dependsOn),
2836
+ $created: node.createdAt,
2837
+ $updated: node.updatedAt
2838
+ });
2839
+ if (res.changes === 1)
2840
+ return;
2841
+ const existingByKey = this.db.query("SELECT id FROM goal_plan_nodes WHERE plan_id = ? AND key = ? LIMIT 1").get(goal.planId, node.key);
2842
+ if (existingByKey)
2843
+ return;
2844
+ id = genId();
2845
+ }
2846
+ throw new Error(`goal plan node was not inserted after id retries: ${goal.planId}/${node.key}`);
2847
+ }
2247
2848
  listGoalPlanNodes(goalIdOrPlanId) {
2248
2849
  const rows = this.db.query("SELECT * FROM goal_plan_nodes WHERE goal_id = ? OR plan_id = ? ORDER BY sequence ASC").all(goalIdOrPlanId, goalIdOrPlanId);
2249
2850
  return rows.map(rowToGoalPlanNode);
@@ -2335,8 +2936,8 @@ class Store {
2335
2936
  $status: input.status,
2336
2937
  $nodeKey: input.nodeKey ?? null,
2337
2938
  $tokensUsed: input.tokensUsed ?? 0,
2338
- $evidence: input.evidence ? JSON.stringify(input.evidence) : null,
2339
- $rawResponse: input.rawResponse === undefined ? null : JSON.stringify(input.rawResponse),
2939
+ $evidence: input.evidence ? scrubSecrets(JSON.stringify(scrubSecretsDeep(input.evidence))) : null,
2940
+ $rawResponse: input.rawResponse === undefined ? null : scrubSecrets(JSON.stringify(scrubSecretsDeep(input.rawResponse))),
2340
2941
  $created: now,
2341
2942
  $updated: now
2342
2943
  });
@@ -2374,7 +2975,6 @@ class Store {
2374
2975
  const targetInput = input.loop?.target.type === "workflow" ? input.loop.target.input : undefined;
2375
2976
  const invocationId = input.invocationId ?? targetInput?.workflowInvocationId ?? targetInput?.invocationId;
2376
2977
  const workItemId = input.workItemId ?? targetInput?.workflowWorkItemId ?? targetInput?.workItemId;
2377
- let manifestPath;
2378
2978
  if (input.idempotencyKey) {
2379
2979
  const existing = this.db.query("SELECT * FROM workflow_runs WHERE workflow_id = ? AND idempotency_key = ? LIMIT 1").get(input.workflow.id, input.idempotencyKey);
2380
2980
  if (existing) {
@@ -2382,6 +2982,28 @@ class Store {
2382
2982
  return rowToWorkflowRun(existing);
2383
2983
  }
2384
2984
  }
2985
+ const runId = genId();
2986
+ const workItem = workItemId ? this.getWorkflowWorkItem(workItemId) : undefined;
2987
+ const invocation = invocationId ? this.getWorkflowInvocation(invocationId) : undefined;
2988
+ const staged = stageWorkflowRunManifest({
2989
+ loopsDataDir: this.rootDir,
2990
+ workflowRunId: runId,
2991
+ workflowId: input.workflow.id,
2992
+ workflowName: input.workflow.name,
2993
+ invocationId,
2994
+ workItemId,
2995
+ projectKey: workItem?.projectKey ?? invocation?.scope?.projectPath,
2996
+ subjectKind: invocation?.subjectRef.kind ?? (input.loop ? "loop" : "workflow"),
2997
+ rawSubjectRef: workItem?.subjectRef ?? invocation?.subjectRef.path ?? invocation?.subjectRef.id ?? invocation?.subjectRef.url ?? input.loop?.name ?? input.workflow.name,
2998
+ payload: {
2999
+ workflowInvocation: invocation,
3000
+ workflowWorkItem: workItem,
3001
+ loopId: input.loop?.id,
3002
+ loopRunId: input.loopRun?.id,
3003
+ scheduledFor: input.scheduledFor ?? input.loopRun?.scheduledFor
3004
+ }
3005
+ });
3006
+ const manifestPath = staged.manifestPath;
2385
3007
  this.db.exec("BEGIN IMMEDIATE");
2386
3008
  try {
2387
3009
  this.assertDaemonLeaseFence(input, now);
@@ -2389,30 +3011,10 @@ class Store {
2389
3011
  const existing = this.db.query("SELECT * FROM workflow_runs WHERE workflow_id = ? AND idempotency_key = ? LIMIT 1").get(input.workflow.id, input.idempotencyKey);
2390
3012
  if (existing) {
2391
3013
  this.db.exec("COMMIT");
3014
+ discardWorkflowRunManifest(staged);
2392
3015
  return rowToWorkflowRun(existing);
2393
3016
  }
2394
3017
  }
2395
- const runId = genId();
2396
- const workItem = workItemId ? this.getWorkflowWorkItem(workItemId) : undefined;
2397
- const invocation = invocationId ? this.getWorkflowInvocation(invocationId) : undefined;
2398
- manifestPath = invocation || workItem ? writeWorkflowRunManifest({
2399
- loopsDataDir: this.rootDir,
2400
- workflowRunId: runId,
2401
- workflowId: input.workflow.id,
2402
- workflowName: input.workflow.name,
2403
- invocationId,
2404
- workItemId,
2405
- projectKey: workItem?.projectKey ?? invocation?.scope?.projectPath,
2406
- subjectKind: invocation?.subjectRef.kind,
2407
- rawSubjectRef: workItem?.subjectRef ?? invocation?.subjectRef.path ?? invocation?.subjectRef.id ?? invocation?.subjectRef.url,
2408
- payload: {
2409
- workflowInvocation: invocation,
2410
- workflowWorkItem: workItem,
2411
- loopId: input.loop?.id,
2412
- loopRunId: input.loopRun?.id,
2413
- scheduledFor: input.scheduledFor ?? input.loopRun?.scheduledFor
2414
- }
2415
- }) : undefined;
2416
3018
  this.db.query(`INSERT INTO workflow_runs (id, workflow_id, workflow_name, loop_id, loop_run_id, invocation_id, work_item_id,
2417
3019
  scheduled_for, idempotency_key, manifest_path, status, started_at, finished_at, duration_ms, error,
2418
3020
  created_at, updated_at)
@@ -2479,6 +3081,7 @@ class Store {
2479
3081
  $created: now
2480
3082
  });
2481
3083
  this.db.exec("COMMIT");
3084
+ commitWorkflowRunManifest(staged);
2482
3085
  const run = this.getWorkflowRun(runId);
2483
3086
  if (!run)
2484
3087
  throw new Error(`workflow run not found after create: ${runId}`);
@@ -2487,8 +3090,7 @@ class Store {
2487
3090
  try {
2488
3091
  this.db.exec("ROLLBACK");
2489
3092
  } catch {}
2490
- if (manifestPath)
2491
- rmSync(manifestPath, { force: true });
3093
+ discardWorkflowRunManifest(staged);
2492
3094
  throw error;
2493
3095
  }
2494
3096
  }
@@ -2586,29 +3188,32 @@ class Store {
2586
3188
  return run;
2587
3189
  }
2588
3190
  recoverWorkflowRun(workflowRunId, reason = "workflow run recovered for retry") {
2589
- const now = nowIso();
2590
- const before = this.listWorkflowStepRuns(workflowRunId).filter((step) => step.status === "running");
2591
- const live = before.filter((step) => step.pid !== undefined && isProcessAlive(step.pid));
2592
- if (live.length > 0) {
2593
- throw new Error(`cannot recover workflow run while step processes are still alive: ${live.map((step) => `${step.stepId} pid=${step.pid}`).join(", ")}`);
2594
- }
2595
- this.db.query(`UPDATE workflow_step_runs
2596
- SET status='pending', started_at=NULL, finished_at=NULL, exit_code=NULL, pid=NULL, duration_ms=NULL,
2597
- stdout=NULL, stderr=NULL, error=$reason, updated_at=$updated
2598
- WHERE workflow_run_id=$workflowRunId AND status='running'`).run({ $workflowRunId: workflowRunId, $reason: reason, $updated: now });
2599
- if (before.length > 0) {
2600
- this.appendWorkflowEvent(workflowRunId, "recovered", undefined, {
2601
- reason,
2602
- recoveredSteps: before.map((step) => step.stepId)
2603
- });
2604
- }
2605
- return {
2606
- run: this.requireWorkflowRun(workflowRunId),
2607
- recoveredSteps: before.map((step) => this.getWorkflowStepRun(workflowRunId, step.stepId)).filter(Boolean)
2608
- };
3191
+ return this.transact(() => {
3192
+ const now = nowIso();
3193
+ const before = this.listWorkflowStepRuns(workflowRunId).filter((step) => step.status === "running");
3194
+ const live = before.filter((step) => step.pid !== undefined && isLiveStepProcess(step.pid, step.startedAt));
3195
+ if (live.length > 0) {
3196
+ throw new Error(`cannot recover workflow run while step processes are still alive: ${live.map((step) => `${step.stepId} pid=${step.pid}`).join(", ")}`);
3197
+ }
3198
+ this.db.query(`UPDATE workflow_step_runs
3199
+ SET status='pending', started_at=NULL, finished_at=NULL, exit_code=NULL, pid=NULL, duration_ms=NULL,
3200
+ stdout=NULL, stderr=NULL, error=$reason, updated_at=$updated
3201
+ WHERE workflow_run_id=$workflowRunId AND status='running'`).run({ $workflowRunId: workflowRunId, $reason: reason, $updated: now });
3202
+ if (before.length > 0) {
3203
+ this.appendWorkflowEvent(workflowRunId, "recovered", undefined, {
3204
+ reason,
3205
+ recoveredSteps: before.map((step) => step.stepId)
3206
+ });
3207
+ }
3208
+ return {
3209
+ run: this.requireWorkflowRun(workflowRunId),
3210
+ recoveredSteps: before.map((step) => this.getWorkflowStepRun(workflowRunId, step.stepId)).filter(Boolean)
3211
+ };
3212
+ });
2609
3213
  }
2610
3214
  finalizeWorkflowStepRun(workflowRunId, stepId, patch, opts = {}) {
2611
3215
  const finishedAt = patch.finishedAt ?? nowIso();
3216
+ const error = patch.error === undefined ? undefined : scrubSecrets(patch.error);
2612
3217
  this.db.exec("BEGIN IMMEDIATE");
2613
3218
  try {
2614
3219
  const res = this.db.query(`UPDATE workflow_step_runs SET status=$status, finished_at=$finished, exit_code=$exitCode, duration_ms=$durationMs,
@@ -2623,9 +3228,9 @@ class Store {
2623
3228
  $finished: finishedAt,
2624
3229
  $exitCode: patch.exitCode ?? null,
2625
3230
  $durationMs: patch.durationMs ?? null,
2626
- $stdout: patch.stdout ?? null,
2627
- $stderr: patch.stderr ?? null,
2628
- $error: patch.error ?? null,
3231
+ $stdout: scrubbedOrNull(patch.stdout),
3232
+ $stderr: scrubbedOrNull(patch.stderr),
3233
+ $error: error ?? null,
2629
3234
  $updated: finishedAt,
2630
3235
  $daemonLeaseId: opts.daemonLeaseId ?? null,
2631
3236
  $now: (opts.now ?? new Date(finishedAt)).toISOString()
@@ -2633,15 +3238,15 @@ class Store {
2633
3238
  if (res.changes === 1) {
2634
3239
  this.appendWorkflowEvent(workflowRunId, `step_${patch.status}`, stepId, {
2635
3240
  exitCode: patch.exitCode,
2636
- error: patch.error
3241
+ error
2637
3242
  });
2638
3243
  }
2639
3244
  this.db.exec("COMMIT");
2640
- } catch (error) {
3245
+ } catch (error2) {
2641
3246
  try {
2642
3247
  this.db.exec("ROLLBACK");
2643
3248
  } catch {}
2644
- throw error;
3249
+ throw error2;
2645
3250
  }
2646
3251
  const run = this.getWorkflowStepRun(workflowRunId, stepId);
2647
3252
  if (!run)
@@ -2681,6 +3286,7 @@ class Store {
2681
3286
  }
2682
3287
  finalizeWorkflowRun(workflowRunId, status, patch = {}, opts = {}) {
2683
3288
  const finishedAt = patch.finishedAt ?? nowIso();
3289
+ const error = patch.error === undefined ? undefined : scrubSecrets(patch.error);
2684
3290
  let changed = false;
2685
3291
  this.db.exec("BEGIN IMMEDIATE");
2686
3292
  try {
@@ -2693,25 +3299,25 @@ class Store {
2693
3299
  $status: status,
2694
3300
  $finished: finishedAt,
2695
3301
  $durationMs: patch.durationMs ?? null,
2696
- $error: patch.error ?? null,
3302
+ $error: error ?? null,
2697
3303
  $updated: finishedAt,
2698
3304
  $daemonLeaseId: opts.daemonLeaseId ?? null,
2699
3305
  $now: (opts.now ?? new Date(finishedAt)).toISOString()
2700
3306
  });
2701
3307
  changed = res.changes === 1;
2702
3308
  if (changed)
2703
- this.appendWorkflowEvent(workflowRunId, status, undefined, { error: patch.error });
3309
+ this.appendWorkflowEvent(workflowRunId, status, undefined, { error });
2704
3310
  if (changed) {
2705
3311
  const itemStatus = status === "succeeded" ? "succeeded" : status === "cancelled" ? "cancelled" : "failed";
2706
- this.setWorkflowWorkItemsForWorkflowRun(workflowRunId, itemStatus, patch.error, finishedAt);
3312
+ this.setWorkflowWorkItemsForWorkflowRun(workflowRunId, itemStatus, error, finishedAt);
2707
3313
  this.maybeArchiveTerminalGeneratedRouteWorkflow(workflowRunId, finishedAt);
2708
3314
  }
2709
3315
  this.db.exec("COMMIT");
2710
- } catch (error) {
3316
+ } catch (error2) {
2711
3317
  try {
2712
3318
  this.db.exec("ROLLBACK");
2713
3319
  } catch {}
2714
- throw error;
3320
+ throw error2;
2715
3321
  }
2716
3322
  const run = this.getWorkflowRun(workflowRunId);
2717
3323
  if (!run)
@@ -2744,24 +3350,26 @@ class Store {
2744
3350
  }
2745
3351
  }
2746
3352
  appendWorkflowEvent(workflowRunId, eventType, stepId, payload) {
2747
- const now = nowIso();
2748
- const current = this.db.query("SELECT MAX(sequence) AS sequence FROM workflow_events WHERE workflow_run_id = ?").get(workflowRunId);
2749
- const sequence = (current?.sequence ?? 0) + 1;
2750
- const id = genId();
2751
- this.db.query(`INSERT INTO workflow_events (id, workflow_run_id, sequence, event_type, step_id, payload_json, created_at)
2752
- VALUES ($id, $workflowRunId, $sequence, $eventType, $stepId, $payload, $created)`).run({
2753
- $id: id,
2754
- $workflowRunId: workflowRunId,
2755
- $sequence: sequence,
2756
- $eventType: eventType,
2757
- $stepId: stepId ?? null,
2758
- $payload: payload ? JSON.stringify(payload) : null,
2759
- $created: now
3353
+ return this.transact(() => {
3354
+ const now = nowIso();
3355
+ const current = this.db.query("SELECT MAX(sequence) AS sequence FROM workflow_events WHERE workflow_run_id = ?").get(workflowRunId);
3356
+ const sequence = (current?.sequence ?? 0) + 1;
3357
+ const id = genId();
3358
+ this.db.query(`INSERT INTO workflow_events (id, workflow_run_id, sequence, event_type, step_id, payload_json, created_at)
3359
+ VALUES ($id, $workflowRunId, $sequence, $eventType, $stepId, $payload, $created)`).run({
3360
+ $id: id,
3361
+ $workflowRunId: workflowRunId,
3362
+ $sequence: sequence,
3363
+ $eventType: eventType,
3364
+ $stepId: stepId ?? null,
3365
+ $payload: payload ? JSON.stringify(payload) : null,
3366
+ $created: now
3367
+ });
3368
+ const event = this.db.query("SELECT * FROM workflow_events WHERE id = ?").get(id);
3369
+ if (!event)
3370
+ throw new Error(`workflow event not found after append: ${id}`);
3371
+ return rowToWorkflowEvent(event);
2760
3372
  });
2761
- const event = this.db.query("SELECT * FROM workflow_events WHERE id = ?").get(id);
2762
- if (!event)
2763
- throw new Error(`workflow event not found after append: ${id}`);
2764
- return rowToWorkflowEvent(event);
2765
3373
  }
2766
3374
  listWorkflowEvents(workflowRunId, limit = 200) {
2767
3375
  const rows = this.db.query("SELECT * FROM workflow_events WHERE workflow_run_id = ? ORDER BY sequence ASC LIMIT ?").all(workflowRunId, limit);
@@ -2777,35 +3385,64 @@ class Store {
2777
3385
  }
2778
3386
  markRunPid(id, pid, claimedBy, opts = {}) {
2779
3387
  const now = (opts.now ?? new Date).toISOString();
2780
- const res = claimedBy ? this.db.query(`UPDATE loop_runs SET pid=$pid, updated_at=$updated
3388
+ const startedMs = processStartTimeMs(pid);
3389
+ const processStartedAt = startedMs === undefined ? null : new Date(startedMs).toISOString();
3390
+ const res = claimedBy ? this.db.query(`UPDATE loop_runs SET pid=$pid, process_started_at=$processStartedAt, updated_at=$updated
2781
3391
  WHERE id=$id AND status='running' AND claimed_by=$claimedBy
2782
3392
  AND ($daemonLeaseId IS NULL OR EXISTS (
2783
3393
  SELECT 1 FROM daemon_lease WHERE id=$daemonLeaseId AND expires_at > $now
2784
3394
  ))`).run({
2785
3395
  $id: id,
2786
3396
  $pid: pid,
3397
+ $processStartedAt: processStartedAt,
2787
3398
  $updated: now,
2788
3399
  $claimedBy: claimedBy,
2789
3400
  $daemonLeaseId: opts.daemonLeaseId ?? null,
2790
3401
  $now: now
2791
- }) : this.db.query(`UPDATE loop_runs SET pid=$pid, updated_at=$updated
3402
+ }) : this.db.query(`UPDATE loop_runs SET pid=$pid, process_started_at=$processStartedAt, updated_at=$updated
2792
3403
  WHERE id=$id AND status='running'
2793
3404
  AND ($daemonLeaseId IS NULL OR EXISTS (
2794
3405
  SELECT 1 FROM daemon_lease WHERE id=$daemonLeaseId AND expires_at > $now
2795
- ))`).run({ $id: id, $pid: pid, $updated: now, $daemonLeaseId: opts.daemonLeaseId ?? null, $now: now });
3406
+ ))`).run({
3407
+ $id: id,
3408
+ $pid: pid,
3409
+ $processStartedAt: processStartedAt,
3410
+ $updated: now,
3411
+ $daemonLeaseId: opts.daemonLeaseId ?? null,
3412
+ $now: now
3413
+ });
2796
3414
  if (res.changes !== 1)
2797
3415
  return;
2798
3416
  return this.getRun(id);
2799
3417
  }
3418
+ recordRunProcess(runId, info, opts = {}) {
3419
+ const now = (opts.now ?? new Date).toISOString();
3420
+ const res = this.db.query(`UPDATE loop_runs SET pid=$pid, pgid=$pgid, process_started_at=$processStartedAt, updated_at=$updated
3421
+ WHERE id=$id AND status='running'
3422
+ AND ($daemonLeaseId IS NULL OR EXISTS (
3423
+ SELECT 1 FROM daemon_lease WHERE id=$daemonLeaseId AND expires_at > $now
3424
+ ))`).run({
3425
+ $id: runId,
3426
+ $pid: info.pid,
3427
+ $pgid: info.pgid ?? null,
3428
+ $processStartedAt: info.processStartedAt ?? isoProcessStart(info.pid) ?? now,
3429
+ $updated: now,
3430
+ $daemonLeaseId: opts.daemonLeaseId ?? null,
3431
+ $now: now
3432
+ });
3433
+ if (res.changes !== 1)
3434
+ return;
3435
+ return this.getRun(runId);
3436
+ }
2800
3437
  hasLiveWorkflowStepProcesses(loopRunId) {
2801
- const liveWorkflowSteps = this.db.query(`SELECT wr.id AS workflow_run_id, wsr.step_id AS step_id, wsr.pid AS pid
3438
+ const liveWorkflowSteps = this.db.query(`SELECT wr.id AS workflow_run_id, wsr.step_id AS step_id, wsr.pid AS pid, wsr.started_at AS started_at
2802
3439
  FROM workflow_runs wr
2803
3440
  JOIN workflow_step_runs wsr ON wsr.workflow_run_id = wr.id
2804
3441
  WHERE wr.loop_run_id = ?
2805
3442
  AND wr.status NOT IN ('succeeded', 'failed', 'timed_out', 'cancelled')
2806
3443
  AND wsr.status = 'running'
2807
3444
  AND wsr.pid IS NOT NULL`).all(loopRunId);
2808
- return liveWorkflowSteps.some((step) => isProcessAlive(step.pid));
3445
+ return liveWorkflowSteps.some((step) => isLiveStepProcess(step.pid, step.started_at));
2809
3446
  }
2810
3447
  createSkippedRun(loop, scheduledFor, reason, opts = {}) {
2811
3448
  const now = nowIso();
@@ -2879,7 +3516,7 @@ class Store {
2879
3516
  const existing = this.getRunBySlot(loop.id, scheduledFor);
2880
3517
  if (existing) {
2881
3518
  if (existing.status === "running") {
2882
- if (existing.leaseExpiresAt && existing.leaseExpiresAt <= startedAt && existing.pid && isProcessAlive(existing.pid)) {
3519
+ if (existing.leaseExpiresAt && existing.leaseExpiresAt <= startedAt && isRecordedProcessAlive(existing.pid, existing.processStartedAt)) {
2883
3520
  this.db.exec("COMMIT");
2884
3521
  return;
2885
3522
  }
@@ -2888,7 +3525,7 @@ class Store {
2888
3525
  return;
2889
3526
  }
2890
3527
  const res3 = this.db.query(`UPDATE loop_runs SET status='running', started_at=$started, finished_at=NULL,
2891
- claimed_by=$claimedBy, lease_expires_at=$lease, pid=NULL, exit_code=NULL,
3528
+ claimed_by=$claimedBy, lease_expires_at=$lease, pid=NULL, pgid=NULL, process_started_at=NULL, exit_code=NULL,
2892
3529
  duration_ms=NULL, stdout=NULL, stderr=NULL, error=NULL, updated_at=$updated
2893
3530
  WHERE id=$id AND status='running' AND lease_expires_at <= $now`).run({
2894
3531
  $id: existing.id,
@@ -2910,7 +3547,7 @@ class Store {
2910
3547
  }
2911
3548
  const attempt = existing.attempt + 1;
2912
3549
  const res2 = this.db.query(`UPDATE loop_runs SET attempt=$attempt, status='running', started_at=$started, finished_at=NULL,
2913
- claimed_by=$claimedBy, lease_expires_at=$lease, pid=NULL, exit_code=NULL,
3550
+ claimed_by=$claimedBy, lease_expires_at=$lease, pid=NULL, pgid=NULL, process_started_at=NULL, exit_code=NULL,
2914
3551
  duration_ms=NULL, stdout=NULL, stderr=NULL, error=NULL, updated_at=$updated
2915
3552
  WHERE id=$id
2916
3553
  AND status IN ('failed', 'timed_out', 'abandoned')
@@ -2958,6 +3595,7 @@ class Store {
2958
3595
  }
2959
3596
  finalizeRun(id, patch, opts = {}) {
2960
3597
  const finishedAt = patch.finishedAt ?? nowIso();
3598
+ const error = patch.error === undefined ? undefined : scrubSecrets(patch.error);
2961
3599
  const params = {
2962
3600
  $id: id,
2963
3601
  $status: patch.status,
@@ -2965,41 +3603,43 @@ class Store {
2965
3603
  $pid: patch.pid ?? null,
2966
3604
  $exitCode: patch.exitCode ?? null,
2967
3605
  $durationMs: patch.durationMs ?? null,
2968
- $stdout: patch.stdout ?? null,
2969
- $stderr: patch.stderr ?? null,
2970
- $error: patch.error ?? null,
3606
+ $stdout: scrubbedOrNull(patch.stdout),
3607
+ $stderr: scrubbedOrNull(patch.stderr),
3608
+ $error: error ?? null,
2971
3609
  $updated: finishedAt,
2972
3610
  $claimedBy: opts.claimedBy ?? null,
2973
3611
  $now: (opts.now ?? new Date).toISOString(),
2974
3612
  $daemonLeaseId: opts.daemonLeaseId ?? null
2975
3613
  };
2976
- const res = opts.claimedBy ? this.db.query(`UPDATE loop_runs SET status=$status, finished_at=$finished, lease_expires_at=NULL, pid=$pid, exit_code=$exitCode,
2977
- duration_ms=$durationMs, stdout=$stdout, stderr=$stderr, error=$error, updated_at=$updated
2978
- WHERE id=$id AND status='running' AND claimed_by=$claimedBy AND lease_expires_at > $now
2979
- AND ($daemonLeaseId IS NULL OR EXISTS (
2980
- SELECT 1 FROM daemon_lease WHERE id=$daemonLeaseId AND expires_at > $now
2981
- ))`).run(params) : this.db.query(`UPDATE loop_runs SET status=$status, finished_at=$finished, lease_expires_at=NULL, pid=$pid, exit_code=$exitCode,
2982
- duration_ms=$durationMs, stdout=$stdout, stderr=$stderr, error=$error, updated_at=$updated WHERE id=$id`).run(params);
2983
- const run = this.getRun(id);
2984
- if (!run)
2985
- throw new Error(`run not found after finalize: ${id}`);
2986
- if (opts.claimedBy && res.changes !== 1)
2987
- return run;
2988
- if (res.changes === 1) {
2989
- this.setWorkflowWorkItemsForLoopRun(run, patch.error, finishedAt);
2990
- const loop = this.getLoop(run.loopId);
2991
- const itemStatus = workItemStatusForLoopRun(run.status, run.attempt, loop?.maxAttempts);
2992
- if (loop?.target.type === "workflow" && itemStatus && itemStatus !== "admitted") {
2993
- const workItemId = loop.target.input?.workflowWorkItemId ?? loop.target.input?.workItemId;
2994
- this.maybeArchiveGeneratedRouteWorkflow({
2995
- workflowId: loop.target.workflowId,
2996
- loopId: loop.id,
2997
- workItemId,
2998
- updated: finishedAt
2999
- });
3614
+ return this.transact(() => {
3615
+ const res = opts.claimedBy ? this.db.query(`UPDATE loop_runs SET status=$status, finished_at=$finished, lease_expires_at=NULL, pid=$pid, exit_code=$exitCode,
3616
+ duration_ms=$durationMs, stdout=$stdout, stderr=$stderr, error=$error, updated_at=$updated
3617
+ WHERE id=$id AND status='running' AND claimed_by=$claimedBy AND lease_expires_at > $now
3618
+ AND ($daemonLeaseId IS NULL OR EXISTS (
3619
+ SELECT 1 FROM daemon_lease WHERE id=$daemonLeaseId AND expires_at > $now
3620
+ ))`).run(params) : this.db.query(`UPDATE loop_runs SET status=$status, finished_at=$finished, lease_expires_at=NULL, pid=$pid, exit_code=$exitCode,
3621
+ duration_ms=$durationMs, stdout=$stdout, stderr=$stderr, error=$error, updated_at=$updated WHERE id=$id`).run(params);
3622
+ const run = this.getRun(id);
3623
+ if (!run)
3624
+ throw new Error(`run not found after finalize: ${id}`);
3625
+ if (opts.claimedBy && res.changes !== 1)
3626
+ return run;
3627
+ if (res.changes === 1) {
3628
+ this.setWorkflowWorkItemsForLoopRun(run, error, finishedAt);
3629
+ const loop = this.getLoop(run.loopId);
3630
+ const itemStatus = workItemStatusForLoopRun(run.status, run.attempt, loop?.maxAttempts);
3631
+ if (loop?.target.type === "workflow" && itemStatus && itemStatus !== "admitted") {
3632
+ const workItemId = loop.target.input?.workflowWorkItemId ?? loop.target.input?.workItemId;
3633
+ this.maybeArchiveGeneratedRouteWorkflow({
3634
+ workflowId: loop.target.workflowId,
3635
+ loopId: loop.id,
3636
+ workItemId,
3637
+ updated: finishedAt
3638
+ });
3639
+ }
3000
3640
  }
3001
- }
3002
- return run;
3641
+ return run;
3642
+ });
3003
3643
  }
3004
3644
  heartbeatRunLease(id, claimedBy, leaseMs, now = new Date, opts = {}) {
3005
3645
  const expiresAt = new Date(now.getTime() + leaseMs).toISOString();
@@ -3049,6 +3689,9 @@ class Store {
3049
3689
  });
3050
3690
  }
3051
3691
  recoverExpiredRunLeases(now = new Date, opts = {}) {
3692
+ return this.recoverExpiredRunLeasesDetailed(now, opts).abandoned;
3693
+ }
3694
+ recoverExpiredRunLeasesDetailed(now = new Date, opts = {}) {
3052
3695
  const limit = Math.max(1, Math.min(1000, Math.floor(opts.limit ?? DEFAULT_RECOVERY_BATCH_LIMIT)));
3053
3696
  const scanLimit = Math.max(limit, Math.min(5000, Math.floor(opts.scanLimit ?? limit * DEFAULT_RECOVERY_SCAN_MULTIPLIER)));
3054
3697
  const rows = this.db.query(`SELECT * FROM loop_runs
@@ -3056,15 +3699,15 @@ class Store {
3056
3699
  ORDER BY lease_expires_at ASC
3057
3700
  LIMIT ?`).all(now.toISOString(), scanLimit);
3058
3701
  const recovered = [];
3702
+ const deferred = [];
3059
3703
  for (const row of rows) {
3060
3704
  if (recovered.length >= limit)
3061
3705
  break;
3062
- if (row.pid && isProcessAlive(row.pid)) {
3063
- this.deferLiveExpiredRun(row.id, now, opts);
3064
- continue;
3065
- }
3066
- if (this.hasLiveWorkflowStepProcesses(row.id)) {
3706
+ if (isRecordedProcessAlive(row.pid, row.process_started_at) || this.hasLiveWorkflowStepProcesses(row.id)) {
3067
3707
  this.deferLiveExpiredRun(row.id, now, opts);
3708
+ const deferredRun = this.getRun(row.id);
3709
+ if (deferredRun)
3710
+ deferred.push(deferredRun);
3068
3711
  continue;
3069
3712
  }
3070
3713
  const finished = now.toISOString();
@@ -3148,7 +3791,7 @@ class Store {
3148
3791
  if (run)
3149
3792
  recovered.push(run);
3150
3793
  }
3151
- return recovered;
3794
+ return { abandoned: recovered, deferred };
3152
3795
  }
3153
3796
  expireLoops(now = new Date, opts = {}) {
3154
3797
  const rows = this.db.query("SELECT * FROM loops WHERE status = 'active' AND archived_at IS NULL AND expires_at IS NOT NULL AND expires_at <= ?").all(now.toISOString());
@@ -3181,6 +3824,83 @@ class Store {
3181
3824
  const row = status ? this.db.query("SELECT COUNT(*) AS count FROM loop_runs WHERE status = ?").get(status) : this.db.query("SELECT COUNT(*) AS count FROM loop_runs").get();
3182
3825
  return row?.count ?? 0;
3183
3826
  }
3827
+ pruneHistory(opts) {
3828
+ const { maxAgeDays, keepPerLoop } = opts;
3829
+ if (maxAgeDays === undefined && keepPerLoop === undefined) {
3830
+ throw new ValidationError("pruneHistory requires maxAgeDays and/or keepPerLoop");
3831
+ }
3832
+ if (maxAgeDays !== undefined && (!Number.isFinite(maxAgeDays) || maxAgeDays < 0)) {
3833
+ throw new ValidationError(`pruneHistory maxAgeDays must be a non-negative number: ${maxAgeDays}`);
3834
+ }
3835
+ if (keepPerLoop !== undefined && (!Number.isInteger(keepPerLoop) || keepPerLoop < 0)) {
3836
+ throw new ValidationError(`pruneHistory keepPerLoop must be a non-negative integer: ${keepPerLoop}`);
3837
+ }
3838
+ const now = opts.now ?? new Date;
3839
+ const dryRun = opts.dryRun ?? false;
3840
+ const cutoff = maxAgeDays === undefined ? undefined : new Date(now.getTime() - maxAgeDays * 86400000).toISOString();
3841
+ const terminal = TERMINAL_RUN_STATUSES.map((status) => `'${status}'`).join(",");
3842
+ const candidateIds = this.db.query(`WITH ranked AS (
3843
+ SELECT id, status, created_at,
3844
+ ROW_NUMBER() OVER (PARTITION BY loop_id ORDER BY created_at DESC, id DESC) AS recency
3845
+ FROM loop_runs
3846
+ )
3847
+ SELECT id FROM ranked
3848
+ WHERE status IN (${terminal})
3849
+ AND ($cutoff IS NULL OR created_at < $cutoff)
3850
+ AND ($keep IS NULL OR recency > $keep)`).all({ $cutoff: cutoff ?? null, $keep: keepPerLoop ?? null }).map((row) => row.id);
3851
+ const summary = {
3852
+ dryRun,
3853
+ cutoff,
3854
+ keepPerLoop,
3855
+ loopRuns: dryRun ? candidateIds.length : 0,
3856
+ workflowRuns: 0,
3857
+ goalRuns: 0
3858
+ };
3859
+ const manifestPaths = [];
3860
+ for (let offset = 0;offset < candidateIds.length; offset += PRUNE_BATCH_SIZE) {
3861
+ const batch = candidateIds.slice(offset, offset + PRUNE_BATCH_SIZE);
3862
+ const batchPlaceholders = batch.map(() => "?").join(",");
3863
+ if (dryRun) {
3864
+ const workflowRunIds = this.db.query(`SELECT id FROM workflow_runs WHERE loop_run_id IN (${batchPlaceholders})`).all(...batch).map((row) => row.id);
3865
+ const workflowPlaceholders = workflowRunIds.map(() => "?").join(",") || "''";
3866
+ summary.workflowRuns += workflowRunIds.length;
3867
+ summary.goalRuns += this.db.query(`SELECT COUNT(*) AS count FROM goal_runs
3868
+ WHERE loop_run_id IN (${batchPlaceholders}) OR workflow_run_id IN (${workflowPlaceholders})`).get(...batch, ...workflowRunIds)?.count ?? 0;
3869
+ continue;
3870
+ }
3871
+ this.transact(() => {
3872
+ const confirmed = this.db.query(`SELECT id FROM loop_runs WHERE id IN (${batchPlaceholders}) AND status IN (${terminal})`).all(...batch).map((row) => row.id);
3873
+ if (confirmed.length === 0)
3874
+ return;
3875
+ const runPlaceholders = confirmed.map(() => "?").join(",");
3876
+ const workflowRuns = this.db.query(`SELECT id, manifest_path FROM workflow_runs WHERE loop_run_id IN (${runPlaceholders})`).all(...confirmed);
3877
+ const workflowRunIds = workflowRuns.map((row) => row.id);
3878
+ const workflowPlaceholders = workflowRunIds.map(() => "?").join(",") || "''";
3879
+ summary.loopRuns += confirmed.length;
3880
+ summary.workflowRuns += workflowRunIds.length;
3881
+ summary.goalRuns += this.db.query(`DELETE FROM goal_runs WHERE loop_run_id IN (${runPlaceholders}) OR workflow_run_id IN (${workflowPlaceholders})`).run(...confirmed, ...workflowRunIds).changes;
3882
+ if (workflowRunIds.length > 0) {
3883
+ this.db.query(`DELETE FROM workflow_runs WHERE id IN (${workflowPlaceholders})`).run(...workflowRunIds);
3884
+ }
3885
+ this.db.query(`DELETE FROM loop_runs WHERE id IN (${runPlaceholders}) AND status IN (${terminal})`).run(...confirmed);
3886
+ for (const row of workflowRuns) {
3887
+ if (row.manifest_path)
3888
+ manifestPaths.push(row.manifest_path);
3889
+ }
3890
+ });
3891
+ }
3892
+ for (const manifestPath of manifestPaths) {
3893
+ const runDir = dirname2(manifestPath);
3894
+ try {
3895
+ if (/^[0-9a-f]{12,64}$/.test(basename2(runDir))) {
3896
+ rmSync2(runDir, { recursive: true, force: true });
3897
+ } else {
3898
+ rmSync2(manifestPath, { force: true });
3899
+ }
3900
+ } catch {}
3901
+ }
3902
+ return summary;
3903
+ }
3184
3904
  acquireDaemonLease(input) {
3185
3905
  const now = input.now ?? new Date;
3186
3906
  const expiresAt = new Date(now.getTime() + input.ttlMs).toISOString();
@@ -3243,16 +3963,245 @@ class Store {
3243
3963
  }
3244
3964
  }
3245
3965
 
3966
+ // src/lib/doctor.ts
3967
+ import { spawnSync as spawnSync5 } from "child_process";
3968
+ import { accessSync as accessSync2, constants as constants2 } from "fs";
3969
+
3970
+ // src/daemon/control.ts
3971
+ import { existsSync as existsSync2, mkdirSync as mkdirSync4, readFileSync as readFileSync3, rmSync as rmSync3, writeFileSync as writeFileSync2 } from "fs";
3972
+ import { spawnSync as spawnSync2 } from "child_process";
3973
+ import { hostname } from "os";
3974
+ import { dirname as dirname3, join as join4 } from "path";
3975
+
3976
+ // src/daemon/loop.ts
3977
+ function realSleep(ms) {
3978
+ return new Promise((resolve2) => setTimeout(resolve2, ms));
3979
+ }
3980
+ async function runLoop(opts) {
3981
+ const sleep = opts.sleep ?? realSleep;
3982
+ const sliceMs = opts.sliceMs ?? 200;
3983
+ while (!opts.shouldStop()) {
3984
+ try {
3985
+ await opts.tickFn();
3986
+ } catch (err) {
3987
+ opts.onTickError?.(err);
3988
+ }
3989
+ let waited = 0;
3990
+ while (waited < opts.intervalMs && !opts.shouldStop()) {
3991
+ const chunk = Math.min(sliceMs, opts.intervalMs - waited);
3992
+ await sleep(chunk);
3993
+ waited += chunk;
3994
+ }
3995
+ }
3996
+ }
3997
+
3998
+ // src/daemon/control.ts
3999
+ function readPidRecord(path = pidFilePath()) {
4000
+ if (!existsSync2(path))
4001
+ return;
4002
+ let raw;
4003
+ try {
4004
+ raw = readFileSync3(path, "utf8").trim();
4005
+ } catch {
4006
+ return;
4007
+ }
4008
+ if (!raw)
4009
+ return;
4010
+ if (raw.startsWith("{")) {
4011
+ try {
4012
+ const parsed = JSON.parse(raw);
4013
+ const pid2 = Number(parsed.pid);
4014
+ if (!Number.isInteger(pid2) || pid2 <= 0)
4015
+ return;
4016
+ const startedAt = Number(parsed.startedAt);
4017
+ return { pid: pid2, startedAt: Number.isFinite(startedAt) ? startedAt : undefined };
4018
+ } catch {
4019
+ return;
4020
+ }
4021
+ }
4022
+ const pid = Number(raw);
4023
+ return Number.isInteger(pid) && pid > 0 ? { pid } : undefined;
4024
+ }
4025
+ function writePid(pid = process.pid, path = pidFilePath()) {
4026
+ mkdirSync4(dirname3(path), { recursive: true, mode: 448 });
4027
+ writeFileSync2(path, JSON.stringify({ pid, startedAt: processStartTimeMs(pid) }));
4028
+ }
4029
+ function removePid(path = pidFilePath()) {
4030
+ rmSync3(path, { force: true });
4031
+ }
4032
+ function isAlive(pid, startedAt) {
4033
+ try {
4034
+ process.kill(pid, 0);
4035
+ } catch (err) {
4036
+ if (err.code !== "EPERM")
4037
+ return false;
4038
+ }
4039
+ return sameProcessStart(startedAt, processStartTimeMs(pid));
4040
+ }
4041
+ function isDaemonRunning(path = pidFilePath()) {
4042
+ const record = readPidRecord(path);
4043
+ if (!record)
4044
+ return { running: false, stale: false };
4045
+ if (isAlive(record.pid, record.startedAt))
4046
+ return { running: true, stale: false, pid: record.pid };
4047
+ return { running: false, stale: true, pid: record.pid };
4048
+ }
4049
+ function toReapableProcess(run) {
4050
+ return { pid: run.pid, pgid: run.pgid, processStartedAt: run.processStartedAt };
4051
+ }
4052
+ function ownProcessGroupId() {
4053
+ if (process.platform === "linux") {
4054
+ const fields = procStatFields("/proc/self/stat");
4055
+ const pgrp = fields ? Number(fields[2]) : Number.NaN;
4056
+ if (Number.isInteger(pgrp) && pgrp > 0)
4057
+ return pgrp;
4058
+ }
4059
+ try {
4060
+ const run = spawnSync2("ps", ["-o", "pgid=", "-p", String(process.pid)], { encoding: "utf8" });
4061
+ const pgid = Number(run.stdout.trim());
4062
+ if (run.status === 0 && Number.isInteger(pgid) && pgid > 0)
4063
+ return pgid;
4064
+ } catch {}
4065
+ return;
4066
+ }
4067
+ function signalReapTarget(target, signal) {
4068
+ try {
4069
+ process.kill(target.group ? -target.id : target.id, signal);
4070
+ return true;
4071
+ } catch (err) {
4072
+ return err.code === "EPERM";
4073
+ }
4074
+ }
4075
+ function processExists(pid) {
4076
+ try {
4077
+ process.kill(pid, 0);
4078
+ return true;
4079
+ } catch (err) {
4080
+ return err.code === "EPERM";
4081
+ }
4082
+ }
4083
+ async function reapProcessGroups(entries, opts = {}) {
4084
+ const sleep = opts.sleep ?? realSleep;
4085
+ const graceMs = Math.max(100, opts.graceMs ?? 2000);
4086
+ const ownPgid = ownProcessGroupId();
4087
+ const targets = new Map;
4088
+ for (const entry of entries) {
4089
+ const leaderPid = entry.pid ?? entry.pgid;
4090
+ if (leaderPid !== undefined && processExists(leaderPid)) {
4091
+ if (entry.processStartedAt === undefined) {
4092
+ opts.log?.(`skipping reap of pid ${leaderPid}: no recorded start-time fingerprint to verify identity`);
4093
+ continue;
4094
+ }
4095
+ if (!verifiedProcessStart(entry.processStartedAt, processStartTimeMs(leaderPid))) {
4096
+ opts.log?.(`skipping reap of pid ${leaderPid}: start-time fingerprint mismatch or unverifiable (pid recycled)`);
4097
+ continue;
4098
+ }
4099
+ }
4100
+ const usableGroup = entry.pgid !== undefined && Number.isInteger(entry.pgid) && entry.pgid > 1 && entry.pgid !== process.pid && entry.pgid !== ownPgid;
4101
+ const usablePid = entry.pid !== undefined && Number.isInteger(entry.pid) && entry.pid > 1 && entry.pid !== process.pid;
4102
+ const target = usableGroup ? { id: entry.pgid, group: true } : usablePid ? { id: entry.pid, group: false } : undefined;
4103
+ if (!target)
4104
+ continue;
4105
+ targets.set(`${target.group ? "g" : "p"}:${target.id}`, target);
4106
+ }
4107
+ const live = [...targets.values()].filter((target) => signalReapTarget(target, "SIGTERM"));
4108
+ if (live.length === 0)
4109
+ return [];
4110
+ const steps = Math.max(1, Math.ceil(graceMs / 100));
4111
+ let remaining = live;
4112
+ for (let i = 0;i < steps && remaining.length > 0; i++) {
4113
+ await sleep(100);
4114
+ remaining = remaining.filter((target) => signalReapTarget(target, 0));
4115
+ }
4116
+ for (const target of remaining) {
4117
+ signalReapTarget(target, "SIGKILL");
4118
+ opts.log?.(`escalated to SIGKILL for ${target.group ? "pgid" : "pid"} ${target.id}`);
4119
+ }
4120
+ return live.map((target) => target.id);
4121
+ }
4122
+ function daemonStatus(store, path = pidFilePath()) {
4123
+ return {
4124
+ ...isDaemonRunning(path),
4125
+ lease: store.getDaemonLease(),
4126
+ host: hostname(),
4127
+ loops: {
4128
+ total: store.countLoops(),
4129
+ active: store.countLoops("active"),
4130
+ paused: store.countLoops("paused"),
4131
+ stopped: store.countLoops("stopped"),
4132
+ expired: store.countLoops("expired"),
4133
+ archived: store.countLoops(undefined, { archived: true })
4134
+ },
4135
+ runs: {
4136
+ total: store.countRuns(),
4137
+ running: store.countRuns("running"),
4138
+ failed: store.countRuns("failed"),
4139
+ succeeded: store.countRuns("succeeded"),
4140
+ abandoned: store.countRuns("abandoned")
4141
+ },
4142
+ logPath: daemonLogPath()
4143
+ };
4144
+ }
4145
+ async function stopDaemon(opts = {}) {
4146
+ const path = opts.path ?? pidFilePath();
4147
+ const record = readPidRecord(path);
4148
+ const state = isDaemonRunning(path);
4149
+ if (state.stale) {
4150
+ removePid(path);
4151
+ return { wasRunning: false, stopped: false, forced: false, pid: state.pid };
4152
+ }
4153
+ if (!state.running || !state.pid)
4154
+ return { wasRunning: false, stopped: false, forced: false };
4155
+ const sleep = opts.sleep ?? realSleep;
4156
+ const store = new Store(join4(dirname3(path), "loops.db"));
4157
+ try {
4158
+ const lease = store.getDaemonLease();
4159
+ if (!lease || lease.pid !== state.pid || new Date(lease.expiresAt).getTime() <= Date.now()) {
4160
+ removePid(path);
4161
+ return { wasRunning: false, stopped: false, forced: false, pid: state.pid };
4162
+ }
4163
+ try {
4164
+ process.kill(state.pid, "SIGTERM");
4165
+ } catch {
4166
+ removePid(path);
4167
+ return { wasRunning: true, stopped: true, forced: false, pid: state.pid };
4168
+ }
4169
+ const steps = Math.max(1, Math.ceil((opts.timeoutMs ?? 6000) / 100));
4170
+ for (let i = 0;i < steps; i++) {
4171
+ await sleep(100);
4172
+ if (!isAlive(state.pid, record?.startedAt)) {
4173
+ removePid(path);
4174
+ return { wasRunning: true, stopped: true, forced: false, pid: state.pid };
4175
+ }
4176
+ }
4177
+ try {
4178
+ process.kill(state.pid, "SIGKILL");
4179
+ } catch {}
4180
+ await sleep(150);
4181
+ removePid(path);
4182
+ const ownedRuns = store.listRuns({ status: "running", limit: 1000 }).filter((run) => run.claimedBy !== undefined && run.claimedBy.endsWith(`:${lease.id}`));
4183
+ const reapedPgids = await reapProcessGroups(ownedRuns.map(toReapableProcess), {
4184
+ sleep,
4185
+ graceMs: opts.reapGraceMs
4186
+ });
4187
+ return { wasRunning: true, stopped: !isAlive(state.pid, record?.startedAt), forced: true, pid: state.pid, reapedPgids };
4188
+ } finally {
4189
+ store.close();
4190
+ }
4191
+ }
4192
+
3246
4193
  // src/lib/executor.ts
3247
- import { spawn, spawnSync as spawnSync2 } from "child_process";
4194
+ import { spawn as spawn2, spawnSync as spawnSync4 } from "child_process";
3248
4195
  import { randomBytes as randomBytes2 } from "crypto";
3249
4196
  import { once } from "events";
3250
- import { resolveMachineCommand } from "@hasna/machines/consumer";
4197
+ import { lstatSync, mkdirSync as mkdirSync5, realpathSync } from "fs";
4198
+ import { dirname as dirname4, resolve as resolve2 } from "path";
3251
4199
 
3252
4200
  // src/lib/accounts.ts
3253
- import { spawnSync } from "child_process";
3254
- import { existsSync as existsSync2 } from "fs";
4201
+ import { spawnSync as spawnSync3 } from "child_process";
4202
+ import { existsSync as existsSync3 } from "fs";
3255
4203
  var EXPORT_RE = /^export\s+([A-Za-z_][A-Za-z0-9_]*)=(.*)$/;
4204
+ var ACCOUNTS_ENV_TIMEOUT_MS = 30000;
3256
4205
  function accountToolForProvider(provider) {
3257
4206
  switch (provider) {
3258
4207
  case "claude":
@@ -3314,21 +4263,17 @@ function accountDirEnvVar(tool) {
3314
4263
  return;
3315
4264
  }
3316
4265
  }
3317
- function resolveAccountEnv(account, toolHint, env) {
3318
- if (!account)
3319
- return {};
4266
+ function requiredAccountTool(account, toolHint) {
3320
4267
  const tool = account.tool ?? toolHint;
3321
4268
  if (!tool)
3322
4269
  throw new Error("account.tool is required when no provider tool can be inferred");
3323
- const result = spawnSync("accounts", ["env", account.profile, "--tool", tool], {
3324
- encoding: "utf8",
3325
- env,
3326
- stdio: ["ignore", "pipe", "pipe"]
3327
- });
4270
+ return tool;
4271
+ }
4272
+ function accountEnvFromResult(account, tool, result) {
3328
4273
  if (result.error) {
3329
- throw new Error(`failed to run accounts env for ${account.profile}/${tool}: ${result.error.message}`);
4274
+ throw new Error(`failed to run accounts env for ${account.profile}/${tool}: ${result.error}`);
3330
4275
  }
3331
- if ((result.status ?? 0) !== 0) {
4276
+ if ((result.status ?? 1) !== 0) {
3332
4277
  const stderr = result.stderr.trim();
3333
4278
  throw new Error(`accounts env failed for ${account.profile}/${tool}${stderr ? `: ${stderr}` : ""}`);
3334
4279
  }
@@ -3336,7 +4281,7 @@ function resolveAccountEnv(account, toolHint, env) {
3336
4281
  const profileDir = (accountDirEnvVar(tool) ? accountEnv[accountDirEnvVar(tool)] : undefined) ?? primaryAccountDir(result.stdout);
3337
4282
  if (!profileDir)
3338
4283
  throw new Error(`accounts env returned no profile directory for ${account.profile}/${tool}`);
3339
- if (!existsSync2(profileDir))
4284
+ if (!existsSync3(profileDir))
3340
4285
  throw new Error(`account profile directory does not exist for ${account.profile}/${tool}: ${profileDir}`);
3341
4286
  return {
3342
4287
  ...accountEnv,
@@ -3344,11 +4289,38 @@ function resolveAccountEnv(account, toolHint, env) {
3344
4289
  LOOPS_ACCOUNT_TOOL: tool
3345
4290
  };
3346
4291
  }
4292
+ async function resolveAccountEnv(account, toolHint, env) {
4293
+ if (!account)
4294
+ return {};
4295
+ const tool = requiredAccountTool(account, toolHint);
4296
+ const result = await spawnCapture("accounts", ["env", account.profile, "--tool", tool], {
4297
+ env,
4298
+ timeoutMs: ACCOUNTS_ENV_TIMEOUT_MS
4299
+ });
4300
+ return accountEnvFromResult(account, tool, result);
4301
+ }
4302
+ function resolveAccountEnvSync(account, toolHint, env) {
4303
+ if (!account)
4304
+ return {};
4305
+ const tool = requiredAccountTool(account, toolHint);
4306
+ const result = spawnSync3("accounts", ["env", account.profile, "--tool", tool], {
4307
+ encoding: "utf8",
4308
+ env,
4309
+ stdio: ["ignore", "pipe", "pipe"],
4310
+ timeout: ACCOUNTS_ENV_TIMEOUT_MS
4311
+ });
4312
+ return accountEnvFromResult(account, tool, {
4313
+ status: result.status,
4314
+ stdout: result.stdout ?? "",
4315
+ stderr: result.stderr ?? "",
4316
+ error: result.error?.message
4317
+ });
4318
+ }
3347
4319
 
3348
4320
  // src/lib/env.ts
3349
4321
  import { accessSync, constants } from "fs";
3350
4322
  import { homedir as homedir2 } from "os";
3351
- import { delimiter, join as join4 } from "path";
4323
+ import { delimiter, join as join5 } from "path";
3352
4324
  function compactPathParts(parts) {
3353
4325
  const seen = new Set;
3354
4326
  const result = [];
@@ -3364,14 +4336,14 @@ function compactPathParts(parts) {
3364
4336
  function commonExecutableDirs(env = process.env) {
3365
4337
  const home = env.HOME || homedir2();
3366
4338
  return compactPathParts([
3367
- join4(home, ".local", "bin"),
3368
- join4(home, ".bun", "bin"),
3369
- join4(home, ".cargo", "bin"),
3370
- join4(home, ".npm-global", "bin"),
3371
- join4(home, "bin"),
3372
- env.BUN_INSTALL ? join4(env.BUN_INSTALL, "bin") : undefined,
4339
+ join5(home, ".local", "bin"),
4340
+ join5(home, ".bun", "bin"),
4341
+ join5(home, ".cargo", "bin"),
4342
+ join5(home, ".npm-global", "bin"),
4343
+ join5(home, "bin"),
4344
+ env.BUN_INSTALL ? join5(env.BUN_INSTALL, "bin") : undefined,
3373
4345
  env.PNPM_HOME,
3374
- env.NPM_CONFIG_PREFIX ? join4(env.NPM_CONFIG_PREFIX, "bin") : undefined,
4346
+ env.NPM_CONFIG_PREFIX ? join5(env.NPM_CONFIG_PREFIX, "bin") : undefined,
3375
4347
  "/opt/homebrew/bin",
3376
4348
  "/usr/local/bin",
3377
4349
  "/usr/bin",
@@ -3395,7 +4367,7 @@ function executableExists(command, env = process.env) {
3395
4367
  if (command.includes("/"))
3396
4368
  return isExecutable(command);
3397
4369
  for (const dir of (env.PATH ?? "").split(delimiter)) {
3398
- if (dir && isExecutable(join4(dir, command)))
4370
+ if (dir && isExecutable(join5(dir, command)))
3399
4371
  return true;
3400
4372
  }
3401
4373
  return false;
@@ -3405,10 +4377,20 @@ function commandNotFoundMessage(command, env = process.env) {
3405
4377
  }
3406
4378
 
3407
4379
  // src/lib/machines.ts
3408
- import {
3409
- discoverMachineTopology,
3410
- resolveMachineRoute
3411
- } from "@hasna/machines/consumer";
4380
+ import { createRequire } from "module";
4381
+ var consumerModule;
4382
+ function machinesConsumer() {
4383
+ if (!consumerModule) {
4384
+ try {
4385
+ const resolved = Bun.resolveSync("@hasna/machines/consumer", import.meta.dir);
4386
+ consumerModule = createRequire(import.meta.url)(resolved);
4387
+ } catch (error) {
4388
+ const detail = error instanceof Error ? error.message : String(error);
4389
+ throw new Error(`@hasna/machines is not available; install the optional dependency to use machine-assigned loops: ${detail}`);
4390
+ }
4391
+ }
4392
+ return consumerModule;
4393
+ }
3412
4394
  function compact(value) {
3413
4395
  const text = value?.trim();
3414
4396
  return text ? text : undefined;
@@ -3445,14 +4427,18 @@ function machineFromRoute(route, topology) {
3445
4427
  };
3446
4428
  }
3447
4429
  function listOpenMachines() {
3448
- const topology = discoverMachineTopology();
4430
+ const topology = machinesConsumer().discoverMachineTopology();
3449
4431
  return topology.machines.map((entry) => entryToSummary(entry, topology));
3450
4432
  }
3451
4433
  function resolveLoopMachine(machineId) {
3452
- const topology = discoverMachineTopology();
3453
- const route = resolveMachineRoute(machineId, { topology });
4434
+ const consumer = machinesConsumer();
4435
+ const topology = consumer.discoverMachineTopology();
4436
+ const route = consumer.resolveMachineRoute(machineId, { topology });
3454
4437
  return machineFromRoute(route, topology);
3455
4438
  }
4439
+ function resolveMachineCommand(machineId, command) {
4440
+ return machinesConsumer().resolveMachineCommand(machineId, command);
4441
+ }
3456
4442
  function refreshLoopMachine(machine) {
3457
4443
  return resolveLoopMachine(machine.id);
3458
4444
  }
@@ -3460,6 +4446,10 @@ function refreshLoopMachine(machine) {
3460
4446
  // src/lib/executor.ts
3461
4447
  var DEFAULT_TIMEOUT_MS = 30 * 60000;
3462
4448
  var DEFAULT_MAX_OUTPUT_BYTES = 256 * 1024;
4449
+ var DEFAULT_AGENT_IDLE_TIMEOUT_MS = 30 * 60000;
4450
+ var BUFFERED_OUTPUT_PROVIDERS = new Set(["claude", "codewith", "opencode", "aicopilot"]);
4451
+ var DEFAULT_BUFFERED_AGENT_IDLE_TIMEOUT_MS = 4 * 60 * 60000;
4452
+ var WORKTREE_GIT_TIMEOUT_MS = 5 * 60000;
3463
4453
  var AUTH_ENV_KEYS = [
3464
4454
  "CLAUDE_CONFIG_DIR",
3465
4455
  "CODEWITH_HOME",
@@ -3495,34 +4485,39 @@ var TRANSPORT_ENV_KEYS = new Set([
3495
4485
  "USER",
3496
4486
  "XDG_RUNTIME_DIR"
3497
4487
  ]);
3498
- function appendBounded(current, chunk, maxBytes) {
3499
- const next = current + chunk.toString("utf8");
3500
- if (Buffer.byteLength(next, "utf8") <= maxBytes)
3501
- return next;
3502
- const overflow = Buffer.byteLength(next, "utf8") - maxBytes;
3503
- return `[truncated ${overflow} bytes]
3504
- ${next.slice(-maxBytes)}`;
3505
- }
3506
- function appendBoundedText(current, chunk, maxBytes) {
3507
- return appendBounded(current, Buffer.from(chunk, "utf8"), maxBytes);
3508
- }
3509
- function killProcessGroup(pid) {
3510
- try {
3511
- process.kill(-pid, "SIGTERM");
3512
- } catch {
3513
- try {
3514
- process.kill(pid, "SIGTERM");
3515
- } catch {}
3516
- }
3517
- setTimeout(() => {
3518
- try {
3519
- process.kill(-pid, "SIGKILL");
3520
- } catch {
3521
- try {
3522
- process.kill(pid, "SIGKILL");
3523
- } catch {}
3524
- }
3525
- }, 2000).unref();
4488
+ function boundedText(text, maxBytes) {
4489
+ const buffer = new BoundedOutputBuffer(maxBytes);
4490
+ buffer.append(text);
4491
+ return buffer.value();
4492
+ }
4493
+ function buildResult(status, startedAt, fields = {}) {
4494
+ const finishedAt = fields.finishedAt ?? nowIso();
4495
+ return {
4496
+ status,
4497
+ exitCode: fields.exitCode,
4498
+ stdout: fields.stdout ?? "",
4499
+ stderr: fields.stderr ?? "",
4500
+ error: fields.error,
4501
+ pid: fields.pid,
4502
+ startedAt,
4503
+ finishedAt,
4504
+ durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4505
+ };
4506
+ }
4507
+ function failureResult(startedAt, error, fields = {}) {
4508
+ return buildResult("failed", startedAt, { ...fields, error });
4509
+ }
4510
+ function timeoutResult(startedAt, error, fields = {}) {
4511
+ return buildResult("timed_out", startedAt, { ...fields, error });
4512
+ }
4513
+ function successResult(startedAt, fields = {}) {
4514
+ return buildResult("succeeded", startedAt, fields);
4515
+ }
4516
+ function notifySpawn(pid, opts) {
4517
+ if (!pid)
4518
+ return;
4519
+ opts.onSpawn?.(pid);
4520
+ opts.onSpawnProcess?.({ pid, pgid: pid, processStartedAt: nowIso() });
3526
4521
  }
3527
4522
  function shellQuote(value) {
3528
4523
  return `'${value.replace(/'/g, `'\\''`)}'`;
@@ -3578,229 +4573,21 @@ function allowlistEnv(allowlist) {
3578
4573
  env.LOOPS_AGENT_ALLOWLIST_ENFORCEMENT = "metadata_only";
3579
4574
  return env;
3580
4575
  }
3581
- function providerCommand(provider) {
3582
- switch (provider) {
3583
- case "claude":
3584
- return "claude";
3585
- case "cursor":
3586
- return "sh";
3587
- case "codewith":
3588
- return "codewith";
3589
- case "aicopilot":
3590
- return "aicopilot";
3591
- case "opencode":
3592
- return "opencode";
3593
- case "codex":
3594
- return "codex";
3595
- }
3596
- }
3597
- function codewithLikeSandbox(target) {
3598
- const sandbox = target.sandbox ?? (target.permissionMode === "bypass" ? "danger-full-access" : "workspace-write");
3599
- if (sandbox !== "read-only" && sandbox !== "workspace-write" && sandbox !== "danger-full-access") {
3600
- throw new Error(`${target.provider} sandbox must be read-only, workspace-write, or danger-full-access`);
3601
- }
3602
- return sandbox;
3603
- }
3604
- function configStringValue(value) {
3605
- return JSON.stringify(value);
3606
- }
3607
- var UNSAFE_CODEWITH_DURABLE_EXTRA_ARGS2 = new Set([
3608
- "e",
3609
- "exec",
3610
- "agent",
3611
- "start",
3612
- "--ephemeral",
3613
- "--ignore-rules",
3614
- "--skip-git-repo-check",
3615
- "--json",
3616
- "--output-last-message",
3617
- "-o",
3618
- "--output-schema",
3619
- "--dangerously-bypass-approvals-and-sandbox"
3620
- ]);
3621
- function assertStringOption(value, label) {
3622
- if (value !== undefined && typeof value !== "string")
3623
- throw new Error(`${label} must be a string`);
3624
- }
3625
- function assertSupportedAgentOptions(target) {
3626
- assertStringOption(target.variant, `${target.provider}.variant`);
3627
- assertStringOption(target.model, `${target.provider}.model`);
3628
- assertStringOption(target.agent, `${target.provider}.agent`);
3629
- assertStringOption(target.authProfile, `${target.provider}.authProfile`);
3630
- assertStringOption(target.configIsolation, `${target.provider}.configIsolation`);
3631
- if (target.provider === "opencode" && (target.model === undefined || target.model.trim() === "")) {
3632
- throw new Error("opencode.model is required; pass a provider/model id such as openrouter/google/gemini-2.5-flash");
3633
- }
3634
- if (target.configIsolation !== undefined && target.configIsolation !== "safe" && target.configIsolation !== "none") {
3635
- throw new Error(`${target.provider}.configIsolation must be safe or none`);
3636
- }
3637
- if (target.authProfile !== undefined && target.provider !== "codewith") {
3638
- throw new Error(`${target.provider}.authProfile is supported only for codewith`);
3639
- }
3640
- if (target.addDirs?.length && !["codewith", "codex"].includes(target.provider)) {
3641
- throw new Error(`${target.provider}.addDirs is currently supported only for codewith or codex`);
3642
- }
3643
- if (target.permissionMode && !["default", "plan", "auto", "bypass"].includes(target.permissionMode)) {
3644
- throw new Error(`${target.provider}.permissionMode must be default, plan, auto, or bypass`);
3645
- }
3646
- if (target.sandbox && !["read-only", "workspace-write", "danger-full-access", "enabled", "disabled"].includes(target.sandbox)) {
3647
- throw new Error(`${target.provider}.sandbox is not supported: ${target.sandbox}`);
3648
- }
3649
- if (target.provider === "codex" && target.agent !== undefined)
3650
- throw new Error("codex.agent is not supported");
3651
- if (target.provider === "codewith" && target.agent !== undefined) {
3652
- throw new Error("codewith.agent is not supported by the durable background-agent adapter");
3653
- }
3654
- if (target.provider === "codewith") {
3655
- const unsafe = target.extraArgs?.find((arg) => UNSAFE_CODEWITH_DURABLE_EXTRA_ARGS2.has(arg));
3656
- if (unsafe)
3657
- throw new Error(`codewith.extraArgs cannot include ${unsafe}; durable agent steps use codewith agent start, not exec/ephemeral flags`);
3658
- }
3659
- if (["codewith", "codex"].includes(target.provider)) {
3660
- if (target.permissionMode && !["default", "bypass"].includes(target.permissionMode)) {
3661
- throw new Error(`${target.provider}.permissionMode supports only default or bypass`);
3662
- }
3663
- if (target.sandbox)
3664
- codewithLikeSandbox(target);
3665
- return;
3666
- }
3667
- if (target.provider === "claude") {
3668
- if (target.sandbox !== undefined)
3669
- throw new Error("claude.sandbox is not supported");
4576
+ function defaultAgentIdleTimeoutMs(target, opts) {
4577
+ if (target.timeoutMs !== undefined || target.idleTimeoutMs !== undefined)
3670
4578
  return;
4579
+ const raw = opts.env?.LOOPS_AGENT_IDLE_TIMEOUT_MS ?? process.env.LOOPS_AGENT_IDLE_TIMEOUT_MS;
4580
+ if (raw !== undefined && raw !== "") {
4581
+ const normalized = raw.trim().toLowerCase();
4582
+ if (normalized === "0" || normalized === "none" || normalized === "off")
4583
+ return;
4584
+ const parsed = Number(normalized);
4585
+ if (Number.isFinite(parsed) && parsed > 0)
4586
+ return parsed;
3671
4587
  }
3672
- if (target.provider === "cursor") {
3673
- if (target.variant !== undefined)
3674
- throw new Error("cursor.variant is not supported");
3675
- if (target.permissionMode === "auto")
3676
- throw new Error("cursor.permissionMode auto is not supported; use provider-specific extraArgs for Cursor auto-review");
3677
- if (target.sandbox !== undefined && target.sandbox !== "enabled" && target.sandbox !== "disabled") {
3678
- throw new Error("cursor.sandbox must be enabled or disabled");
3679
- }
3680
- return;
3681
- }
3682
- if (target.permissionMode && !["default", "bypass"].includes(target.permissionMode)) {
3683
- throw new Error(`${target.provider}.permissionMode supports only default or bypass`);
3684
- }
3685
- if (target.sandbox !== undefined)
3686
- throw new Error(`${target.provider}.sandbox is not supported`);
3687
- }
3688
- function agentArgs(target) {
3689
- assertSupportedAgentOptions(target);
3690
- const isolation = target.configIsolation ?? "safe";
3691
- const permissionMode = target.permissionMode ?? "default";
3692
- const args = [];
3693
- switch (target.provider) {
3694
- case "claude":
3695
- if (isolation === "safe")
3696
- args.push("--safe-mode", "--setting-sources", "local", "--no-session-persistence");
3697
- if (permissionMode !== "default") {
3698
- const mode = permissionMode === "bypass" ? "bypassPermissions" : permissionMode === "plan" || permissionMode === "auto" ? permissionMode : undefined;
3699
- if (mode)
3700
- args.push("--permission-mode", mode);
3701
- }
3702
- args.push("-p", "--output-format", "json");
3703
- if (target.model)
3704
- args.push("--model", target.model);
3705
- if (target.variant)
3706
- args.push("--effort", target.variant);
3707
- if (target.agent)
3708
- args.push("--agent", target.agent);
3709
- args.push(...target.extraArgs ?? []);
3710
- return args;
3711
- case "cursor":
3712
- args.push("-c", [
3713
- "set -eu",
3714
- "if command -v agent >/dev/null 2>&1; then",
3715
- ' exec agent "$@"',
3716
- "else",
3717
- " echo 'Executable not found in PATH: agent' >&2",
3718
- " exit 127",
3719
- "fi"
3720
- ].join(`
3721
- `), "openloops-cursor", "-p");
3722
- if (permissionMode === "plan")
3723
- args.push("--mode", "plan");
3724
- if (permissionMode === "bypass")
3725
- args.push("--force");
3726
- const cursorSandbox = target.sandbox ?? (isolation === "safe" ? "enabled" : undefined);
3727
- if (cursorSandbox) {
3728
- if (cursorSandbox !== "enabled" && cursorSandbox !== "disabled")
3729
- throw new Error("cursor sandbox must be enabled or disabled");
3730
- args.push("--sandbox", cursorSandbox);
3731
- }
3732
- if (target.model)
3733
- args.push("--model", target.model);
3734
- if (target.agent)
3735
- args.push("--agent", target.agent);
3736
- args.push(...target.extraArgs ?? []);
3737
- return args;
3738
- case "codewith":
3739
- args.push(...target.authProfile ? ["--auth-profile", target.authProfile] : []);
3740
- if (target.variant)
3741
- args.push("-c", `model_reasoning_effort=${configStringValue(target.variant)}`);
3742
- args.push("--ask-for-approval", "never", "--sandbox", codewithLikeSandbox(target));
3743
- if (target.cwd)
3744
- args.push("--cd", target.cwd);
3745
- for (const dir of target.addDirs ?? [])
3746
- args.push("--add-dir", dir);
3747
- if (target.model)
3748
- args.push("--model", target.model);
3749
- args.push(...target.extraArgs ?? []);
3750
- args.push("agent", "start");
3751
- if (target.cwd)
3752
- args.push("--cwd", target.cwd);
3753
- args.push(target.prompt);
3754
- return args;
3755
- case "codex":
3756
- if (target.variant)
3757
- args.push("-c", `model_reasoning_effort=${configStringValue(target.variant)}`);
3758
- args.push("--ask-for-approval", "never", "exec", "--json", "--ephemeral", "--sandbox", codewithLikeSandbox(target), "--skip-git-repo-check");
3759
- if (isolation === "safe")
3760
- args.push("--ignore-rules");
3761
- if (target.cwd)
3762
- args.push("--cd", target.cwd);
3763
- for (const dir of target.addDirs ?? [])
3764
- args.push("--add-dir", dir);
3765
- if (target.model)
3766
- args.push("--model", target.model);
3767
- args.push(...target.extraArgs ?? []);
3768
- return args;
3769
- case "aicopilot":
3770
- args.push("run", "--format", "json");
3771
- if (isolation === "safe")
3772
- args.push("--pure");
3773
- if (permissionMode === "bypass")
3774
- args.push("--dangerously-skip-permissions");
3775
- if (target.cwd)
3776
- args.push("--dir", target.cwd);
3777
- if (target.model)
3778
- args.push("--model", target.model);
3779
- if (target.variant)
3780
- args.push("--variant", target.variant);
3781
- if (target.agent)
3782
- args.push("--agent", target.agent);
3783
- args.push(...target.extraArgs ?? []);
3784
- return args;
3785
- case "opencode":
3786
- args.push("run", "--format", "json");
3787
- if (isolation === "safe")
3788
- args.push("--pure");
3789
- if (permissionMode === "bypass")
3790
- args.push("--dangerously-skip-permissions");
3791
- if (target.cwd)
3792
- args.push("--dir", target.cwd);
3793
- if (target.model)
3794
- args.push("--model", target.model);
3795
- if (target.variant)
3796
- args.push("--variant", target.variant);
3797
- if (target.agent)
3798
- args.push("--agent", target.agent);
3799
- args.push(...target.extraArgs ?? []);
3800
- return args;
3801
- }
4588
+ return BUFFERED_OUTPUT_PROVIDERS.has(target.provider) ? DEFAULT_BUFFERED_AGENT_IDLE_TIMEOUT_MS : DEFAULT_AGENT_IDLE_TIMEOUT_MS;
3802
4589
  }
3803
- function commandSpec(target) {
4590
+ function commandSpec(target, opts) {
3804
4591
  if (target.type === "command") {
3805
4592
  const commandTarget = target;
3806
4593
  return {
@@ -3816,25 +4603,27 @@ function commandSpec(target) {
3816
4603
  };
3817
4604
  }
3818
4605
  const agentTarget = target;
4606
+ const adapter = providerAdapter(agentTarget.provider);
4607
+ const invocation = adapter.buildInvocation(agentTarget);
3819
4608
  return {
3820
- command: providerCommand(agentTarget.provider),
3821
- args: agentArgs(agentTarget),
4609
+ command: invocation.command,
4610
+ args: invocation.args,
3822
4611
  cwd: agentTarget.cwd,
3823
4612
  timeoutMs: agentTarget.timeoutMs ?? null,
3824
- idleTimeoutMs: agentTarget.idleTimeoutMs,
4613
+ idleTimeoutMs: agentTarget.idleTimeoutMs ?? defaultAgentIdleTimeoutMs(agentTarget, opts),
3825
4614
  account: agentTarget.account,
3826
4615
  accountTool: agentTarget.account?.tool ?? accountToolForProvider(agentTarget.provider),
3827
4616
  nativeAuthProfile: agentTarget.authProfile ? { provider: agentTarget.provider, profile: agentTarget.authProfile } : undefined,
3828
- preflightAnyOf: agentTarget.provider === "cursor" ? ["agent"] : undefined,
3829
- stdin: agentTarget.provider === "codewith" ? undefined : agentTarget.prompt,
4617
+ preflightAnyOf: invocation.preflightAnyOf,
4618
+ stdin: invocation.stdin,
3830
4619
  allowlist: agentTarget.allowlist,
3831
- codewithDurableAgent: agentTarget.provider === "codewith" ? { target: agentTarget } : undefined
4620
+ worktree: agentTarget.worktree,
4621
+ codewithDurableAgent: adapter.capabilities.durable ? { target: agentTarget } : undefined
3832
4622
  };
3833
4623
  }
3834
- function executionEnv(spec, metadata, opts) {
4624
+ function composeExecutionEnv(spec, metadata, opts, accountEnv) {
3835
4625
  const env = { ...opts.env ?? process.env };
3836
- if (spec.account) {
3837
- const accountEnv = resolveAccountEnv(spec.account, spec.accountTool, env);
4626
+ if (accountEnv) {
3838
4627
  for (const key of AUTH_ENV_KEYS)
3839
4628
  delete env[key];
3840
4629
  Object.assign(env, accountEnv);
@@ -3845,6 +4634,14 @@ function executionEnv(spec, metadata, opts) {
3845
4634
  Object.assign(env, metadataEnv(metadata));
3846
4635
  return env;
3847
4636
  }
4637
+ async function executionEnv(spec, metadata, opts) {
4638
+ const accountEnv = spec.account ? await resolveAccountEnv(spec.account, spec.accountTool, { ...opts.env ?? process.env }) : undefined;
4639
+ return composeExecutionEnv(spec, metadata, opts, accountEnv);
4640
+ }
4641
+ function executionEnvSync(spec, metadata, opts) {
4642
+ const accountEnv = spec.account ? resolveAccountEnvSync(spec.account, spec.accountTool, { ...opts.env ?? process.env }) : undefined;
4643
+ return composeExecutionEnv(spec, metadata, opts, accountEnv);
4644
+ }
3848
4645
  function resolvedMachine(opts) {
3849
4646
  if (!opts.machine)
3850
4647
  return;
@@ -3862,13 +4659,21 @@ function hereDoc(value) {
3862
4659
  }
3863
4660
  return [`cat > "$__OPENLOOPS_STDIN" <<'${delimiter2}'`, value, delimiter2];
3864
4661
  }
3865
- function remoteBootstrapLines(spec, metadata) {
4662
+ function remoteBootstrapLines(spec, metadata, opts = {}) {
3866
4663
  const lines = [
3867
4664
  "set -e",
3868
4665
  'export PATH="$HOME/.local/bin:$HOME/.bun/bin:$HOME/.cargo/bin:$HOME/.npm-global/bin:$HOME/bin:/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:/usr/sbin:/sbin${PATH:+:$PATH}"'
3869
4666
  ];
3870
- if (spec.cwd)
4667
+ const worktree = spec.worktree;
4668
+ const worktreeManaged = worktree?.enabled && (worktree.mode === "auto" || worktree.mode === "required");
4669
+ if (opts.worktree && worktree && worktreeManaged) {
4670
+ lines.push(...remoteWorktreePrepareLines(worktree));
4671
+ lines.push(...remoteWorktreeEnterLines(worktree, spec.cwd));
4672
+ } else if (worktree && worktreeManaged) {
4673
+ lines.push(`cd ${shellQuote(worktree.originalCwd)}`);
4674
+ } else if (spec.cwd) {
3871
4675
  lines.push(`cd ${shellQuote(spec.cwd)}`);
4676
+ }
3872
4677
  if (spec.account) {
3873
4678
  if (!spec.accountTool)
3874
4679
  throw new Error("account.tool is required when no provider tool can be inferred");
@@ -3884,16 +4689,79 @@ function remoteBootstrapLines(spec, metadata) {
3884
4689
  }
3885
4690
  return lines;
3886
4691
  }
3887
- function remoteScript(spec, metadata) {
3888
- const lines = remoteBootstrapLines(spec, metadata);
4692
+ function remoteWorktreePrepareLines(worktree) {
4693
+ const { repoRoot, path, branch } = worktree;
4694
+ if (!repoRoot || !path || !branch) {
4695
+ return [
4696
+ "__openloops_prepare_worktree() {",
4697
+ ` echo ${shellQuote("worktree preparation requires repoRoot, path, and branch metadata")} >&2`,
4698
+ " return 1",
4699
+ "}"
4700
+ ];
4701
+ }
4702
+ return [
4703
+ "__openloops_prepare_worktree() {",
4704
+ ` local repo=${shellQuote(repoRoot)} path=${shellQuote(path)} branch=${shellQuote(branch)}`,
4705
+ " local top expected_common actual_common current",
4706
+ ' if [ -L "$path" ]; then echo "refusing symlinked worktree path $path" >&2; return 1; fi',
4707
+ ' if [ -e "$path" ]; then',
4708
+ ' top="$(git -C "$path" rev-parse --show-toplevel 2>/dev/null)" || { echo "refusing to reuse non-worktree path: $path" >&2; return 1; }',
4709
+ ' if [ "$(cd "$top" 2>/dev/null && pwd -P)" != "$(cd "$path" 2>/dev/null && pwd -P)" ]; then echo "existing worktree top-level mismatch for $path: $top" >&2; return 1; fi',
4710
+ ' expected_common="$(cd "$repo" 2>/dev/null && cd "$(git rev-parse --git-common-dir 2>/dev/null)" 2>/dev/null && pwd -P)"',
4711
+ ' actual_common="$(cd "$path" 2>/dev/null && cd "$(git rev-parse --git-common-dir 2>/dev/null)" 2>/dev/null && pwd -P)"',
4712
+ ' if [ -z "$expected_common" ] || [ "$expected_common" != "$actual_common" ]; then echo "existing worktree $path belongs to a different git common dir" >&2; return 1; fi',
4713
+ ' current="$(git -C "$path" branch --show-current 2>/dev/null)"',
4714
+ ' if [ "$current" != "$branch" ]; then echo "existing worktree $path is on branch ${current:-unknown}, expected $branch" >&2; return 1; fi',
4715
+ " return 0",
4716
+ " fi",
4717
+ ' git -C "$repo" rev-parse --is-inside-work-tree >/dev/null 2>&1 || { echo "worktree repoRoot is not a git repository: $repo" >&2; return 1; }',
4718
+ ' mkdir -p "$(dirname "$path")" || return 1',
4719
+ " # Preparation chatter goes to stderr so run stdout stays the agent's.",
4720
+ ' if git -C "$repo" show-ref --verify --quiet "refs/heads/$branch"; then',
4721
+ ' git -C "$repo" worktree add "$path" "$branch" 1>&2 || return 1',
4722
+ " else",
4723
+ ' git -C "$repo" worktree add -b "$branch" "$path" HEAD 1>&2 || return 1',
4724
+ " fi",
4725
+ "}"
4726
+ ];
4727
+ }
4728
+ function remoteWorktreeEnterLines(worktree, cwd) {
4729
+ const workdir = cwd ?? worktree.cwd;
4730
+ if (worktree.mode === "required") {
4731
+ return [
4732
+ "if ! __openloops_prepare_worktree; then",
4733
+ ` echo ${shellQuote("worktree preparation failed (mode=required)")} >&2`,
4734
+ " exit 1",
4735
+ "fi",
4736
+ "__OPENLOOPS_WORKTREE_OK=1",
4737
+ `cd ${shellQuote(workdir)}`
4738
+ ];
4739
+ }
4740
+ return [
4741
+ "if __openloops_prepare_worktree; then",
4742
+ " __OPENLOOPS_WORKTREE_OK=1",
4743
+ ` cd ${shellQuote(workdir)}`,
4744
+ "else",
4745
+ ` echo ${shellQuote(`worktree preparation failed (mode=${worktree.mode}); falling back to ${worktree.originalCwd}`)} >&2`,
4746
+ " __OPENLOOPS_WORKTREE_OK=0",
4747
+ ` cd ${shellQuote(worktree.originalCwd)}`,
4748
+ "fi"
4749
+ ];
4750
+ }
4751
+ function remoteScript(spec, metadata, fallbackSpec) {
4752
+ const lines = remoteBootstrapLines(spec, metadata, { worktree: true });
3889
4753
  let stdinRedirect = "";
3890
4754
  if (spec.stdin !== undefined) {
3891
4755
  lines.push('__OPENLOOPS_STDIN="$(mktemp -t openloops-stdin.XXXXXX)"', `trap 'rm -f "$__OPENLOOPS_STDIN"' EXIT`);
3892
4756
  lines.push(...hereDoc(spec.stdin));
3893
4757
  stdinRedirect = ' < "$__OPENLOOPS_STDIN"';
3894
4758
  }
3895
- const invocation = spec.shell ? `sh -c ${shellQuote(commandForShell(spec))}${stdinRedirect}` : `${[spec.command, ...spec.args].map(shellQuote).join(" ")}${stdinRedirect}`;
3896
- lines.push(invocation);
4759
+ const invocationFor = (invocationSpec) => invocationSpec.shell ? `sh -c ${shellQuote(commandForShell(invocationSpec))}${stdinRedirect}` : `${[invocationSpec.command, ...invocationSpec.args].map(shellQuote).join(" ")}${stdinRedirect}`;
4760
+ if (spec.worktree?.enabled && spec.worktree.mode === "auto" && fallbackSpec) {
4761
+ lines.push('if [ "${__OPENLOOPS_WORKTREE_OK:-0}" = 1 ]; then', ` ${invocationFor(spec)}`, "else", ` ${invocationFor(fallbackSpec)}`, "fi");
4762
+ } else {
4763
+ lines.push(invocationFor(spec));
4764
+ }
3897
4765
  return `${lines.join(`
3898
4766
  `)}
3899
4767
  `;
@@ -3925,31 +4793,43 @@ function transportEnv(opts) {
3925
4793
  env.PATH = normalizeExecutionPath(env);
3926
4794
  return env;
3927
4795
  }
3928
- function preflightNativeAuthProfile(spec, env) {
3929
- if (!spec.nativeAuthProfile)
3930
- return;
3931
- if (spec.nativeAuthProfile.provider !== "codewith")
3932
- return;
3933
- const result = spawnSync2(spec.command, ["profile", "list"], {
3934
- encoding: "utf8",
3935
- env,
3936
- stdio: ["ignore", "pipe", "pipe"],
3937
- timeout: 15000
3938
- });
4796
+ function assertCodewithProfileListed(profile, result) {
3939
4797
  if (result.error) {
3940
- throw new Error(`codewith auth profile preflight failed: ${result.error.message}`);
4798
+ throw new Error(`codewith auth profile preflight failed: ${result.error}`);
3941
4799
  }
3942
4800
  if ((result.status ?? 1) !== 0) {
3943
4801
  const detail = (result.stderr || result.stdout || `exit ${result.status ?? "unknown"}`).trim();
3944
4802
  throw new Error(`codewith auth profile preflight failed${detail ? `: ${detail}` : ""}`);
3945
4803
  }
3946
4804
  const profiles = new Set((result.stdout || "").split(/\r?\n/).slice(1).map((line) => line.trim().split(/\s+/)[0]).filter(Boolean));
3947
- if (!profiles.has(spec.nativeAuthProfile.profile)) {
3948
- throw new Error(`codewith auth profile not found: ${spec.nativeAuthProfile.profile}`);
4805
+ if (!profiles.has(profile)) {
4806
+ throw new Error(`codewith auth profile not found: ${profile}`);
3949
4807
  }
3950
4808
  }
4809
+ function preflightNativeAuthProfileSync(spec, env) {
4810
+ if (spec.nativeAuthProfile?.provider !== "codewith")
4811
+ return;
4812
+ const result = spawnSync4(spec.command, ["profile", "list"], {
4813
+ encoding: "utf8",
4814
+ env,
4815
+ stdio: ["ignore", "pipe", "pipe"],
4816
+ timeout: 15000
4817
+ });
4818
+ assertCodewithProfileListed(spec.nativeAuthProfile.profile, {
4819
+ status: result.status,
4820
+ stdout: result.stdout ?? "",
4821
+ stderr: result.stderr ?? "",
4822
+ error: result.error?.message
4823
+ });
4824
+ }
4825
+ async function preflightNativeAuthProfile(spec, env) {
4826
+ if (spec.nativeAuthProfile?.provider !== "codewith")
4827
+ return;
4828
+ const result = await spawnCapture(spec.command, ["profile", "list"], { env, timeoutMs: 15000 });
4829
+ assertCodewithProfileListed(spec.nativeAuthProfile.profile, result);
4830
+ }
3951
4831
  function codewithAgentStartArgs(target, idempotencyKey) {
3952
- const args = agentArgs(target);
4832
+ const args = providerAdapter(target.provider).buildInvocation(target).args;
3953
4833
  const startIndex = args.findIndex((arg, index) => arg === "start" && args[index - 1] === "agent");
3954
4834
  if (startIndex === -1)
3955
4835
  throw new Error("internal error: codewith durable agent args missing agent start");
@@ -4023,7 +4903,7 @@ function codewithAgentEvidence(startJson, readJson, logsJson) {
4023
4903
  }, null, 2);
4024
4904
  }
4025
4905
  function sleep(ms) {
4026
- return new Promise((resolve2) => setTimeout(resolve2, ms));
4906
+ return new Promise((resolve3) => setTimeout(resolve3, ms));
4027
4907
  }
4028
4908
  async function executeCodewithDurableAgent(spec, metadata, opts, env, startedAt) {
4029
4909
  const target = spec.codewithDurableAgent?.target;
@@ -4031,116 +4911,84 @@ async function executeCodewithDurableAgent(spec, metadata, opts, env, startedAt)
4031
4911
  throw new Error("internal error: missing codewith durable target");
4032
4912
  const maxOutputBytes = opts.maxOutputBytes ?? DEFAULT_MAX_OUTPUT_BYTES;
4033
4913
  const idempotencyKey = codewithAgentIdempotencyKey(metadata);
4034
- const start = spawnSync2(spec.command, codewithAgentStartArgs(target, idempotencyKey), {
4914
+ const start = await spawnCapture(spec.command, codewithAgentStartArgs(target, idempotencyKey), {
4035
4915
  cwd: spec.cwd,
4036
4916
  env,
4037
- encoding: "utf8",
4038
- stdio: ["ignore", "pipe", "pipe"],
4039
- timeout: 30000
4917
+ timeoutMs: 30000,
4918
+ maxOutputBytes
4040
4919
  });
4041
- let stderr = start.stderr || "";
4920
+ const stderr = new BoundedOutputBuffer(maxOutputBytes);
4921
+ stderr.append(start.stderr);
4042
4922
  if (start.error || (start.status ?? 1) !== 0) {
4043
- const finishedAt = nowIso();
4044
- return {
4045
- status: "failed",
4923
+ return failureResult(startedAt, start.error ?? `codewith agent start exited with code ${start.status ?? "unknown"}`, {
4046
4924
  exitCode: start.status ?? undefined,
4047
- stdout: start.stdout || "",
4048
- stderr,
4049
- error: start.error?.message ?? `codewith agent start exited with code ${start.status ?? "unknown"}`,
4050
- startedAt,
4051
- finishedAt,
4052
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4053
- };
4925
+ stdout: start.stdout,
4926
+ stderr: stderr.value()
4927
+ });
4054
4928
  }
4055
4929
  let startJson;
4056
4930
  try {
4057
4931
  startJson = parseJsonOutput(start.stdout || "{}", "codewith agent start");
4058
4932
  } catch (error) {
4059
- const finishedAt = nowIso();
4060
- return {
4061
- status: "failed",
4062
- stdout: start.stdout || "",
4063
- stderr,
4064
- error: error instanceof Error ? error.message : String(error),
4065
- startedAt,
4066
- finishedAt,
4067
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4068
- };
4933
+ return failureResult(startedAt, error instanceof Error ? error.message : String(error), { stdout: start.stdout, stderr: stderr.value() });
4069
4934
  }
4070
4935
  const agentId = stringField(recordField(startJson, "agent"), "agentId");
4071
4936
  if (!agentId) {
4072
- const finishedAt = nowIso();
4073
- return {
4074
- status: "failed",
4075
- stdout: start.stdout || "",
4076
- stderr,
4077
- error: "codewith agent start did not return agent.agentId",
4078
- startedAt,
4079
- finishedAt,
4080
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4081
- };
4937
+ return failureResult(startedAt, "codewith agent start did not return agent.agentId", { stdout: start.stdout, stderr: stderr.value() });
4082
4938
  }
4939
+ const stopAgent = async () => {
4940
+ await spawnCapture(spec.command, codewithAgentControlArgs(target, "stop", agentId), {
4941
+ cwd: spec.cwd,
4942
+ env,
4943
+ timeoutMs: 15000,
4944
+ maxOutputBytes
4945
+ });
4946
+ };
4083
4947
  const pollMs = Math.max(100, Number(opts.env?.LOOPS_CODEWITH_AGENT_POLL_MS ?? process.env.LOOPS_CODEWITH_AGENT_POLL_MS ?? 2000) || 2000);
4084
4948
  let lastReadJson = startJson;
4085
4949
  let lastLogsJson;
4086
- let stdout = "";
4950
+ let lastFingerprint;
4951
+ let lastProgressAt = Date.now();
4952
+ const evidence = () => boundedText(codewithAgentEvidence(startJson, lastReadJson, lastLogsJson), maxOutputBytes);
4087
4953
  while (true) {
4088
4954
  if (opts.signal?.aborted) {
4089
- spawnSync2(spec.command, codewithAgentControlArgs(target, "stop", agentId), { cwd: spec.cwd, env, stdio: "ignore", timeout: 15000 });
4090
- const finishedAt = nowIso();
4091
- return {
4092
- status: "failed",
4093
- stdout: appendBoundedText(stdout, codewithAgentEvidence(startJson, lastReadJson, lastLogsJson), maxOutputBytes),
4094
- stderr,
4095
- error: "cancelled",
4096
- startedAt,
4097
- finishedAt,
4098
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4099
- };
4955
+ await stopAgent();
4956
+ return failureResult(startedAt, "cancelled", { stdout: evidence(), stderr: stderr.value() });
4100
4957
  }
4101
- const read = spawnSync2(spec.command, codewithAgentControlArgs(target, "read", agentId), {
4958
+ const read = await spawnCapture(spec.command, codewithAgentControlArgs(target, "read", agentId), {
4102
4959
  cwd: spec.cwd,
4103
4960
  env,
4104
- encoding: "utf8",
4105
- stdio: ["ignore", "pipe", "pipe"],
4106
- timeout: 30000
4961
+ timeoutMs: 30000,
4962
+ maxOutputBytes
4107
4963
  });
4108
- stderr = appendBoundedText(stderr, read.stderr || "", maxOutputBytes);
4964
+ stderr.append(read.stderr);
4109
4965
  if (read.error || (read.status ?? 1) !== 0) {
4110
- const finishedAt = nowIso();
4111
- return {
4112
- status: "failed",
4966
+ return failureResult(startedAt, read.error ?? `codewith agent read exited with code ${read.status ?? "unknown"}`, {
4113
4967
  exitCode: read.status ?? undefined,
4114
- stdout: appendBoundedText(stdout, codewithAgentEvidence(startJson, lastReadJson, lastLogsJson), maxOutputBytes),
4115
- stderr,
4116
- error: read.error?.message ?? `codewith agent read exited with code ${read.status ?? "unknown"}`,
4117
- startedAt,
4118
- finishedAt,
4119
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4120
- };
4968
+ stdout: evidence(),
4969
+ stderr: stderr.value()
4970
+ });
4121
4971
  }
4122
4972
  try {
4123
4973
  lastReadJson = parseJsonOutput(read.stdout || "{}", "codewith agent read");
4124
4974
  } catch (error) {
4125
- const finishedAt = nowIso();
4126
- return {
4127
- status: "failed",
4128
- stdout: appendBoundedText(stdout, read.stdout || "", maxOutputBytes),
4129
- stderr,
4130
- error: error instanceof Error ? error.message : String(error),
4131
- startedAt,
4132
- finishedAt,
4133
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4134
- };
4975
+ return failureResult(startedAt, error instanceof Error ? error.message : String(error), {
4976
+ stdout: boundedText(read.stdout, maxOutputBytes),
4977
+ stderr: stderr.value()
4978
+ });
4135
4979
  }
4136
4980
  const status = codewithAgentStatus(lastReadJson);
4981
+ const fingerprint = JSON.stringify({ status, snapshot: recordField(lastReadJson, "statusSnapshot") ?? null });
4982
+ if (fingerprint !== lastFingerprint) {
4983
+ lastFingerprint = fingerprint;
4984
+ lastProgressAt = Date.now();
4985
+ }
4137
4986
  if (status === "completed" || status === "failed" || status === "cancelled") {
4138
- const logs = spawnSync2(spec.command, codewithAgentControlArgs(target, "logs", agentId), {
4987
+ const logs = await spawnCapture(spec.command, codewithAgentControlArgs(target, "logs", agentId), {
4139
4988
  cwd: spec.cwd,
4140
4989
  env,
4141
- encoding: "utf8",
4142
- stdio: ["ignore", "pipe", "pipe"],
4143
- timeout: 30000
4990
+ timeoutMs: 30000,
4991
+ maxOutputBytes
4144
4992
  });
4145
4993
  if (!logs.error && (logs.status ?? 1) === 0) {
4146
4994
  try {
@@ -4149,40 +4997,129 @@ async function executeCodewithDurableAgent(spec, metadata, opts, env, startedAt)
4149
4997
  lastLogsJson = undefined;
4150
4998
  }
4151
4999
  } else {
4152
- stderr = appendBoundedText(stderr, logs.stderr || logs.error?.message || "", maxOutputBytes);
5000
+ stderr.append(logs.stderr || logs.error || "");
4153
5001
  }
4154
- const finishedAt = nowIso();
4155
- const resultStatus = status === "completed" ? "succeeded" : "failed";
4156
- return {
4157
- status: resultStatus,
4158
- exitCode: resultStatus === "succeeded" ? 0 : 1,
4159
- stdout: appendBoundedText(stdout, codewithAgentEvidence(startJson, lastReadJson, lastLogsJson), maxOutputBytes),
4160
- stderr,
4161
- error: resultStatus === "succeeded" ? undefined : stringField(recordField(lastReadJson, "agent"), "statusReason") ?? `codewith agent ${status}`,
4162
- startedAt,
4163
- finishedAt,
4164
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4165
- };
5002
+ if (status === "completed") {
5003
+ return successResult(startedAt, { exitCode: 0, stdout: evidence(), stderr: stderr.value() });
5004
+ }
5005
+ return failureResult(startedAt, stringField(recordField(lastReadJson, "agent"), "statusReason") ?? `codewith agent ${status}`, { exitCode: 1, stdout: evidence(), stderr: stderr.value() });
4166
5006
  }
4167
5007
  if (typeof spec.timeoutMs === "number" && new Date(nowIso()).getTime() - new Date(startedAt).getTime() >= spec.timeoutMs) {
4168
- spawnSync2(spec.command, codewithAgentControlArgs(target, "stop", agentId), { cwd: spec.cwd, env, stdio: "ignore", timeout: 15000 });
4169
- const finishedAt = nowIso();
4170
- return {
4171
- status: "timed_out",
4172
- stdout: appendBoundedText(stdout, codewithAgentEvidence(startJson, lastReadJson, lastLogsJson), maxOutputBytes),
4173
- stderr,
4174
- error: `timed out after ${spec.timeoutMs}ms`,
4175
- startedAt,
4176
- finishedAt,
4177
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
4178
- };
5008
+ await stopAgent();
5009
+ return timeoutResult(startedAt, `timed out after ${spec.timeoutMs}ms`, { stdout: evidence(), stderr: stderr.value() });
5010
+ }
5011
+ if (typeof spec.idleTimeoutMs === "number" && Date.now() - lastProgressAt >= spec.idleTimeoutMs) {
5012
+ await stopAgent();
5013
+ return timeoutResult(startedAt, `idle timed out after ${spec.idleTimeoutMs}ms without agent progress`, {
5014
+ stdout: evidence(),
5015
+ stderr: stderr.value()
5016
+ });
4179
5017
  }
4180
5018
  await sleep(pollMs);
4181
5019
  }
4182
5020
  }
5021
+ function resolvedDirEquals(left, right) {
5022
+ try {
5023
+ return realpathSync(left) === realpathSync(right);
5024
+ } catch {
5025
+ return false;
5026
+ }
5027
+ }
5028
+ async function ensureLocalWorktree(worktree, env) {
5029
+ const { repoRoot, path, branch } = worktree;
5030
+ if (!repoRoot || !path || !branch) {
5031
+ return { error: "worktree preparation requires repoRoot, path, and branch metadata" };
5032
+ }
5033
+ const git = (args) => spawnCapture("git", args, { env, timeoutMs: WORKTREE_GIT_TIMEOUT_MS });
5034
+ let stats;
5035
+ try {
5036
+ stats = lstatSync(path);
5037
+ } catch {
5038
+ stats = undefined;
5039
+ }
5040
+ if (stats?.isSymbolicLink())
5041
+ return { error: `refusing symlinked worktree path ${path}` };
5042
+ const commonDir = async (base) => {
5043
+ const result = await git(["-C", base, "rev-parse", "--git-common-dir"]);
5044
+ if (result.error || (result.status ?? 1) !== 0)
5045
+ return;
5046
+ const raw = result.stdout.trim();
5047
+ if (!raw)
5048
+ return;
5049
+ try {
5050
+ return realpathSync(resolve2(base, raw));
5051
+ } catch {
5052
+ return resolve2(base, raw);
5053
+ }
5054
+ };
5055
+ if (stats) {
5056
+ const top = await git(["-C", path, "rev-parse", "--show-toplevel"]);
5057
+ if (top.error || (top.status ?? 1) !== 0) {
5058
+ return { error: `refusing to reuse non-worktree path: ${path}` };
5059
+ }
5060
+ if (!resolvedDirEquals(top.stdout.trim(), path)) {
5061
+ return { error: `existing worktree top-level mismatch for ${path}: ${top.stdout.trim()}` };
5062
+ }
5063
+ const expectedCommon = await commonDir(repoRoot);
5064
+ const actualCommon = await commonDir(path);
5065
+ if (!expectedCommon || expectedCommon !== actualCommon) {
5066
+ return { error: `existing worktree ${path} belongs to a different git common dir` };
5067
+ }
5068
+ const current = await git(["-C", path, "branch", "--show-current"]);
5069
+ const actualBranch = current.stdout.trim();
5070
+ if (current.error || (current.status ?? 1) !== 0 || actualBranch !== branch) {
5071
+ return { error: `existing worktree ${path} is on branch ${actualBranch || "unknown"}, expected ${branch}` };
5072
+ }
5073
+ return { cwd: worktree.cwd };
5074
+ }
5075
+ const inside = await git(["-C", repoRoot, "rev-parse", "--is-inside-work-tree"]);
5076
+ if (inside.error || (inside.status ?? 1) !== 0) {
5077
+ return { error: `worktree repoRoot is not a git repository: ${repoRoot}` };
5078
+ }
5079
+ try {
5080
+ mkdirSync5(dirname4(path), { recursive: true });
5081
+ } catch (error) {
5082
+ return { error: `could not create worktree parent directory: ${error instanceof Error ? error.message : String(error)}` };
5083
+ }
5084
+ const hasBranch = await git(["-C", repoRoot, "show-ref", "--verify", "--quiet", `refs/heads/${branch}`]);
5085
+ const add = hasBranch.status === 0 ? await git(["-C", repoRoot, "worktree", "add", path, branch]) : await git(["-C", repoRoot, "worktree", "add", "-b", branch, path, "HEAD"]);
5086
+ if (add.error || (add.status ?? 1) !== 0) {
5087
+ const detail = (add.stderr || add.stdout || add.error || "").toString().trim();
5088
+ return { error: `git worktree add failed for ${path}${detail ? `: ${detail}` : ""}` };
5089
+ }
5090
+ return { cwd: worktree.cwd };
5091
+ }
5092
+ async function enterWorktree(spec, opts, env, startedAt) {
5093
+ const worktree = spec.worktree;
5094
+ if (!worktree?.enabled || worktree.mode === "off" || worktree.mode === "main")
5095
+ return;
5096
+ const prepared = await ensureLocalWorktree(worktree, env);
5097
+ if (prepared.error) {
5098
+ if (worktree.mode === "required") {
5099
+ return { failure: failureResult(startedAt, `worktree preparation failed (mode=required): ${prepared.error}`) };
5100
+ }
5101
+ opts.log?.(`worktree preparation failed (mode=${worktree.mode}); falling back to ${worktree.originalCwd}: ${prepared.error}`);
5102
+ spec.cwd = worktree.originalCwd;
5103
+ return { fallbackCwd: worktree.originalCwd };
5104
+ }
5105
+ spec.cwd = prepared.cwd ?? spec.cwd;
5106
+ opts.log?.(`entered worktree ${worktree.path ?? spec.cwd}${worktree.branch ? ` branch ${worktree.branch}` : ""}`);
5107
+ return;
5108
+ }
5109
+ function worktreeFallbackSpec(target, opts, fallbackCwd) {
5110
+ if (target.type !== "agent")
5111
+ return;
5112
+ const agentTarget = target;
5113
+ const fallbackTarget = {
5114
+ ...agentTarget,
5115
+ cwd: fallbackCwd,
5116
+ worktree: agentTarget.worktree ? { ...agentTarget.worktree, enabled: false, cwd: fallbackCwd, reason: "auto worktree preparation failed" } : undefined
5117
+ };
5118
+ return commandSpec(fallbackTarget, opts);
5119
+ }
4183
5120
  function preflightRemoteSpec(spec, machine, metadata, opts) {
4184
5121
  const plan = (opts.machineCommandResolver ?? resolveMachineCommand)(machine.id, "bash -s");
4185
- const result = spawnSync2(plan.command, plan.args, {
5122
+ const result = spawnSync4(plan.command, plan.args, {
4186
5123
  encoding: "utf8",
4187
5124
  env: transportEnv(opts),
4188
5125
  input: remotePreflightScript(spec, metadata),
@@ -4196,11 +5133,11 @@ function preflightRemoteSpec(spec, machine, metadata, opts) {
4196
5133
  throw new Error(`remote preflight failed on ${machine.id}${detail ? `: ${detail}` : ""}`);
4197
5134
  }
4198
5135
  }
4199
- async function executeRemoteSpec(spec, machine, metadata, opts) {
5136
+ async function executeRemoteSpec(spec, machine, metadata, opts, fallbackSpec) {
4200
5137
  const maxOutputBytes = opts.maxOutputBytes ?? DEFAULT_MAX_OUTPUT_BYTES;
4201
5138
  const startedAt = nowIso();
4202
- let stdout = "";
4203
- let stderr = "";
5139
+ const stdout = new BoundedOutputBuffer(maxOutputBytes);
5140
+ const stderr = new BoundedOutputBuffer(maxOutputBytes);
4204
5141
  let timedOut = false;
4205
5142
  let idleTimedOut = false;
4206
5143
  let exitCode;
@@ -4209,25 +5146,16 @@ async function executeRemoteSpec(spec, machine, metadata, opts) {
4209
5146
  let script;
4210
5147
  try {
4211
5148
  plan = (opts.machineCommandResolver ?? resolveMachineCommand)(machine.id, "bash -s");
4212
- script = remoteScript(spec, metadata);
5149
+ script = remoteScript(spec, metadata, fallbackSpec);
4213
5150
  } catch (err) {
4214
- return {
4215
- status: "failed",
4216
- stdout: "",
4217
- stderr: "",
4218
- error: err instanceof Error ? err.message : String(err),
4219
- startedAt,
4220
- finishedAt: nowIso(),
4221
- durationMs: 0
4222
- };
5151
+ return failureResult(startedAt, err instanceof Error ? err.message : String(err));
4223
5152
  }
4224
- const child = spawn(plan.command, plan.args, {
5153
+ const child = spawn2(plan.command, plan.args, {
4225
5154
  env: transportEnv(opts),
4226
5155
  detached: true,
4227
5156
  stdio: ["pipe", "pipe", "pipe"]
4228
5157
  });
4229
- if (child.pid)
4230
- opts.onSpawn?.(child.pid);
5158
+ notifySpawn(child.pid, opts);
4231
5159
  child.stdin?.on("error", (err) => {
4232
5160
  if (err.code !== "EPIPE")
4233
5161
  error = err.message;
@@ -4261,12 +5189,14 @@ async function executeRemoteSpec(spec, machine, metadata, opts) {
4261
5189
  idleTimer.unref();
4262
5190
  };
4263
5191
  resetIdleTimer();
5192
+ child.stdout?.setEncoding("utf8");
5193
+ child.stderr?.setEncoding("utf8");
4264
5194
  child.stdout?.on("data", (chunk) => {
4265
- stdout = appendBounded(stdout, chunk, maxOutputBytes);
5195
+ stdout.append(chunk);
4266
5196
  resetIdleTimer();
4267
5197
  });
4268
5198
  child.stderr?.on("data", (chunk) => {
4269
- stderr = appendBounded(stderr, chunk, maxOutputBytes);
5199
+ stderr.append(chunk);
4270
5200
  resetIdleTimer();
4271
5201
  });
4272
5202
  try {
@@ -4284,47 +5214,17 @@ async function executeRemoteSpec(spec, machine, metadata, opts) {
4284
5214
  clearTimeout(idleTimer);
4285
5215
  opts.signal?.removeEventListener("abort", abortHandler);
4286
5216
  }
4287
- const finishedAt = nowIso();
4288
- const durationMs = new Date(finishedAt).getTime() - new Date(startedAt).getTime();
5217
+ const fields = { exitCode, stdout: stdout.value(), stderr: stderr.value(), pid: child.pid };
4289
5218
  if (timedOut || idleTimedOut) {
4290
- return {
4291
- status: "timed_out",
4292
- exitCode,
4293
- stdout,
4294
- stderr,
4295
- error: idleTimedOut ? `idle timed out after ${spec.idleTimeoutMs}ms without stdout/stderr` : `timed out after ${spec.timeoutMs}ms`,
4296
- pid: child.pid,
4297
- startedAt,
4298
- finishedAt,
4299
- durationMs
4300
- };
5219
+ return timeoutResult(startedAt, idleTimedOut ? `idle timed out after ${spec.idleTimeoutMs}ms without stdout/stderr` : `timed out after ${spec.timeoutMs}ms`, fields);
4301
5220
  }
4302
5221
  if (error || exitCode !== 0) {
4303
- return {
4304
- status: "failed",
4305
- exitCode,
4306
- stdout,
4307
- stderr,
4308
- error: error ?? `remote process on ${machine.id} exited with code ${exitCode ?? "unknown"}`,
4309
- pid: child.pid,
4310
- startedAt,
4311
- finishedAt,
4312
- durationMs
4313
- };
5222
+ return failureResult(startedAt, error ?? `remote process on ${machine.id} exited with code ${exitCode ?? "unknown"}`, fields);
4314
5223
  }
4315
- return {
4316
- status: "succeeded",
4317
- exitCode,
4318
- stdout,
4319
- stderr,
4320
- pid: child.pid,
4321
- startedAt,
4322
- finishedAt,
4323
- durationMs
4324
- };
5224
+ return successResult(startedAt, fields);
4325
5225
  }
4326
5226
  function preflightTarget(target, metadata = {}, opts = {}) {
4327
- const spec = commandSpec(target);
5227
+ const spec = commandSpec(target, opts);
4328
5228
  const machine = resolvedMachine(opts);
4329
5229
  if (machine && !machine.local) {
4330
5230
  preflightRemoteSpec(spec, machine, metadata, opts);
@@ -4334,14 +5234,14 @@ function preflightTarget(target, metadata = {}, opts = {}) {
4334
5234
  accountTool: spec.accountTool
4335
5235
  };
4336
5236
  }
4337
- const env = executionEnv(spec, metadata, opts);
5237
+ const env = executionEnvSync(spec, metadata, opts);
4338
5238
  if (!spec.shell && !executableExists(spec.command, env)) {
4339
5239
  throw new Error(commandNotFoundMessage(spec.command, env));
4340
5240
  }
4341
5241
  if (spec.preflightAnyOf?.length && !spec.preflightAnyOf.some((command) => executableExists(command, env))) {
4342
5242
  throw new Error(`none of required executables found: ${spec.preflightAnyOf.join(", ")}`);
4343
5243
  }
4344
- preflightNativeAuthProfile(spec, env);
5244
+ preflightNativeAuthProfileSync(spec, env);
4345
5245
  return {
4346
5246
  command: spec.command,
4347
5247
  accountProfile: spec.account?.profile,
@@ -4349,79 +5249,52 @@ function preflightTarget(target, metadata = {}, opts = {}) {
4349
5249
  };
4350
5250
  }
4351
5251
  async function executeTarget(target, metadata = {}, opts = {}) {
4352
- const spec = commandSpec(target);
5252
+ let spec = commandSpec(target, opts);
4353
5253
  const machine = resolvedMachine(opts);
4354
5254
  if (machine && !machine.local && spec.codewithDurableAgent) {
4355
- const startedAt2 = nowIso();
4356
- const finishedAt2 = nowIso();
4357
- return {
4358
- status: "failed",
4359
- stdout: "",
4360
- stderr: "",
4361
- error: "remote Codewith durable background-agent steps require remote status polling support; run this Codewith step locally or add remote durable readback before dispatch",
4362
- startedAt: startedAt2,
4363
- finishedAt: finishedAt2,
4364
- durationMs: new Date(finishedAt2).getTime() - new Date(startedAt2).getTime()
4365
- };
5255
+ return failureResult(nowIso(), "remote Codewith durable background-agent steps require remote status polling support; run this Codewith step locally or add remote durable readback before dispatch");
5256
+ }
5257
+ if (machine && !machine.local) {
5258
+ const remoteFallbackSpec = spec.worktree?.enabled && spec.worktree.mode === "auto" ? worktreeFallbackSpec(target, opts, spec.worktree.originalCwd) : undefined;
5259
+ return executeRemoteSpec(spec, machine, metadata, opts, remoteFallbackSpec);
4366
5260
  }
4367
- if (machine && !machine.local)
4368
- return executeRemoteSpec(spec, machine, metadata, opts);
4369
5261
  const maxOutputBytes = opts.maxOutputBytes ?? DEFAULT_MAX_OUTPUT_BYTES;
4370
5262
  const startedAt = nowIso();
4371
- let stdout = "";
4372
- let stderr = "";
5263
+ const stdout = new BoundedOutputBuffer(maxOutputBytes);
5264
+ const stderr = new BoundedOutputBuffer(maxOutputBytes);
4373
5265
  let timedOut = false;
4374
5266
  let idleTimedOut = false;
4375
5267
  let exitCode;
4376
5268
  let error;
4377
- const env = executionEnv(spec, metadata, opts);
5269
+ const env = await executionEnv(spec, metadata, opts);
4378
5270
  if (!spec.shell && !executableExists(spec.command, env)) {
4379
- return {
4380
- status: "failed",
4381
- stdout: "",
4382
- stderr: "",
4383
- error: commandNotFoundMessage(spec.command, env),
4384
- startedAt,
4385
- finishedAt: nowIso(),
4386
- durationMs: 0
4387
- };
5271
+ return failureResult(startedAt, commandNotFoundMessage(spec.command, env));
4388
5272
  }
4389
5273
  if (spec.preflightAnyOf?.length && !spec.preflightAnyOf.some((command) => executableExists(command, env))) {
4390
- return {
4391
- status: "failed",
4392
- stdout,
4393
- stderr,
4394
- error: `none of required executables found: ${spec.preflightAnyOf.join(", ")}`,
4395
- startedAt,
4396
- finishedAt: nowIso(),
4397
- durationMs: 0
4398
- };
5274
+ return failureResult(startedAt, `none of required executables found: ${spec.preflightAnyOf.join(", ")}`);
4399
5275
  }
4400
5276
  try {
4401
- preflightNativeAuthProfile(spec, env);
5277
+ await preflightNativeAuthProfile(spec, env);
4402
5278
  } catch (err) {
4403
- return {
4404
- status: "failed",
4405
- stdout: "",
4406
- stderr: "",
4407
- error: err instanceof Error ? err.message : String(err),
4408
- startedAt,
4409
- finishedAt: nowIso(),
4410
- durationMs: 0
4411
- };
5279
+ return failureResult(startedAt, err instanceof Error ? err.message : String(err));
5280
+ }
5281
+ const worktreeEntry = await enterWorktree(spec, opts, env, startedAt);
5282
+ if (worktreeEntry?.failure)
5283
+ return worktreeEntry.failure;
5284
+ if (worktreeEntry?.fallbackCwd) {
5285
+ spec = worktreeFallbackSpec(target, opts, worktreeEntry.fallbackCwd) ?? spec;
4412
5286
  }
4413
5287
  if (spec.codewithDurableAgent) {
4414
5288
  return executeCodewithDurableAgent(spec, metadata, opts, env, startedAt);
4415
5289
  }
4416
- const child = spawn(spec.command, spec.args, {
5290
+ const child = spawn2(spec.command, spec.args, {
4417
5291
  cwd: spec.cwd,
4418
5292
  env,
4419
5293
  shell: spec.shell ?? false,
4420
5294
  detached: true,
4421
5295
  stdio: spec.stdin === undefined ? ["ignore", "pipe", "pipe"] : ["pipe", "pipe", "pipe"]
4422
5296
  });
4423
- if (child.pid)
4424
- opts.onSpawn?.(child.pid);
5297
+ notifySpawn(child.pid, opts);
4425
5298
  if (spec.stdin !== undefined && child.stdin) {
4426
5299
  child.stdin.on("error", (err) => {
4427
5300
  if (err.code !== "EPIPE")
@@ -4457,12 +5330,14 @@ async function executeTarget(target, metadata = {}, opts = {}) {
4457
5330
  idleTimer.unref();
4458
5331
  };
4459
5332
  resetIdleTimer();
5333
+ child.stdout?.setEncoding("utf8");
5334
+ child.stderr?.setEncoding("utf8");
4460
5335
  child.stdout?.on("data", (chunk) => {
4461
- stdout = appendBounded(stdout, chunk, maxOutputBytes);
5336
+ stdout.append(chunk);
4462
5337
  resetIdleTimer();
4463
5338
  });
4464
5339
  child.stderr?.on("data", (chunk) => {
4465
- stderr = appendBounded(stderr, chunk, maxOutputBytes);
5340
+ stderr.append(chunk);
4466
5341
  resetIdleTimer();
4467
5342
  });
4468
5343
  try {
@@ -4480,97 +5355,706 @@ async function executeTarget(target, metadata = {}, opts = {}) {
4480
5355
  clearTimeout(idleTimer);
4481
5356
  opts.signal?.removeEventListener("abort", abortHandler);
4482
5357
  }
4483
- const finishedAt = nowIso();
4484
- const durationMs = new Date(finishedAt).getTime() - new Date(startedAt).getTime();
5358
+ const fields = { exitCode, stdout: stdout.value(), stderr: stderr.value(), pid: child.pid };
4485
5359
  if (timedOut || idleTimedOut) {
5360
+ return timeoutResult(startedAt, idleTimedOut ? `idle timed out after ${spec.idleTimeoutMs}ms without stdout/stderr` : `timed out after ${spec.timeoutMs}ms`, fields);
5361
+ }
5362
+ if (error || exitCode !== 0) {
5363
+ return failureResult(startedAt, error ?? `process exited with code ${exitCode ?? "unknown"}`, fields);
5364
+ }
5365
+ return successResult(startedAt, fields);
5366
+ }
5367
+ async function executeLoop(loop, run, opts = {}) {
5368
+ if (loop.target.type === "workflow") {
5369
+ throw new Error("workflow loop targets must be executed with executeLoopTarget");
5370
+ }
5371
+ if (loop.target.preflight?.beforeRun) {
5372
+ const startedAt = nowIso();
5373
+ try {
5374
+ preflightTarget(loop.target, {
5375
+ loopId: loop.id,
5376
+ loopName: loop.name,
5377
+ runId: run.id,
5378
+ scheduledFor: run.scheduledFor
5379
+ }, { ...opts, machine: opts.machine ?? loop.machine });
5380
+ } catch (error) {
5381
+ return failureResult(startedAt, `runtime preflight failed: ${error instanceof Error ? error.message : String(error)}`);
5382
+ }
5383
+ }
5384
+ return executeTarget(loop.target, {
5385
+ loopId: loop.id,
5386
+ loopName: loop.name,
5387
+ runId: run.id,
5388
+ scheduledFor: run.scheduledFor
5389
+ }, { ...opts, machine: opts.machine ?? loop.machine });
5390
+ }
5391
+
5392
+ // src/lib/doctor.ts
5393
+ var PROVIDER_COMMANDS = [
5394
+ "claude",
5395
+ "agent",
5396
+ "codewith",
5397
+ "aicopilot",
5398
+ "opencode",
5399
+ "codex"
5400
+ ];
5401
+ function hasCommand(command) {
5402
+ const result = spawnSync5("sh", ["-c", 'command -v "$1" >/dev/null', "sh", command], { stdio: "ignore" });
5403
+ return (result.status ?? 1) === 0;
5404
+ }
5405
+ function commandVersion(command) {
5406
+ const result = spawnSync5(command, ["--version"], {
5407
+ encoding: "utf8",
5408
+ stdio: ["ignore", "pipe", "pipe"]
5409
+ });
5410
+ if ((result.status ?? 1) !== 0)
5411
+ return;
5412
+ return (result.stdout || result.stderr).trim().split(/\r?\n/)[0];
5413
+ }
5414
+ function runDoctor(store) {
5415
+ const checks = [];
5416
+ try {
5417
+ const dir = ensureDataDir();
5418
+ accessSync2(dir, constants2.R_OK | constants2.W_OK);
5419
+ checks.push({ id: "data-dir", status: "ok", message: "data directory is writable", detail: dir });
5420
+ } catch (error) {
5421
+ checks.push({
5422
+ id: "data-dir",
5423
+ status: "fail",
5424
+ message: "data directory is not writable",
5425
+ detail: error instanceof Error ? error.message : String(error)
5426
+ });
5427
+ }
5428
+ const bunVersion = commandVersion("bun");
5429
+ checks.push(bunVersion ? { id: "bun", status: "ok", message: "bun is available", detail: bunVersion } : { id: "bun", status: "fail", message: "bun is not available on PATH" });
5430
+ const accountsVersion = commandVersion("accounts");
5431
+ checks.push(accountsVersion ? { id: "accounts", status: "ok", message: "accounts is available", detail: accountsVersion } : { id: "accounts", status: "warn", message: "accounts CLI is not available; account-routed steps will fail" });
5432
+ try {
5433
+ const machines = listOpenMachines();
5434
+ const local = machines.find((machine) => machine.local);
5435
+ checks.push({
5436
+ id: "machines",
5437
+ status: "ok",
5438
+ message: `OpenMachines topology available (${machines.length} machine(s))`,
5439
+ detail: local ? `local=${local.id}` : undefined
5440
+ });
5441
+ } catch (error) {
5442
+ checks.push({
5443
+ id: "machines",
5444
+ status: "warn",
5445
+ message: "OpenMachines topology is not available; machine-assigned loops will fail",
5446
+ detail: error instanceof Error ? error.message : String(error)
5447
+ });
5448
+ }
5449
+ for (const command of PROVIDER_COMMANDS) {
5450
+ checks.push(hasCommand(command) ? { id: `provider:${command}`, status: "ok", message: `${command} is available` } : { id: `provider:${command}`, status: "warn", message: `${command} is not on PATH` });
5451
+ }
5452
+ const status = daemonStatus(store);
5453
+ checks.push(status.running ? { id: "daemon", status: "ok", message: `daemon is running pid=${status.pid}` } : { id: "daemon", status: status.stale ? "warn" : "ok", message: status.stale ? "daemon pid file is stale" : "daemon is not running" });
5454
+ const failedRuns = store.countRuns("failed");
5455
+ checks.push(failedRuns === 0 ? { id: "loop-runs", status: "ok", message: "no failed loop runs recorded" } : { id: "loop-runs", status: "warn", message: `${failedRuns} failed loop run(s) recorded` });
5456
+ for (const loop of store.listLoops({ status: "active" })) {
5457
+ try {
5458
+ if (loop.target.type === "workflow") {
5459
+ const workflow = store.requireWorkflow(loop.target.workflowId);
5460
+ for (const step of workflowExecutionOrder(workflow)) {
5461
+ preflightTarget({
5462
+ ...step.target,
5463
+ account: step.account ?? step.target.account,
5464
+ timeoutMs: step.timeoutMs !== undefined ? step.timeoutMs : step.target.timeoutMs
5465
+ }, { loopId: loop.id, loopName: loop.name, workflowId: workflow.id, workflowName: workflow.name, workflowStepId: step.id }, { machine: loop.machine });
5466
+ }
5467
+ } else {
5468
+ preflightTarget(loop.target, { loopId: loop.id, loopName: loop.name }, { machine: loop.machine });
5469
+ }
5470
+ checks.push({ id: `loop:${loop.id}:preflight`, status: "ok", message: `active loop target is ready: ${loop.name}` });
5471
+ } catch (error) {
5472
+ checks.push({
5473
+ id: `loop:${loop.id}:preflight`,
5474
+ status: "fail",
5475
+ message: `active loop target preflight failed: ${loop.name}`,
5476
+ detail: error instanceof Error ? error.message : String(error)
5477
+ });
5478
+ }
5479
+ }
5480
+ return {
5481
+ ok: checks.every((check) => check.status !== "fail"),
5482
+ checks
5483
+ };
5484
+ }
5485
+
5486
+ // src/lib/health.ts
5487
+ import { createHash as createHash2 } from "crypto";
5488
+ import { existsSync as existsSync4, readFileSync as readFileSync4 } from "fs";
5489
+
5490
+ // src/lib/format.ts
5491
+ var TEXT_OUTPUT_LIMIT = 32 * 1024;
5492
+ var SENSITIVE_PAYLOAD_KEYS = new Set(["env", "error", "prompt", "reason", "stderr", "stdout"]);
5493
+ function redact(value, visible = 0) {
5494
+ if (!value)
5495
+ return value;
5496
+ if (value.length <= visible)
5497
+ return value;
5498
+ if (visible <= 0)
5499
+ return `[redacted ${value.length} chars]`;
5500
+ return `${value.slice(0, visible)}... [redacted ${value.length - visible} chars]`;
5501
+ }
5502
+ function truncateTextOutput(value) {
5503
+ if (value.length <= TEXT_OUTPUT_LIMIT)
5504
+ return value;
5505
+ return `${value.slice(0, TEXT_OUTPUT_LIMIT)}
5506
+ [truncated ${value.length - TEXT_OUTPUT_LIMIT} chars]`;
5507
+ }
5508
+ function redactSensitivePayload(value, key) {
5509
+ if (key && SENSITIVE_PAYLOAD_KEYS.has(key)) {
5510
+ if (typeof value === "string")
5511
+ return redact(value);
5512
+ if (value === undefined || value === null)
5513
+ return value;
5514
+ return "[redacted]";
5515
+ }
5516
+ if (Array.isArray(value))
5517
+ return value.map((item) => redactSensitivePayload(item));
5518
+ if (value && typeof value === "object") {
5519
+ return Object.fromEntries(Object.entries(value).map(([entryKey, entryValue]) => [entryKey, redactSensitivePayload(entryValue, entryKey)]));
5520
+ }
5521
+ return value;
5522
+ }
5523
+ function textOutputBlocks(value, opts = {}) {
5524
+ const indent = opts.indent ?? "";
5525
+ const nested = `${indent} `;
5526
+ const blocks = [];
5527
+ for (const [label, output] of [
5528
+ ["stdout", value.stdout],
5529
+ ["stderr", value.stderr]
5530
+ ]) {
5531
+ if (!output)
5532
+ continue;
5533
+ blocks.push(`${indent}${label}:`);
5534
+ for (const line of truncateTextOutput(output).replace(/\s+$/, "").split(/\r?\n/)) {
5535
+ blocks.push(`${nested}${line}`);
5536
+ }
5537
+ }
5538
+ return blocks;
5539
+ }
5540
+ function publicLoop(loop) {
5541
+ const target = loop.target.type === "command" ? { ...loop.target, env: loop.target.env ? "[redacted]" : undefined } : loop.target.type === "agent" ? { ...loop.target, prompt: redact(loop.target.prompt) } : loop.target;
5542
+ return {
5543
+ ...loop,
5544
+ target
5545
+ };
5546
+ }
5547
+ function publicRun(run, showOutput = false) {
5548
+ return {
5549
+ ...run,
5550
+ stdout: showOutput ? run.stdout : run.stdout ? `[redacted ${run.stdout.length} chars]` : undefined,
5551
+ stderr: showOutput ? run.stderr : run.stderr ? `[redacted ${run.stderr.length} chars]` : undefined
5552
+ };
5553
+ }
5554
+ function publicExecutorResult(result, showOutput = false) {
5555
+ return {
5556
+ ...result,
5557
+ stdout: showOutput ? result.stdout : result.stdout ? `[redacted ${result.stdout.length} chars]` : undefined,
5558
+ stderr: showOutput ? result.stderr : result.stderr ? `[redacted ${result.stderr.length} chars]` : undefined,
5559
+ error: redact(result.error)
5560
+ };
5561
+ }
5562
+ function publicWorkflow(workflow) {
5563
+ return {
5564
+ ...workflow,
5565
+ steps: workflow.steps.map((step) => ({
5566
+ ...step,
5567
+ target: step.target.type === "agent" ? { ...step.target, prompt: redact(step.target.prompt) } : step.target.type === "command" && step.target.env ? { ...step.target, env: "[redacted]" } : step.target
5568
+ }))
5569
+ };
5570
+ }
5571
+ function publicWorkflowRun(run) {
5572
+ return { ...run, error: redact(run.error) };
5573
+ }
5574
+ function publicWorkflowInvocation(invocation) {
5575
+ return redactSensitivePayload(invocation);
5576
+ }
5577
+ function publicWorkflowWorkItem(item) {
5578
+ return { ...item, lastReason: redact(item.lastReason, 240) };
5579
+ }
5580
+ function publicWorkflowStepRun(run, showOutput = false) {
5581
+ return {
5582
+ ...run,
5583
+ stdout: showOutput ? run.stdout : run.stdout ? `[redacted ${run.stdout.length} chars]` : undefined,
5584
+ stderr: showOutput ? run.stderr : run.stderr ? `[redacted ${run.stderr.length} chars]` : undefined,
5585
+ error: redact(run.error)
5586
+ };
5587
+ }
5588
+ function publicWorkflowEvent(event) {
5589
+ return { ...event, payload: redactSensitivePayload(event.payload) };
5590
+ }
5591
+ function publicGoal(goal) {
5592
+ return {
5593
+ ...goal,
5594
+ objective: redact(goal.objective, 120)
5595
+ };
5596
+ }
5597
+ function publicGoalRun(run) {
5598
+ return {
5599
+ ...run,
5600
+ evidence: redactSensitivePayload(run.evidence),
5601
+ rawResponse: redactSensitivePayload(run.rawResponse)
5602
+ };
5603
+ }
5604
+
5605
+ // src/lib/health.ts
5606
+ var EVIDENCE_CHARS = 2000;
5607
+ var FINGERPRINT_EVIDENCE_CHARS = 120;
5608
+ var CLASSIFICATIONS = [
5609
+ "rate_limit",
5610
+ "auth",
5611
+ "model_not_found",
5612
+ "context_length",
5613
+ "schema_response_format",
5614
+ "node_init",
5615
+ "preflight",
5616
+ "route_functional",
5617
+ "timeout",
5618
+ "sigsegv",
5619
+ "skipped_previous_active",
5620
+ "circuit_breaker",
5621
+ "unknown"
5622
+ ];
5623
+ function bounded(value, limit = EVIDENCE_CHARS) {
5624
+ if (!value)
5625
+ return;
5626
+ if (value.length <= limit)
5627
+ return value;
5628
+ return `${value.slice(0, limit)}
5629
+ [truncated ${value.length - limit} chars]`;
5630
+ }
5631
+ function redactedEvidence(value) {
5632
+ return redact(bounded(value));
5633
+ }
5634
+ function searchableText(run) {
5635
+ return [run.error, run.stderr, run.stdout].filter(Boolean).join(`
5636
+ `).toLowerCase();
5637
+ }
5638
+ function isRecord(value) {
5639
+ return Boolean(value && typeof value === "object" && !Array.isArray(value));
5640
+ }
5641
+ function stringValue(value) {
5642
+ return typeof value === "string" && value.trim() ? value : undefined;
5643
+ }
5644
+ function objectField(value, key) {
5645
+ if (!value)
5646
+ return;
5647
+ const field = value[key];
5648
+ return isRecord(field) ? field : undefined;
5649
+ }
5650
+ function tagsFromValue(value) {
5651
+ if (Array.isArray(value))
5652
+ return value.map((entry) => String(entry).trim()).filter(Boolean);
5653
+ if (typeof value === "string")
5654
+ return value.split(",").map((entry) => entry.trim()).filter(Boolean);
5655
+ return [];
5656
+ }
5657
+ function stableFingerprint(parts) {
5658
+ return createHash2("sha256").update(parts.join(`
5659
+ `)).digest("hex").slice(0, 16);
5660
+ }
5661
+ function stableFailureFingerprint(run, classification) {
5662
+ return stableFingerprint([
5663
+ run.loopId,
5664
+ classification,
5665
+ String(run.status),
5666
+ String(run.exitCode ?? ""),
5667
+ (run.error ?? run.stderr ?? run.stdout ?? "").replace(/\d{4}-\d{2}-\d{2}T\S+/g, "<timestamp>").slice(0, FINGERPRINT_EVIDENCE_CHARS)
5668
+ ]);
5669
+ }
5670
+ function stableRouteFunctionalFingerprint(loop, reason) {
5671
+ return stableFingerprint([
5672
+ loop.id,
5673
+ "route_functional",
5674
+ reason.replace(/\d{4}-\d{2}-\d{2}T\S+/g, "<timestamp>").slice(0, FINGERPRINT_EVIDENCE_CHARS)
5675
+ ]);
5676
+ }
5677
+ function healthRun(run) {
5678
+ return {
5679
+ ...run,
5680
+ error: redactedEvidence(run.error),
5681
+ stdout: redactedEvidence(run.stdout),
5682
+ stderr: redactedEvidence(run.stderr)
5683
+ };
5684
+ }
5685
+ function classifyRunFailure(run) {
5686
+ if (run.status === "succeeded" || run.status === "running")
5687
+ return;
5688
+ const text = searchableText(run);
5689
+ let classification = "unknown";
5690
+ if (run.status === "timed_out")
5691
+ classification = "timeout";
5692
+ else if (run.status === "skipped" && /circuit breaker open/.test(text))
5693
+ classification = "circuit_breaker";
5694
+ else if (run.status === "skipped" && /previous run still active/.test(text))
5695
+ classification = "skipped_previous_active";
5696
+ else if (/runtime preflight failed|preflight failed|executable not found in path|none of required executables found|auth profile preflight failed|profile not found/.test(text))
5697
+ classification = "preflight";
5698
+ else if (/rate limit|too many requests|429\b|quota exceeded/.test(text))
5699
+ classification = "rate_limit";
5700
+ else if (/unauthorized|authentication|auth\b|api key|invalid token|permission denied|401\b|403\b/.test(text))
5701
+ classification = "auth";
5702
+ else if (/model .*not found|model_not_found|unknown model|invalid model|404.*model/.test(text))
5703
+ classification = "model_not_found";
5704
+ else if (/context length|context_length|context window|maximum context|token limit|too many tokens/.test(text))
5705
+ classification = "context_length";
5706
+ else if (/response_format|json schema|schema validation|invalid schema|structured output/.test(text))
5707
+ classification = "schema_response_format";
5708
+ else if (/cannot find module|module not found|node:internal|bun: command not found|node: command not found|npm err!|err_module_not_found/.test(text))
5709
+ classification = "node_init";
5710
+ else if (/sigsegv|segmentation fault|signal 11/.test(text))
5711
+ classification = "sigsegv";
5712
+ return {
5713
+ classification,
5714
+ fingerprint: stableFailureFingerprint(run, classification),
5715
+ evidence: {
5716
+ error: redactedEvidence(run.error),
5717
+ stdout: redactedEvidence(run.stdout),
5718
+ stderr: redactedEvidence(run.stderr),
5719
+ exitCode: run.exitCode
5720
+ }
5721
+ };
5722
+ }
5723
+ var ROUTE_FUNCTIONAL_DISALLOWED_TAGS = new Set([
5724
+ "no-auto",
5725
+ "manual",
5726
+ "manual-required",
5727
+ "approval-required",
5728
+ "blocked",
5729
+ "completed",
5730
+ "done",
5731
+ "cancelled",
5732
+ "canceled",
5733
+ "failed",
5734
+ "archived"
5735
+ ]);
5736
+ var ROUTE_FUNCTIONAL_DISALLOWED_STATUSES = new Set([
5737
+ "blocked",
5738
+ "completed",
5739
+ "done",
5740
+ "cancelled",
5741
+ "canceled",
5742
+ "failed",
5743
+ "archived"
5744
+ ]);
5745
+ function parseJsonObject(raw) {
5746
+ if (!raw?.trim())
5747
+ return;
5748
+ try {
5749
+ const parsed = JSON.parse(raw);
5750
+ return isRecord(parsed) ? parsed : undefined;
5751
+ } catch {
5752
+ return;
5753
+ }
5754
+ }
5755
+ function routeEvidenceReport(run) {
5756
+ const stdoutReport = parseJsonObject(run.stdout);
5757
+ const evidencePath = stringValue(stdoutReport?.evidencePath);
5758
+ if (evidencePath && existsSync4(evidencePath)) {
5759
+ try {
5760
+ return parseJsonObject(readFileSync4(evidencePath, "utf8")) ?? stdoutReport;
5761
+ } catch {
5762
+ return stdoutReport;
5763
+ }
5764
+ }
5765
+ return stdoutReport;
5766
+ }
5767
+ function commandName(command) {
5768
+ return command.split(/[\\/]/).at(-1) ?? command;
5769
+ }
5770
+ function argsContainSequence(args, sequence) {
5771
+ for (let index = 0;index <= args.length - sequence.length; index += 1) {
5772
+ if (sequence.every((part, offset) => args[index + offset] === part))
5773
+ return true;
5774
+ }
5775
+ return false;
5776
+ }
5777
+ function isRouteDrainLoop(loop) {
5778
+ if (loop.target.type !== "command")
5779
+ return false;
5780
+ if (commandName(loop.target.command) !== "loops")
5781
+ return false;
5782
+ const args = loop.target.args ?? [];
5783
+ return argsContainSequence(args, ["events", "drain", "todos-task"]) || argsContainSequence(args, ["routes", "drain", "todos-task"]) || argsContainSequence(args, ["route", "drain", "todos-task"]);
5784
+ }
5785
+ function routeResultTaskState(result) {
5786
+ const event = objectField(result, "event");
5787
+ const data = objectField(event, "data");
5788
+ const task = objectField(data, "task");
5789
+ const payload = objectField(data, "payload");
5790
+ const payloadTask = objectField(payload, "task");
5791
+ const metadata = objectField(data, "metadata");
5792
+ const records = [data, task, payload, payloadTask, metadata].filter(isRecord);
5793
+ const tags = new Set;
5794
+ for (const record of records) {
5795
+ for (const tag of tagsFromValue(record.tags ?? record.task_tags ?? record.taskTags)) {
5796
+ tags.add(tag.toLowerCase());
5797
+ }
5798
+ }
5799
+ const status = records.map((record) => stringValue(record.status ?? record.task_status ?? record.taskStatus)?.toLowerCase()).find(Boolean);
5800
+ return {
5801
+ taskId: stringValue(event?.subject) ?? stringValue(data?.id) ?? stringValue(task?.id) ?? stringValue(payloadTask?.id),
5802
+ tags: [...tags],
5803
+ status
5804
+ };
5805
+ }
5806
+ function detectRouteFunctionalFailure(store, loop, run) {
5807
+ if (run.status !== "succeeded")
5808
+ return;
5809
+ if (!isRouteDrainLoop(loop))
5810
+ return;
5811
+ const report = routeEvidenceReport(run);
5812
+ const rawResults = Array.isArray(report?.results) ? report.results.filter(isRecord) : [];
5813
+ for (const result of rawResults) {
5814
+ const kind = stringValue(result.kind);
5815
+ const task = routeResultTaskState(result);
5816
+ const disallowedTag = task.tags.find((tag) => ROUTE_FUNCTIONAL_DISALLOWED_TAGS.has(tag));
5817
+ if (kind && kind !== "skipped" && disallowedTag) {
5818
+ const reason2 = `route drain ${kind} task ${task.taskId ?? "unknown"} with disallowed tag ${disallowedTag}`;
5819
+ return {
5820
+ classification: "route_functional",
5821
+ fingerprint: stableRouteFunctionalFingerprint(loop, reason2),
5822
+ evidence: { error: reason2, stdout: redactedEvidence(run.stdout), exitCode: run.exitCode }
5823
+ };
5824
+ }
5825
+ if (kind && kind !== "skipped" && task.status && ROUTE_FUNCTIONAL_DISALLOWED_STATUSES.has(task.status)) {
5826
+ const reason2 = `route drain ${kind} task ${task.taskId ?? "unknown"} with non-routable status ${task.status}`;
5827
+ return {
5828
+ classification: "route_functional",
5829
+ fingerprint: stableRouteFunctionalFingerprint(loop, reason2),
5830
+ evidence: { error: reason2, stdout: redactedEvidence(run.stdout), exitCode: run.exitCode }
5831
+ };
5832
+ }
5833
+ const reason = stringValue(result.reason);
5834
+ if (kind === "skipped" && reason === "task metadata requires manual or approval-gated handling") {
5835
+ const message = `route drain skipped task ${task.taskId ?? "unknown"} with ambiguous manual-gate reason`;
5836
+ return {
5837
+ classification: "route_functional",
5838
+ fingerprint: stableRouteFunctionalFingerprint(loop, message),
5839
+ evidence: { error: message, stdout: redactedEvidence(run.stdout), exitCode: run.exitCode }
5840
+ };
5841
+ }
5842
+ const sourceTaskUpdate = objectField(result, "sourceTaskUpdate");
5843
+ if (kind === "skipped" && sourceTaskUpdate && sourceTaskUpdate.ok === false) {
5844
+ const updateError = stringValue(sourceTaskUpdate.error);
5845
+ const message = `route drain skipped task ${task.taskId ?? "unknown"} but failed to update source task${updateError ? `: ${updateError}` : ""}`;
5846
+ return {
5847
+ classification: "route_functional",
5848
+ fingerprint: stableRouteFunctionalFingerprint(loop, message),
5849
+ evidence: { error: message, stdout: redactedEvidence(run.stdout), exitCode: run.exitCode }
5850
+ };
5851
+ }
5852
+ const childLoopId = stringValue(objectField(result, "loop")?.id) ?? stringValue(result.loopId);
5853
+ const childRun = childLoopId ? store.listRuns({ loopId: childLoopId, limit: 1 })[0] : undefined;
5854
+ if (childRun && !["succeeded", "running"].includes(childRun.status)) {
5855
+ const message = `route drain ${kind ?? "handled"} task ${task.taskId ?? "unknown"} but child loop ${childLoopId} latest run is ${childRun.status}`;
5856
+ return {
5857
+ classification: "route_functional",
5858
+ fingerprint: stableRouteFunctionalFingerprint(loop, message),
5859
+ evidence: { error: message, stdout: redactedEvidence(run.stdout), stderr: redactedEvidence(childRun.stderr), exitCode: childRun.exitCode }
5860
+ };
5861
+ }
5862
+ }
5863
+ return;
5864
+ }
5865
+ function targetRoute(loop) {
5866
+ if (loop.target.type === "agent") {
4486
5867
  return {
4487
- status: "timed_out",
4488
- exitCode,
4489
- stdout,
4490
- stderr,
4491
- error: idleTimedOut ? `idle timed out after ${spec.idleTimeoutMs}ms without stdout/stderr` : `timed out after ${spec.timeoutMs}ms`,
4492
- pid: child.pid,
4493
- startedAt,
4494
- finishedAt,
4495
- durationMs
5868
+ source: "openloops",
5869
+ kind: "loop_expectation",
5870
+ loopId: loop.id,
5871
+ loopName: loop.name,
5872
+ cwd: loop.target.cwd,
5873
+ provider: loop.target.provider
4496
5874
  };
4497
5875
  }
4498
- if (error || exitCode !== 0) {
5876
+ if (loop.target.type === "command") {
4499
5877
  return {
4500
- status: "failed",
4501
- exitCode,
4502
- stdout,
4503
- stderr,
4504
- error: error ?? `process exited with code ${exitCode ?? "unknown"}`,
4505
- pid: child.pid,
4506
- startedAt,
4507
- finishedAt,
4508
- durationMs
5878
+ source: "openloops",
5879
+ kind: "loop_expectation",
5880
+ loopId: loop.id,
5881
+ loopName: loop.name,
5882
+ cwd: loop.target.cwd
4509
5883
  };
4510
5884
  }
4511
5885
  return {
4512
- status: "succeeded",
4513
- exitCode,
4514
- stdout,
4515
- stderr,
4516
- pid: child.pid,
4517
- startedAt,
4518
- finishedAt,
4519
- durationMs
5886
+ source: "openloops",
5887
+ kind: "loop_expectation",
5888
+ loopId: loop.id,
5889
+ loopName: loop.name
4520
5890
  };
4521
5891
  }
4522
- async function executeLoop(loop, run, opts = {}) {
4523
- if (loop.target.type === "workflow") {
4524
- throw new Error("workflow loop targets must be executed with executeLoopTarget");
5892
+ function recommendedTask(loop, run, failure, route) {
5893
+ const title = `BUG: open-loops loop failure - ${loop.name}`;
5894
+ const description = [
5895
+ `OpenLoops expectation failed for loop ${loop.name} (${loop.id}).`,
5896
+ `Run: ${run.id}`,
5897
+ `Status: ${run.status}`,
5898
+ `Classification: ${failure.classification}`,
5899
+ `Fingerprint: ${failure.fingerprint}`,
5900
+ `No-tmux routing: Do not dispatch or paste prompts into tmux panes; use task-triggered headless worker/verifier workflows only.`,
5901
+ route.cwd ? `Route cwd: ${route.cwd}` : undefined,
5902
+ route.provider ? `Provider: ${route.provider}` : undefined,
5903
+ failure.evidence.error ? `Error:
5904
+ ${failure.evidence.error}` : undefined,
5905
+ failure.evidence.stderr ? `Stderr:
5906
+ ${failure.evidence.stderr}` : undefined
5907
+ ].filter(Boolean).join(`
5908
+
5909
+ `);
5910
+ const dedupeKey = `openloops:${loop.id}:${failure.fingerprint}`;
5911
+ const tags = ["bug", "openloops", "loop-health", failure.classification];
5912
+ const priority = failure.classification === "auth" || failure.classification === "rate_limit" ? "high" : "medium";
5913
+ return {
5914
+ title,
5915
+ description,
5916
+ priority,
5917
+ tags,
5918
+ dedupeKey,
5919
+ search: { query: dedupeKey },
5920
+ compatibilityFallback: {
5921
+ search: ["todos", "search", dedupeKey, "--json"],
5922
+ add: ["todos", "add", title, "--description", description, "--tag", tags.join(","), "--priority", priority],
5923
+ comment: ["todos", "comment", "<task-id>", description]
5924
+ },
5925
+ futureNativeUpsert: {
5926
+ command: "todos upsert",
5927
+ fields: {
5928
+ title,
5929
+ description,
5930
+ priority,
5931
+ tags,
5932
+ dedupeKey,
5933
+ routeSource: route.source,
5934
+ routeKind: route.kind,
5935
+ routeLoopId: route.loopId,
5936
+ routeLoopName: route.loopName
5937
+ }
5938
+ }
5939
+ };
5940
+ }
5941
+ function expectationForLoop(store, loop) {
5942
+ const latestRun = store.listRuns({ loopId: loop.id, limit: 1 })[0];
5943
+ const route = targetRoute(loop);
5944
+ if (!latestRun) {
5945
+ return {
5946
+ loop: { id: loop.id, name: loop.name, status: loop.status, nextRunAt: loop.nextRunAt },
5947
+ ok: true,
5948
+ check: { id: "latest-run-succeeded", status: "warn", message: "loop has no recorded runs yet" },
5949
+ route
5950
+ };
4525
5951
  }
4526
- if (loop.target.preflight?.beforeRun) {
4527
- const startedAt = nowIso();
4528
- try {
4529
- preflightTarget(loop.target, {
4530
- loopId: loop.id,
4531
- loopName: loop.name,
4532
- runId: run.id,
4533
- scheduledFor: run.scheduledFor
4534
- }, { ...opts, machine: opts.machine ?? loop.machine });
4535
- } catch (error) {
4536
- const finishedAt = nowIso();
5952
+ const routeFailure = detectRouteFunctionalFailure(store, loop, latestRun);
5953
+ if (routeFailure) {
5954
+ return {
5955
+ loop: { id: loop.id, name: loop.name, status: loop.status, nextRunAt: loop.nextRunAt },
5956
+ ok: false,
5957
+ check: { id: "route-functional-health", status: "fail", message: routeFailure.evidence.error ?? "route functional blocker detected" },
5958
+ latestRun: healthRun(latestRun),
5959
+ failure: routeFailure,
5960
+ route,
5961
+ recommendedTask: recommendedTask(loop, latestRun, routeFailure, route)
5962
+ };
5963
+ }
5964
+ if (latestRun.status === "succeeded") {
5965
+ return {
5966
+ loop: { id: loop.id, name: loop.name, status: loop.status, nextRunAt: loop.nextRunAt },
5967
+ ok: true,
5968
+ check: { id: "latest-run-succeeded", status: "pass", message: "latest run succeeded" },
5969
+ latestRun: healthRun(latestRun),
5970
+ route
5971
+ };
5972
+ }
5973
+ const failure = classifyRunFailure(latestRun);
5974
+ if (failure?.classification === "circuit_breaker") {
5975
+ if (loop.status !== "paused") {
4537
5976
  return {
4538
- status: "failed",
4539
- stdout: "",
4540
- stderr: "",
4541
- error: `runtime preflight failed: ${error instanceof Error ? error.message : String(error)}`,
4542
- startedAt,
4543
- finishedAt,
4544
- durationMs: new Date(finishedAt).getTime() - new Date(startedAt).getTime()
5977
+ loop: { id: loop.id, name: loop.name, status: loop.status, nextRunAt: loop.nextRunAt },
5978
+ ok: true,
5979
+ check: { id: "latest-run-succeeded", status: "warn", message: "circuit breaker cleared by resume; awaiting next run" },
5980
+ latestRun: healthRun(latestRun),
5981
+ route
4545
5982
  };
4546
5983
  }
5984
+ return {
5985
+ loop: { id: loop.id, name: loop.name, status: loop.status, nextRunAt: loop.nextRunAt },
5986
+ ok: false,
5987
+ check: { id: "latest-run-succeeded", status: "fail", message: latestRun.error ?? "circuit breaker open; loop auto-paused" },
5988
+ latestRun: healthRun(latestRun),
5989
+ failure,
5990
+ route,
5991
+ recommendedTask: recommendedTask(loop, latestRun, failure, route)
5992
+ };
4547
5993
  }
4548
- return executeTarget(loop.target, {
4549
- loopId: loop.id,
4550
- loopName: loop.name,
4551
- runId: run.id,
4552
- scheduledFor: run.scheduledFor
4553
- }, { ...opts, machine: opts.machine ?? loop.machine });
5994
+ return {
5995
+ loop: { id: loop.id, name: loop.name, status: loop.status, nextRunAt: loop.nextRunAt },
5996
+ ok: false,
5997
+ check: { id: "latest-run-succeeded", status: "fail", message: `latest run is ${latestRun.status}` },
5998
+ latestRun: healthRun(latestRun),
5999
+ failure,
6000
+ route,
6001
+ recommendedTask: failure ? recommendedTask(loop, latestRun, failure, route) : undefined
6002
+ };
4554
6003
  }
4555
-
4556
- // src/lib/goal/runner.ts
4557
- import { generateObject } from "ai";
4558
- import { z } from "zod";
4559
-
4560
- // src/lib/goal/model-factory.ts
4561
- import { createOpenRouter } from "@openrouter/ai-sdk-provider";
4562
- var DEFAULT_GOAL_MODEL = "openai/gpt-4o-mini";
4563
- function resolveGoalModel(opts = {}) {
4564
- const env = opts.env ?? process.env;
4565
- const apiKey = opts.apiKey ?? env.OPENROUTER_API_KEY;
4566
- if (!apiKey) {
4567
- throw new Error("OPENROUTER_API_KEY is required to run goals with the OpenRouter AI SDK provider");
6004
+ function buildHealthReport(store, opts = {}) {
6005
+ const loops = store.listLoops({ includeArchived: opts.includeArchived, limit: opts.limit ?? 200 }).filter((loop) => opts.includeInactive || loop.status === "active" || loop.status === "paused");
6006
+ const expectations = loops.map((loop) => expectationForLoop(store, loop));
6007
+ const classifications = Object.fromEntries(CLASSIFICATIONS.map((key) => [key, 0]));
6008
+ for (const expectation of expectations) {
6009
+ if (expectation.failure)
6010
+ classifications[expectation.failure.classification] += 1;
4568
6011
  }
4569
- const provider = createOpenRouter({
4570
- apiKey,
4571
- baseURL: opts.baseURL ?? env.LOOPS_GOAL_BASE_URL ?? env.OPENROUTER_BASE_URL
4572
- });
4573
- return provider.chat(opts.model ?? env.LOOPS_GOAL_MODEL ?? DEFAULT_GOAL_MODEL);
6012
+ const unhealthy = expectations.filter((expectation) => !expectation.ok).length;
6013
+ const warnings = expectations.filter((expectation) => expectation.check.status === "warn").length;
6014
+ return {
6015
+ ok: unhealthy === 0,
6016
+ generatedAt: new Date().toISOString(),
6017
+ summary: {
6018
+ loops: expectations.length,
6019
+ healthy: expectations.length - unhealthy,
6020
+ unhealthy,
6021
+ warnings
6022
+ },
6023
+ classifications,
6024
+ expectations
6025
+ };
6026
+ }
6027
+
6028
+ // src/lib/goal/metadata.ts
6029
+ function goalExecutionContext(parts) {
6030
+ return {
6031
+ loopId: parts.loop?.id,
6032
+ loopName: parts.loop?.name,
6033
+ loopRunId: parts.loopRun?.id,
6034
+ scheduledFor: parts.loopRun?.scheduledFor ?? parts.scheduledFor,
6035
+ workflowId: parts.workflow?.id,
6036
+ workflowName: parts.workflow?.name,
6037
+ workflowRunId: parts.workflowRunId,
6038
+ workflowStepId: parts.workflowStepId
6039
+ };
6040
+ }
6041
+ function executionMetadata(context, goal, node) {
6042
+ return {
6043
+ loopId: context?.loopId,
6044
+ loopName: context?.loopName,
6045
+ runId: context?.loopRunId,
6046
+ scheduledFor: context?.scheduledFor,
6047
+ workflowId: context?.workflowId,
6048
+ workflowName: context?.workflowName,
6049
+ workflowRunId: context?.workflowRunId,
6050
+ workflowStepId: context?.workflowStepId,
6051
+ goalId: goal?.goalId,
6052
+ goalObjective: goal?.objective,
6053
+ goalNodeKey: node?.key
6054
+ };
6055
+ }
6056
+ function withGoalNodeEnv(env, node) {
6057
+ return { ...env ?? process.env, LOOPS_GOAL_NODE_KEY: node.key, LOOPS_GOAL_NODE_OBJECTIVE: node.objective };
4574
6058
  }
4575
6059
 
4576
6060
  // src/lib/goal/prompts.ts
@@ -4586,6 +6070,23 @@ function planPrompt(spec) {
4586
6070
  ].join(`
4587
6071
  `);
4588
6072
  }
6073
+ function iterationPrompt(goal, node) {
6074
+ return [
6075
+ "Continue executing the goal plan without losing budget discipline.",
6076
+ `Goal: ${goal.objective}`,
6077
+ `Plan node: ${node.key}`,
6078
+ `Node objective: ${node.objective}`,
6079
+ "Work only on this node's objective in this run; other plan nodes run in their own iterations."
6080
+ ].join(`
6081
+ `);
6082
+ }
6083
+ function goalNodeTarget(target, goal, node) {
6084
+ if (target.type !== "agent")
6085
+ return target;
6086
+ return { ...target, prompt: `${target.prompt}
6087
+
6088
+ ${iterationPrompt(goal, node)}` };
6089
+ }
4589
6090
  function achievementPrompt(goal, nodes, evidence) {
4590
6091
  return [
4591
6092
  "Run an adversarial achievement audit.",
@@ -4605,6 +6106,79 @@ ${evidence.join(`
4605
6106
  `);
4606
6107
  }
4607
6108
 
6109
+ // src/lib/goal/runner.ts
6110
+ import { generateObject } from "ai";
6111
+ import { z } from "zod";
6112
+
6113
+ // src/lib/run-envelope.ts
6114
+ var ENVELOPE_EXCERPT_CHARS = 2048;
6115
+ var BLOCKED_STEP_ERROR_PREFIX = "blocked:";
6116
+ function boundedExcerpt(text, limit = ENVELOPE_EXCERPT_CHARS) {
6117
+ if (!text)
6118
+ return;
6119
+ const scrubbed = scrubSecrets(text);
6120
+ if (scrubbed.length <= limit * 2)
6121
+ return scrubbed;
6122
+ const omitted = scrubbed.length - limit * 2;
6123
+ return `${scrubbed.slice(0, limit)}
6124
+ [... ${omitted} chars omitted ...]
6125
+ ${scrubbed.slice(-limit)}`;
6126
+ }
6127
+ function summarizeOutput(stdout, stderr) {
6128
+ return {
6129
+ stdoutBytes: stdout ? Buffer.byteLength(stdout, "utf8") : 0,
6130
+ stderrBytes: stderr ? Buffer.byteLength(stderr, "utf8") : 0,
6131
+ stdoutExcerpt: boundedExcerpt(stdout),
6132
+ stderrExcerpt: boundedExcerpt(stderr)
6133
+ };
6134
+ }
6135
+ function summarizeExecutorResult(result) {
6136
+ return {
6137
+ ...summarizeOutput(result.stdout, result.stderr),
6138
+ status: result.status,
6139
+ exitCode: result.exitCode,
6140
+ durationMs: result.durationMs,
6141
+ error: boundedExcerpt(result.error)
6142
+ };
6143
+ }
6144
+ function isBlockedStepRun(step) {
6145
+ return step?.status === "skipped" && (step.error?.startsWith(BLOCKED_STEP_ERROR_PREFIX) ?? false);
6146
+ }
6147
+ function summarizeWorkflowStepRun(step) {
6148
+ return {
6149
+ ...summarizeOutput(step.stdout, step.stderr),
6150
+ stepId: step.stepId,
6151
+ status: step.status,
6152
+ exitCode: step.exitCode,
6153
+ durationMs: step.durationMs,
6154
+ error: boundedExcerpt(step.error),
6155
+ blocked: isBlockedStepRun(step)
6156
+ };
6157
+ }
6158
+ function workflowRunEnvelope(run, steps) {
6159
+ return JSON.stringify({ workflowRun: run, steps: steps.map(summarizeWorkflowStepRun) }, null, 2);
6160
+ }
6161
+
6162
+ // src/lib/goal/model-factory.ts
6163
+ import { createOpenRouter } from "@openrouter/ai-sdk-provider";
6164
+ var DEFAULT_GOAL_MODEL = "openai/gpt-4o-mini";
6165
+ function resolveGoalModel(opts = {}) {
6166
+ const env = opts.env ?? process.env;
6167
+ const apiKey = opts.apiKey ?? env.OPENROUTER_API_KEY;
6168
+ if (!apiKey) {
6169
+ throw new Error("OPENROUTER_API_KEY is required to run goals with the OpenRouter AI SDK provider");
6170
+ }
6171
+ const provider = createOpenRouter({
6172
+ apiKey,
6173
+ baseURL: opts.baseURL ?? env.LOOPS_GOAL_BASE_URL ?? env.OPENROUTER_BASE_URL
6174
+ });
6175
+ return provider.chat(opts.model ?? env.LOOPS_GOAL_MODEL ?? DEFAULT_GOAL_MODEL);
6176
+ }
6177
+ function resolveGoalVerifierModel(opts = {}) {
6178
+ const env = opts.env ?? process.env;
6179
+ return resolveGoalModel({ ...opts, model: opts.model ?? env.LOOPS_GOAL_VERIFIER_MODEL });
6180
+ }
6181
+
4608
6182
  // src/lib/goal/runner.ts
4609
6183
  var DEFAULT_MAX_TURNS = 10;
4610
6184
  var PlanNodeSchema = z.object({
@@ -4732,34 +6306,39 @@ function noReadyDiagnostic(goal, nodes) {
4732
6306
  incompleteNodes: incompleteDetails
4733
6307
  };
4734
6308
  }
4735
- function metadataFor(goal, node, context) {
4736
- return {
4737
- loopId: context?.loopId,
4738
- loopName: context?.loopName,
4739
- runId: context?.loopRunId,
4740
- scheduledFor: context?.scheduledFor,
4741
- workflowId: context?.workflowId,
4742
- workflowName: context?.workflowName,
4743
- workflowRunId: context?.workflowRunId,
4744
- workflowStepId: context?.workflowStepId,
4745
- goalId: goal.goalId,
4746
- goalObjective: goal.objective,
4747
- goalNodeKey: node.key
4748
- };
4749
- }
4750
6309
  async function executeUnderlyingTarget(target, goal, node, opts) {
4751
- const metadata = metadataFor(goal, node, opts.context);
6310
+ const metadata = executionMetadata(opts.context, goal, node);
4752
6311
  if (opts.executeNode)
4753
6312
  return opts.executeNode(node, metadata);
4754
6313
  if (!target)
4755
6314
  throw new Error("runGoal requires either target or executeNode");
4756
- return executeTarget(target, metadata, {
4757
- env: opts.env,
6315
+ return executeTarget(goalNodeTarget(target, goal, node), metadata, {
6316
+ env: withGoalNodeEnv(opts.env, node),
4758
6317
  daemonLeaseId: opts.daemonLeaseId,
4759
6318
  beforePersist: opts.beforePersist,
4760
6319
  signal: opts.signal
4761
6320
  });
4762
6321
  }
6322
+ function verifierModelFor(spec, opts, planner) {
6323
+ if (opts.verifierModel)
6324
+ return opts.verifierModel;
6325
+ const env = opts.env ?? process.env;
6326
+ const configured = spec.verifierModel ?? env.LOOPS_GOAL_VERIFIER_MODEL;
6327
+ if (!configured)
6328
+ return planner;
6329
+ return resolveGoalVerifierModel({ model: configured, env: opts.env });
6330
+ }
6331
+ function nodeEvidence(node, result) {
6332
+ const summary = summarizeExecutorResult(result);
6333
+ return [
6334
+ `node ${node.key} ${summary.status} (exit ${summary.exitCode ?? "unknown"}, ${summary.durationMs}ms, ` + `stdout ${summary.stdoutBytes}B, stderr ${summary.stderrBytes}B)`,
6335
+ summary.stdoutExcerpt ? `stdout excerpt:
6336
+ ${summary.stdoutExcerpt}` : undefined,
6337
+ summary.stderrExcerpt ? `stderr excerpt:
6338
+ ${summary.stderrExcerpt}` : undefined
6339
+ ].filter(Boolean).join(`
6340
+ `);
6341
+ }
4763
6342
  async function planGoal(store, goal, spec, model, opts) {
4764
6343
  const existing = store.listGoalPlanNodes(goal.goalId);
4765
6344
  if (existing.length > 0)
@@ -4793,18 +6372,19 @@ async function planGoal(store, goal, spec, model, opts) {
4793
6372
  return store.createGoalPlanNodes(goal.goalId, rawNodes, { daemonLeaseId: opts.daemonLeaseId });
4794
6373
  }
4795
6374
  function stdoutFor(goal, nodes, evidence, validation, diagnostics) {
4796
- return JSON.stringify({
6375
+ return scrubSecrets(JSON.stringify(scrubSecretsDeep({
4797
6376
  goal,
4798
6377
  rollup: rollupSummary(nodes),
4799
6378
  nodes,
4800
6379
  evidence,
4801
6380
  validation,
4802
6381
  diagnostics
4803
- }, null, 2);
6382
+ }), null, 2));
4804
6383
  }
4805
6384
  async function runGoal(store, input, opts = {}) {
4806
6385
  const spec = normalizeGoalSpec2(input);
4807
6386
  const model = opts.model ?? resolveGoalModel({ model: spec.model, env: opts.env });
6387
+ const verifier = verifierModelFor(spec, opts, model);
4808
6388
  const startedAt = nowIso();
4809
6389
  const existing = store.findGoalByContext({
4810
6390
  loopRunId: opts.context?.loopRunId,
@@ -4837,6 +6417,13 @@ async function runGoal(store, input, opts = {}) {
4837
6417
  goal = store.updateGoalStatus(goal.goalId, "budgetLimited", { daemonLeaseId: opts.daemonLeaseId });
4838
6418
  return resultFromGoal(goal, "failed", stdoutFor(goal, nodes, evidence), "goal token budget exhausted after planning", startedAt);
4839
6419
  }
6420
+ if ((goal.autoExecute ?? spec.autoExecute) === "off") {
6421
+ nodes = syncReadyFlags(store, goal, nodes, opts);
6422
+ return resultFromGoal(goal, "succeeded", stdoutFor(goal, nodes, evidence, undefined, {
6423
+ autoExecute: "off",
6424
+ note: "autoExecute is off: goal plan persisted without executing any nodes"
6425
+ }), undefined, startedAt);
6426
+ }
4840
6427
  for (let turn = 1;turn <= (spec.maxTurns ?? DEFAULT_MAX_TURNS); turn++) {
4841
6428
  if (opts.signal?.aborted) {
4842
6429
  goal = store.updateGoalStatus(goal.goalId, "cancelled", { daemonLeaseId: opts.daemonLeaseId });
@@ -4875,11 +6462,7 @@ async function runGoal(store, input, opts = {}) {
4875
6462
  }
4876
6463
  }, { daemonLeaseId: opts.daemonLeaseId });
4877
6464
  if (result.status === "succeeded") {
4878
- evidence.push(`node ${node.key} succeeded
4879
- stdout:
4880
- ${result.stdout}
4881
- stderr:
4882
- ${result.stderr}`);
6465
+ evidence.push(nodeEvidence(node, result));
4883
6466
  store.updateGoalPlanNode(goal.goalId, node.key, {
4884
6467
  status: "complete",
4885
6468
  timeUsedSeconds: Math.round(result.durationMs / 1000)
@@ -4906,7 +6489,7 @@ ${result.stderr}`);
4906
6489
  }
4907
6490
  if (nodes.every((node) => node.status === "complete")) {
4908
6491
  const judged = await generateObject({
4909
- model,
6492
+ model: verifier,
4910
6493
  schema: AchievementSchema,
4911
6494
  temperature: 0,
4912
6495
  prompt: achievementPrompt(goal, nodes, evidence),
@@ -4975,19 +6558,31 @@ ${result.stderr}`);
4975
6558
  }
4976
6559
 
4977
6560
  // src/lib/workflow-runner.ts
4978
- function targetWithStepAccount(step) {
6561
+ var DEFAULT_BLOCKED_EXIT_CODES = [12];
6562
+ var GATE_STEP_PATTERN = /(^|[^a-z])gate([^a-z]|$)/i;
6563
+ function blockedExitCodesForStep(step) {
6564
+ const explicit = step.blockedExitCodes;
6565
+ if (explicit !== undefined)
6566
+ return explicit.filter((code) => Number.isInteger(code));
6567
+ return GATE_STEP_PATTERN.test(step.name ? `${step.id} ${step.name}` : step.id) ? DEFAULT_BLOCKED_EXIT_CODES : [];
6568
+ }
6569
+ function targetWithStepAccount(step, goalNodePrompt) {
4979
6570
  const account = step.account ?? step.target.account;
4980
6571
  const timeoutMs = step.timeoutMs !== undefined ? step.timeoutMs : step.target.timeoutMs;
4981
- if (!account && timeoutMs === step.target.timeoutMs)
4982
- return step.target;
4983
- return { ...step.target, account, timeoutMs };
6572
+ const target = !account && timeoutMs === step.target.timeoutMs ? step.target : { ...step.target, account, timeoutMs };
6573
+ if (goalNodePrompt && target.type === "agent") {
6574
+ return { ...target, prompt: `${target.prompt}
6575
+
6576
+ ${goalNodePrompt}` };
6577
+ }
6578
+ return target;
4984
6579
  }
4985
- function workflowResult(workflowRun, status, startedAt, finishedAt, stdout, error) {
6580
+ function workflowResult(workflowRun, status, startedAt, finishedAt, steps, error) {
4986
6581
  const executorStatus = status === "succeeded" ? "succeeded" : status === "timed_out" ? "timed_out" : "failed";
4987
6582
  return {
4988
6583
  status: executorStatus,
4989
6584
  exitCode: executorStatus === "succeeded" ? 0 : 1,
4990
- stdout,
6585
+ stdout: workflowRunEnvelope(workflowRun, steps),
4991
6586
  stderr: "",
4992
6587
  error,
4993
6588
  startedAt,
@@ -4997,20 +6592,16 @@ function workflowResult(workflowRun, status, startedAt, finishedAt, stdout, erro
4997
6592
  }
4998
6593
  async function executeWorkflow(store, workflow, opts = {}) {
4999
6594
  if (workflow.goal) {
6595
+ const goalSpec = workflow.goal;
5000
6596
  const workflowWithoutGoal = { ...workflow, goal: undefined };
5001
- return runGoal(store, workflow.goal, {
6597
+ return runGoal(store, goalSpec, {
5002
6598
  ...opts,
5003
6599
  model: opts.goalModel,
5004
- context: {
5005
- loopId: opts.loop?.id,
5006
- loopName: opts.loop?.name,
5007
- loopRunId: opts.loopRun?.id,
5008
- scheduledFor: opts.loopRun?.scheduledFor ?? opts.scheduledFor,
5009
- workflowId: workflow.id,
5010
- workflowName: workflow.name
5011
- },
6600
+ context: goalExecutionContext({ loop: opts.loop, loopRun: opts.loopRun, scheduledFor: opts.scheduledFor, workflow }),
5012
6601
  executeNode: async (node) => executeWorkflow(store, workflowWithoutGoal, {
5013
6602
  ...opts,
6603
+ env: withGoalNodeEnv(opts.env, node),
6604
+ goalNodePrompt: iterationPrompt(goalSpec, node),
5014
6605
  idempotencyKey: `${opts.idempotencyKey ?? workflow.id}:goal:${node.key}`
5015
6606
  })
5016
6607
  });
@@ -5025,8 +6616,7 @@ async function executeWorkflow(store, workflow, opts = {}) {
5025
6616
  });
5026
6617
  const startedAt = run.startedAt ?? nowIso();
5027
6618
  if (run.status === "succeeded" || run.status === "failed" || run.status === "timed_out" || run.status === "cancelled") {
5028
- const steps2 = store.listWorkflowStepRuns(run.id);
5029
- return workflowResult(run, run.status, startedAt, run.finishedAt ?? nowIso(), JSON.stringify({ workflowRun: run, steps: steps2 }, null, 2), run.error);
6619
+ return workflowResult(run, run.status, startedAt, run.finishedAt ?? nowIso(), store.listWorkflowStepRuns(run.id), run.error);
5030
6620
  }
5031
6621
  const ordered = workflowExecutionOrder(workflow);
5032
6622
  const byId = new Map(workflow.steps.map((step) => [step.id, step]));
@@ -5056,6 +6646,12 @@ async function executeWorkflow(store, workflow, opts = {}) {
5056
6646
  });
5057
6647
  if (blockedBy) {
5058
6648
  opts.beforePersist?.();
6649
+ if (isBlockedStepRun(store.getWorkflowStepRun(run.id, blockedBy))) {
6650
+ store.skipWorkflowStepRun(run.id, step.id, `${BLOCKED_STEP_ERROR_PREFIX} upstream step ${blockedBy} was blocked`, {
6651
+ daemonLeaseId: opts.daemonLeaseId
6652
+ });
6653
+ continue;
6654
+ }
5059
6655
  store.skipWorkflowStepRun(run.id, step.id, `dependency did not succeed: ${blockedBy}`, { daemonLeaseId: opts.daemonLeaseId });
5060
6656
  blockingError ??= `step ${step.id} blocked by dependency ${blockedBy}`;
5061
6657
  terminalStatus = "failed";
@@ -5068,16 +6664,14 @@ async function executeWorkflow(store, workflow, opts = {}) {
5068
6664
  blockingError = `step ${step.id} could not start because workflow is no longer running`;
5069
6665
  break;
5070
6666
  }
5071
- const metadata = {
5072
- loopId: opts.loop?.id,
5073
- loopName: opts.loop?.name,
5074
- runId: opts.loopRun?.id,
5075
- scheduledFor: opts.loopRun?.scheduledFor ?? opts.scheduledFor,
5076
- workflowId: workflow.id,
5077
- workflowName: workflow.name,
6667
+ const stepContext = goalExecutionContext({
6668
+ loop: opts.loop,
6669
+ loopRun: opts.loopRun,
6670
+ scheduledFor: opts.scheduledFor,
6671
+ workflow,
5078
6672
  workflowRunId: run.id,
5079
6673
  workflowStepId: step.id
5080
- };
6674
+ });
5081
6675
  let result;
5082
6676
  const controller = new AbortController;
5083
6677
  const externalAbort = () => controller.abort();
@@ -5096,19 +6690,10 @@ async function executeWorkflow(store, workflow, opts = {}) {
5096
6690
  model: opts.goalModel,
5097
6691
  target: targetWithStepAccount(step),
5098
6692
  signal: controller.signal,
5099
- context: {
5100
- loopId: opts.loop?.id,
5101
- loopName: opts.loop?.name,
5102
- loopRunId: opts.loopRun?.id,
5103
- scheduledFor: opts.loopRun?.scheduledFor ?? opts.scheduledFor,
5104
- workflowId: workflow.id,
5105
- workflowName: workflow.name,
5106
- workflowRunId: run.id,
5107
- workflowStepId: step.id
5108
- }
6693
+ context: stepContext
5109
6694
  });
5110
6695
  } else {
5111
- result = await executeTarget(targetWithStepAccount(step), metadata, {
6696
+ result = await executeTarget(targetWithStepAccount(step, opts.goalNodePrompt), executionMetadata(stepContext), {
5112
6697
  ...opts,
5113
6698
  machine: opts.machine ?? opts.loop?.machine,
5114
6699
  signal: controller.signal,
@@ -5144,6 +6729,21 @@ async function executeWorkflow(store, workflow, opts = {}) {
5144
6729
  break;
5145
6730
  }
5146
6731
  opts.beforePersist?.();
6732
+ const blockedExit = result.status === "failed" && result.exitCode !== undefined && blockedExitCodesForStep(step).includes(result.exitCode);
6733
+ if (blockedExit) {
6734
+ store.finalizeWorkflowStepRun(run.id, step.id, {
6735
+ status: "skipped",
6736
+ finishedAt: result.finishedAt,
6737
+ durationMs: result.durationMs,
6738
+ stdout: result.stdout,
6739
+ stderr: result.stderr,
6740
+ exitCode: result.exitCode,
6741
+ error: `${BLOCKED_STEP_ERROR_PREFIX} step ${step.id} exited with blocked exit code ${result.exitCode}`
6742
+ }, {
6743
+ daemonLeaseId: opts.daemonLeaseId
6744
+ });
6745
+ continue;
6746
+ }
5147
6747
  store.finalizeWorkflowStepRun(run.id, step.id, {
5148
6748
  status: result.status,
5149
6749
  finishedAt: result.finishedAt,
@@ -5174,8 +6774,7 @@ async function executeWorkflow(store, workflow, opts = {}) {
5174
6774
  const finishedAt = nowIso();
5175
6775
  if (store.isWorkflowRunTerminal(run.id)) {
5176
6776
  const terminalRun = store.requireWorkflowRun(run.id);
5177
- const steps2 = store.listWorkflowStepRuns(run.id);
5178
- return workflowResult(terminalRun, terminalRun.status, startedAt, terminalRun.finishedAt ?? finishedAt, JSON.stringify({ workflowRun: terminalRun, steps: steps2 }, null, 2), terminalRun.error ?? blockingError);
6777
+ return workflowResult(terminalRun, terminalRun.status, startedAt, terminalRun.finishedAt ?? finishedAt, store.listWorkflowStepRuns(run.id), terminalRun.error ?? blockingError);
5179
6778
  }
5180
6779
  opts.beforePersist?.();
5181
6780
  const finalRun = store.finalizeWorkflowRun(run.id, terminalStatus, {
@@ -5185,8 +6784,7 @@ async function executeWorkflow(store, workflow, opts = {}) {
5185
6784
  }, {
5186
6785
  daemonLeaseId: opts.daemonLeaseId
5187
6786
  });
5188
- const steps = store.listWorkflowStepRuns(run.id);
5189
- return workflowResult(finalRun, terminalStatus, startedAt, finishedAt, JSON.stringify({ workflowRun: finalRun, steps }, null, 2), blockingError);
6787
+ return workflowResult(finalRun, terminalStatus, startedAt, finishedAt, store.listWorkflowStepRuns(run.id), blockingError);
5190
6788
  }
5191
6789
  function preflightWorkflow(workflow, opts = {}) {
5192
6790
  return workflowExecutionOrder(workflow).map((step) => {
@@ -5237,12 +6835,7 @@ async function executeLoopTarget(store, loop, run, opts = {}) {
5237
6835
  ...opts,
5238
6836
  model: opts.goalModel,
5239
6837
  target: loop.target,
5240
- context: {
5241
- loopId: loop.id,
5242
- loopName: loop.name,
5243
- loopRunId: run.id,
5244
- scheduledFor: run.scheduledFor
5245
- }
6838
+ context: goalExecutionContext({ loop, loopRun: run })
5246
6839
  });
5247
6840
  }
5248
6841
  return executeLoop(loop, run, opts);
@@ -5257,23 +6850,19 @@ async function executeLoopTarget(store, loop, run, opts = {}) {
5257
6850
  }
5258
6851
  }
5259
6852
  if (loop.goal) {
6853
+ const loopGoal = loop.goal;
5260
6854
  const workflowForLoopGoal = workflow.goal ? { ...workflow, goal: undefined } : workflow;
5261
- return runGoal(store, loop.goal, {
6855
+ return runGoal(store, loopGoal, {
5262
6856
  ...opts,
5263
6857
  model: opts.goalModel,
5264
- context: {
5265
- loopId: loop.id,
5266
- loopName: loop.name,
5267
- loopRunId: run.id,
5268
- scheduledFor: run.scheduledFor,
5269
- workflowId: workflow.id,
5270
- workflowName: workflow.name
5271
- },
6858
+ context: goalExecutionContext({ loop, loopRun: run, workflow }),
5272
6859
  executeNode: async (node) => executeWorkflow(store, workflowForLoopGoal, {
5273
6860
  ...opts,
5274
6861
  loop,
5275
6862
  loopRun: run,
5276
6863
  scheduledFor: run.scheduledFor,
6864
+ env: withGoalNodeEnv(opts.env, node),
6865
+ goalNodePrompt: iterationPrompt(loopGoal, node),
5277
6866
  idempotencyKey: `${loop.id}:${run.scheduledFor}:attempt:${run.attempt}:goal:${node.key}`
5278
6867
  })
5279
6868
  });
@@ -5338,12 +6927,129 @@ function manualRunSource(loop, scheduledFor, now = new Date) {
5338
6927
  return "retry_slot";
5339
6928
  return "due_slot";
5340
6929
  }
5341
- function nextAfterRetry(loop, now) {
5342
- return new Date(now.getTime() + loop.retryDelayMs).toISOString();
6930
+ var INLINE_RUNNER_ID_PATTERN = /^[^:\s]+:(\d+)$/;
6931
+ function inlineRunnerOwnerPid(claimedBy) {
6932
+ const match = claimedBy ? INLINE_RUNNER_ID_PATTERN.exec(claimedBy) : null;
6933
+ return match ? Number(match[1]) : undefined;
6934
+ }
6935
+ async function runLoopNow(deps) {
6936
+ const { store, runnerId } = deps;
6937
+ const loop = store.requireLoop(deps.idOrName);
6938
+ if (loop.archivedAt)
6939
+ throw new LoopArchivedError(loop.name || deps.idOrName);
6940
+ const now = deps.now?.() ?? new Date;
6941
+ if (deps.mode === "schedule") {
6942
+ const scheduledFor2 = now.toISOString();
6943
+ const updated = store.updateLoop(loop.id, { status: "active", nextRunAt: scheduledFor2 });
6944
+ return { mode: "schedule", loop: updated, scheduledFor: scheduledFor2 };
6945
+ }
6946
+ let scheduledFor = manualRunScheduledFor(loop, now);
6947
+ let source = manualRunSource(loop, scheduledFor, now);
6948
+ let shouldAdvance = shouldAdvanceManualRun(loop, scheduledFor, now);
6949
+ let claim = store.claimRun(loop, scheduledFor, runnerId, now);
6950
+ if (!claim && shouldAdvance) {
6951
+ const existing = store.getRunBySlot(loop.id, scheduledFor);
6952
+ if (existing && existing.status !== "running") {
6953
+ scheduledFor = now.toISOString();
6954
+ source = "ad_hoc";
6955
+ shouldAdvance = false;
6956
+ claim = store.claimRun(loop, scheduledFor, runnerId, now);
6957
+ }
6958
+ }
6959
+ if (!claim)
6960
+ throw new Error(`could not claim manual run for ${deps.idOrName}`);
6961
+ const run = await executeClaimedRun({
6962
+ store,
6963
+ runnerId,
6964
+ loop: claim.loop,
6965
+ run: claim.run,
6966
+ now: deps.now,
6967
+ execute: deps.execute
6968
+ });
6969
+ if (shouldAdvance) {
6970
+ advanceLoop(store, claim.loop, run, new Date(run.finishedAt ?? new Date), run.status === "succeeded");
6971
+ }
6972
+ return { mode: "inline", loop: claim.loop, run, source, advancedLoop: shouldAdvance };
6973
+ }
6974
+ var MAX_RETRY_DELAY_MS = 6 * 60 * 60 * 1000;
6975
+ var DEFAULT_CIRCUIT_BREAKER_THRESHOLD = 5;
6976
+ var CIRCUIT_BREAKER_REASON_PREFIX = "circuit breaker open";
6977
+ var MAX_SKIPS_PER_LOOP_PER_TICK = 10;
6978
+ var THROTTLED_RETRY_MULTIPLIER = 4;
6979
+ var MAX_RETRY_EXPONENT = 20;
6980
+ function retryBackoffDelayMs(loop, run, random = Math.random) {
6981
+ const attempt = Math.max(1, run.attempt);
6982
+ const failure = classifyRunFailure(run);
6983
+ const throttled = failure?.classification === "rate_limit" || failure?.classification === "auth";
6984
+ const growth = 2 ** Math.min(attempt - 1, MAX_RETRY_EXPONENT);
6985
+ const base = loop.retryDelayMs * growth * (throttled ? THROTTLED_RETRY_MULTIPLIER : 1);
6986
+ const jitter = 0.5 + random();
6987
+ return Math.min(MAX_RETRY_DELAY_MS, Math.round(base * jitter));
6988
+ }
6989
+ function nextAfterRetry(loop, run, now, random) {
6990
+ return new Date(now.getTime() + retryBackoffDelayMs(loop, run, random)).toISOString();
5343
6991
  }
5344
6992
  function isDaemonLeaseLost(error) {
5345
6993
  return error instanceof Error && error.message === "daemon lease lost";
5346
6994
  }
6995
+ function resolveBreakerThreshold(loop, override) {
6996
+ const perLoop = loop.circuitBreakerThreshold;
6997
+ if (typeof perLoop === "number" && Number.isFinite(perLoop))
6998
+ return Math.floor(perLoop);
6999
+ const resolved = typeof override === "function" ? override(loop) : override;
7000
+ if (typeof resolved === "number" && Number.isFinite(resolved))
7001
+ return Math.floor(resolved);
7002
+ return DEFAULT_CIRCUIT_BREAKER_THRESHOLD;
7003
+ }
7004
+ function consecutiveFailureCount(store, loopId, maxAttempts = 1, scanLimit = 50) {
7005
+ const runs = store.listRuns({ loopId, limit: scanLimit });
7006
+ let watermark;
7007
+ for (const run of runs) {
7008
+ if (run.status !== "skipped" || !run.error?.startsWith(CIRCUIT_BREAKER_REASON_PREFIX))
7009
+ continue;
7010
+ const at = new Date(run.scheduledFor).getTime();
7011
+ if (watermark === undefined || at > watermark)
7012
+ watermark = at;
7013
+ }
7014
+ let count = 0;
7015
+ for (const run of runs) {
7016
+ if (run.status === "running" || run.status === "skipped")
7017
+ continue;
7018
+ if (watermark !== undefined && new Date(run.scheduledFor).getTime() <= watermark)
7019
+ continue;
7020
+ if (run.status === "succeeded")
7021
+ break;
7022
+ if (run.attempt < maxAttempts)
7023
+ continue;
7024
+ count += 1;
7025
+ }
7026
+ return count;
7027
+ }
7028
+ function awaitStrictlyNewerRunTimestamp(store, loopId) {
7029
+ const latest = store.listRuns({ loopId, limit: 1 })[0];
7030
+ if (!latest)
7031
+ return;
7032
+ const latestMs = new Date(latest.createdAt).getTime();
7033
+ for (let spin = 0;spin < 1e6 && Date.now() <= latestMs; spin += 1) {}
7034
+ }
7035
+ function tripCircuitBreaker(store, loop, run, finishedAt, failures, opts) {
7036
+ awaitStrictlyNewerRunTimestamp(store, loop.id);
7037
+ const reason = `${CIRCUIT_BREAKER_REASON_PREFIX}: ${failures} consecutive failed runs; loop auto-paused (resume with 'loops resume ${loop.name}')`;
7038
+ let markerAtMs = finishedAt.getTime();
7039
+ for (let probe = 0;probe < 1000 && store.getRunBySlot(loop.id, new Date(markerAtMs).toISOString()); probe += 1) {
7040
+ markerAtMs += 1;
7041
+ }
7042
+ const marker = store.createSkippedRun(loop, new Date(markerAtMs).toISOString(), reason, {
7043
+ daemonLeaseId: opts.daemonLeaseId
7044
+ });
7045
+ const nextRunAt = computeNextAfter(loop.schedule, new Date(run.scheduledFor), finishedAt);
7046
+ store.updateLoop(loop.id, {
7047
+ status: "paused",
7048
+ nextRunAt,
7049
+ retryScheduledFor: undefined
7050
+ }, { daemonLeaseId: opts.daemonLeaseId });
7051
+ opts.onRun?.(marker);
7052
+ }
5347
7053
  function advanceLoop(store, loop, run, finishedAt, succeeded, opts = {}) {
5348
7054
  if (run.status === "running")
5349
7055
  return;
@@ -5356,7 +7062,7 @@ function advanceLoop(store, loop, run, finishedAt, succeeded, opts = {}) {
5356
7062
  if (shouldRetry) {
5357
7063
  store.updateLoop(current.id, {
5358
7064
  status: "active",
5359
- nextRunAt: nextAfterRetry(current, finishedAt),
7065
+ nextRunAt: nextAfterRetry(current, run, finishedAt, opts.random),
5360
7066
  retryScheduledFor: run.scheduledFor
5361
7067
  }, { daemonLeaseId: opts.daemonLeaseId });
5362
7068
  return;
@@ -5365,11 +7071,21 @@ function advanceLoop(store, loop, run, finishedAt, succeeded, opts = {}) {
5365
7071
  if (deferredRetry) {
5366
7072
  store.updateLoop(current.id, {
5367
7073
  status: "active",
5368
- nextRunAt: nextAfterRetry(current, finishedAt),
7074
+ nextRunAt: nextAfterRetry(current, deferredRetry, finishedAt, opts.random),
5369
7075
  retryScheduledFor: deferredRetry.scheduledFor
5370
7076
  }, { daemonLeaseId: opts.daemonLeaseId });
5371
7077
  return;
5372
7078
  }
7079
+ if (!succeeded) {
7080
+ const threshold = resolveBreakerThreshold(current, opts.circuitBreakerThreshold);
7081
+ if (threshold > 0) {
7082
+ const failures = consecutiveFailureCount(store, current.id, current.maxAttempts, Math.max(threshold * 4, 50));
7083
+ if (failures >= threshold) {
7084
+ tripCircuitBreaker(store, current, run, finishedAt, failures, opts);
7085
+ return;
7086
+ }
7087
+ }
7088
+ }
5373
7089
  const nextRunAt = computeNextAfter(current.schedule, new Date(run.scheduledFor), finishedAt);
5374
7090
  store.updateLoop(current.id, {
5375
7091
  status: nextRunAt ? "active" : "stopped",
@@ -5431,6 +7147,14 @@ async function executeClaimedRun(deps) {
5431
7147
  clearInterval(heartbeat);
5432
7148
  }
5433
7149
  }
7150
+ function advanceOptions(deps) {
7151
+ return {
7152
+ daemonLeaseId: deps.daemonLeaseId,
7153
+ random: deps.random,
7154
+ circuitBreakerThreshold: deps.circuitBreakerThreshold,
7155
+ onRun: deps.onRun
7156
+ };
7157
+ }
5434
7158
  async function runSlot(deps, loop, scheduledFor) {
5435
7159
  const now = deps.now?.() ?? new Date;
5436
7160
  deps.beforeRun?.(loop, scheduledFor);
@@ -5445,7 +7169,7 @@ async function runSlot(deps, loop, scheduledFor) {
5445
7169
  return;
5446
7170
  throw error;
5447
7171
  }
5448
- advanceLoop(deps.store, loop, skipped, now, true, { daemonLeaseId: deps.daemonLeaseId });
7172
+ advanceLoop(deps.store, loop, skipped, now, true, advanceOptions(deps));
5449
7173
  deps.onRun?.(skipped);
5450
7174
  return skipped;
5451
7175
  }
@@ -5472,7 +7196,7 @@ async function runSlot(deps, loop, scheduledFor) {
5472
7196
  daemonLeaseId: deps.daemonLeaseId,
5473
7197
  onError: deps.onError
5474
7198
  });
5475
- advanceLoop(deps.store, claim.loop, finalRun, new Date(finalRun.finishedAt ?? new Date), finalRun.status === "succeeded", { daemonLeaseId: deps.daemonLeaseId });
7199
+ advanceLoop(deps.store, claim.loop, finalRun, new Date(finalRun.finishedAt ?? new Date), finalRun.status === "succeeded", advanceOptions(deps));
5476
7200
  deps.onRun?.(finalRun);
5477
7201
  return finalRun;
5478
7202
  }
@@ -5492,7 +7216,7 @@ function claimSlot(deps, loop, scheduledFor) {
5492
7216
  return;
5493
7217
  throw error;
5494
7218
  }
5495
- advanceLoop(deps.store, loop, skipped, now, true, { daemonLeaseId: deps.daemonLeaseId });
7219
+ advanceLoop(deps.store, loop, skipped, now, true, advanceOptions(deps));
5496
7220
  deps.onRun?.(skipped);
5497
7221
  return skipped;
5498
7222
  }
@@ -5510,8 +7234,7 @@ function claimSlot(deps, loop, scheduledFor) {
5510
7234
  deps.onRun?.(claim.run);
5511
7235
  return claim;
5512
7236
  }
5513
- function claimDueRuns(deps) {
5514
- const now = deps.now?.() ?? new Date;
7237
+ function recoverAndExpire(deps, now) {
5515
7238
  const recovered = deps.store.recoverExpiredRunLeases(now, { daemonLeaseId: deps.daemonLeaseId });
5516
7239
  const recoveredByLoop = new Map;
5517
7240
  for (const run of recovered) {
@@ -5523,21 +7246,22 @@ function claimDueRuns(deps) {
5523
7246
  continue;
5524
7247
  const retryable = runs.filter((run) => run.attempt < loop.maxAttempts).sort((a, b) => new Date(a.scheduledFor).getTime() - new Date(b.scheduledFor).getTime())[0];
5525
7248
  if (retryable) {
5526
- advanceLoop(deps.store, loop, retryable, new Date(retryable.finishedAt ?? now), false, {
5527
- daemonLeaseId: deps.daemonLeaseId
5528
- });
7249
+ advanceLoop(deps.store, loop, retryable, new Date(retryable.finishedAt ?? now), false, advanceOptions(deps));
5529
7250
  continue;
5530
7251
  }
5531
7252
  for (const run of runs) {
5532
7253
  const current = deps.store.getLoop(run.loopId);
5533
7254
  if (current) {
5534
- advanceLoop(deps.store, current, run, new Date(run.finishedAt ?? now), false, {
5535
- daemonLeaseId: deps.daemonLeaseId
5536
- });
7255
+ advanceLoop(deps.store, current, run, new Date(run.finishedAt ?? now), false, advanceOptions(deps));
5537
7256
  }
5538
7257
  }
5539
7258
  }
5540
7259
  const expired = deps.store.expireLoops(now, { daemonLeaseId: deps.daemonLeaseId });
7260
+ return { recovered, expired };
7261
+ }
7262
+ function claimDueRuns(deps) {
7263
+ const now = deps.now?.() ?? new Date;
7264
+ const { recovered, expired } = recoverAndExpire(deps, now);
5541
7265
  const claims = [];
5542
7266
  const claimed = [];
5543
7267
  const skipped = [];
@@ -5545,11 +7269,14 @@ function claimDueRuns(deps) {
5545
7269
  if (maxClaims === 0)
5546
7270
  return { claims, claimed, completed: [], skipped, recovered, expired };
5547
7271
  for (const loop of deps.store.dueLoops(now)) {
5548
- if (claims.length + skipped.length >= maxClaims)
7272
+ if (claims.length >= maxClaims)
5549
7273
  break;
5550
7274
  const plan = dueSlots(loop, now);
7275
+ let loopSkips = 0;
5551
7276
  for (const slot of plan.slots) {
5552
- if (claims.length + skipped.length >= maxClaims)
7277
+ if (claims.length >= maxClaims)
7278
+ break;
7279
+ if (loopSkips >= MAX_SKIPS_PER_LOOP_PER_TICK)
5553
7280
  break;
5554
7281
  const run = claimSlot(deps, loop, slot);
5555
7282
  if (!run)
@@ -5559,6 +7286,7 @@ function claimDueRuns(deps) {
5559
7286
  claimed.push(run.run);
5560
7287
  } else if (run.status === "skipped") {
5561
7288
  skipped.push(run);
7289
+ loopSkips += 1;
5562
7290
  }
5563
7291
  }
5564
7292
  }
@@ -5566,46 +7294,25 @@ function claimDueRuns(deps) {
5566
7294
  }
5567
7295
  async function tick(deps) {
5568
7296
  const now = deps.now?.() ?? new Date;
5569
- const recovered = deps.store.recoverExpiredRunLeases(now, { daemonLeaseId: deps.daemonLeaseId });
5570
- const recoveredByLoop = new Map;
5571
- for (const run of recovered) {
5572
- recoveredByLoop.set(run.loopId, [...recoveredByLoop.get(run.loopId) ?? [], run]);
5573
- }
5574
- for (const runs of recoveredByLoop.values()) {
5575
- const loop = deps.store.getLoop(runs[0].loopId);
5576
- if (!loop)
5577
- continue;
5578
- const retryable = runs.filter((run) => run.attempt < loop.maxAttempts).sort((a, b) => new Date(a.scheduledFor).getTime() - new Date(b.scheduledFor).getTime())[0];
5579
- if (retryable) {
5580
- advanceLoop(deps.store, loop, retryable, new Date(retryable.finishedAt ?? now), false, {
5581
- daemonLeaseId: deps.daemonLeaseId
5582
- });
5583
- continue;
5584
- }
5585
- for (const run of runs) {
5586
- const current = deps.store.getLoop(run.loopId);
5587
- if (current) {
5588
- advanceLoop(deps.store, current, run, new Date(run.finishedAt ?? now), false, {
5589
- daemonLeaseId: deps.daemonLeaseId
5590
- });
5591
- }
5592
- }
5593
- }
5594
- const expired = deps.store.expireLoops(now, { daemonLeaseId: deps.daemonLeaseId });
7297
+ const { recovered, expired } = recoverAndExpire(deps, now);
5595
7298
  const claimed = [];
5596
7299
  const completed = [];
5597
7300
  const skipped = [];
5598
7301
  for (const loop of deps.store.dueLoops(now)) {
5599
7302
  const plan = dueSlots(loop, now);
7303
+ let loopSkips = 0;
5600
7304
  for (const slot of plan.slots) {
7305
+ if (loopSkips >= MAX_SKIPS_PER_LOOP_PER_TICK)
7306
+ break;
5601
7307
  const run = await runSlot(deps, loop, slot);
5602
7308
  if (!run)
5603
7309
  continue;
5604
7310
  if (run.status === "running")
5605
7311
  claimed.push(run);
5606
- else if (run.status === "skipped")
7312
+ else if (run.status === "skipped") {
5607
7313
  skipped.push(run);
5608
- else
7314
+ loopSkips += 1;
7315
+ } else
5609
7316
  completed.push(run);
5610
7317
  if (["failed", "timed_out", "abandoned"].includes(run.status) && run.attempt < loop.maxAttempts)
5611
7318
  break;
@@ -5627,29 +7334,25 @@ class LoopsClient {
5627
7334
  create(input) {
5628
7335
  return this.store.createLoop(input);
5629
7336
  }
5630
- list() {
5631
- return this.store.listLoops();
7337
+ list(filters = {}) {
7338
+ return this.store.listLoops({
7339
+ status: filters.status,
7340
+ limit: filters.limit,
7341
+ includeArchived: filters.includeArchived,
7342
+ archived: filters.archivedOnly
7343
+ });
5632
7344
  }
5633
7345
  get(idOrName) {
5634
7346
  return this.store.requireLoop(idOrName);
5635
7347
  }
5636
7348
  pause(idOrName) {
5637
- const loop = this.get(idOrName);
5638
- if (loop.archivedAt)
5639
- throw new Error(`loop is archived; unarchive it before pausing: ${idOrName}`);
5640
- return this.store.updateLoop(loop.id, { status: "paused" });
7349
+ return this.store.updateLoop(this.get(idOrName).id, { status: "paused" });
5641
7350
  }
5642
7351
  resume(idOrName) {
5643
- const loop = this.get(idOrName);
5644
- if (loop.archivedAt)
5645
- throw new Error(`loop is archived; unarchive it before resuming: ${idOrName}`);
5646
- return this.store.updateLoop(loop.id, { status: "active" });
7352
+ return this.store.updateLoop(this.get(idOrName).id, { status: "active" });
5647
7353
  }
5648
7354
  stop(idOrName) {
5649
- const loop = this.get(idOrName);
5650
- if (loop.archivedAt)
5651
- throw new Error(`loop is archived; unarchive it before stopping: ${idOrName}`);
5652
- return this.store.updateLoop(loop.id, { status: "stopped", nextRunAt: undefined });
7355
+ return this.store.updateLoop(this.get(idOrName).id, { status: "stopped", nextRunAt: undefined });
5653
7356
  }
5654
7357
  archive(idOrName) {
5655
7358
  return this.store.archiveLoop(idOrName);
@@ -5660,8 +7363,24 @@ class LoopsClient {
5660
7363
  delete(idOrName) {
5661
7364
  return this.store.deleteLoop(idOrName);
5662
7365
  }
5663
- runs(loopId) {
5664
- return this.store.listRuns({ loopId });
7366
+ runs(idOrName, filters = {}) {
7367
+ let loopId;
7368
+ if (idOrName) {
7369
+ try {
7370
+ loopId = this.get(idOrName).id;
7371
+ } catch (error) {
7372
+ if (error instanceof LoopNotFoundError)
7373
+ return [];
7374
+ throw error;
7375
+ }
7376
+ }
7377
+ return this.store.listRuns({ loopId, status: filters.status, limit: filters.limit });
7378
+ }
7379
+ doctor() {
7380
+ return runDoctor(this.store);
7381
+ }
7382
+ health(opts = {}) {
7383
+ return buildHealthReport(this.store, opts);
5665
7384
  }
5666
7385
  goal(idOrName) {
5667
7386
  const goal = this.store.getGoal(idOrName) ?? this.store.findGoalByLoop(idOrName) ?? this.store.findGoalByRunId(idOrName);
@@ -5674,28 +7393,8 @@ class LoopsClient {
5674
7393
  return tick({ store: this.store, runnerId: this.runnerId });
5675
7394
  }
5676
7395
  async runNow(idOrName) {
5677
- const loop = this.get(idOrName);
5678
- if (loop.archivedAt)
5679
- throw new Error(`loop is archived; unarchive it before running: ${idOrName}`);
5680
- const now = new Date;
5681
- let scheduledFor = manualRunScheduledFor(loop, now);
5682
- let shouldAdvance = shouldAdvanceManualRun(loop, scheduledFor, now);
5683
- let claim = this.store.claimRun(loop, scheduledFor, this.runnerId, now);
5684
- if (!claim && shouldAdvance) {
5685
- const existing = this.store.getRunBySlot(loop.id, scheduledFor);
5686
- if (existing && existing.status !== "running") {
5687
- scheduledFor = now.toISOString();
5688
- shouldAdvance = false;
5689
- claim = this.store.claimRun(loop, scheduledFor, this.runnerId, now);
5690
- }
5691
- }
5692
- if (!claim)
5693
- throw new Error(`could not claim manual run for ${idOrName}`);
5694
- const run = await executeClaimedRun({ store: this.store, runnerId: this.runnerId, loop: claim.loop, run: claim.run });
5695
- if (shouldAdvance) {
5696
- advanceLoop(this.store, claim.loop, run, new Date(run.finishedAt ?? new Date), run.status === "succeeded");
5697
- }
5698
- return run;
7396
+ const result = await runLoopNow({ store: this.store, idOrName, runnerId: this.runnerId });
7397
+ return result.run;
5699
7398
  }
5700
7399
  close() {
5701
7400
  if (this.ownStore)