@tea-agent/loop-agent 0.2.1 → 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 (135) hide show
  1. package/AGENTS.md +91 -87
  2. package/CHANGELOG.md +89 -52
  3. package/README.md +195 -180
  4. package/bin/agent-worker.js +22 -0
  5. package/bin/loop-agent.js +21 -21
  6. package/dist/application/dag/args.js +6 -0
  7. package/dist/application/dag/generate-task-dag.js +2 -0
  8. package/dist/application/dag/run-dag.js +3 -0
  9. package/dist/application/dag/validate-dag.js +40 -0
  10. package/dist/cli/command-definitions.js +2 -2
  11. package/dist/cli/program.js +24 -4
  12. package/dist/commands/init.js +1011 -459
  13. package/dist/commands/loop-benchmark.js +11 -11
  14. package/dist/commands/pi-reuse-benchmark.js +16 -16
  15. package/dist/executors/cursor-executor.js +1 -1
  16. package/dist/executors/dag-pi-executor.js +8 -1
  17. package/dist/task/runtime.js +27 -27
  18. package/dist/worker/cli.js +119 -0
  19. package/dist/worker/loop-agent/command-result.js +1 -0
  20. package/dist/worker/loop-agent/loop-agent-client.js +105 -0
  21. package/dist/worker/loop-agent/parse-json.js +14 -0
  22. package/dist/worker/materialize/harness-task-materializer.js +157 -0
  23. package/dist/worker/pool/failure-routing.js +98 -0
  24. package/dist/worker/pool/run-store.js +117 -0
  25. package/dist/worker/pool/types.js +1 -0
  26. package/dist/worker/preflight.js +108 -0
  27. package/dist/worker/profile-mapping.js +76 -0
  28. package/dist/worker/progress-reporter.js +81 -0
  29. package/dist/worker/report/morning-report.js +69 -0
  30. package/dist/worker/repos/repo-resolver.js +23 -0
  31. package/dist/worker/run-task/run-task.js +359 -0
  32. package/dist/worker/runner/run-ready.js +216 -0
  33. package/dist/worker/task-graph/acceptance-schema.js +25 -0
  34. package/dist/worker/task-graph/ready-queue.js +23 -0
  35. package/dist/worker/task-graph/task-graph-schema.js +28 -0
  36. package/dist/worker/task-graph/types.js +1 -0
  37. package/dist/worker/task-graph/validate.js +188 -0
  38. package/dist/worker/task-spec/complexity-mapping.js +8 -0
  39. package/dist/worker/task-spec/schema.js +116 -0
  40. package/dist/worker/task-spec/types.js +1 -0
  41. package/dist/worker/task-spec/validate.js +352 -0
  42. package/dist/workflows/dag/canvas-observer.js +275 -275
  43. package/dist/workflows/dag/dynamic-runtime/loop-until.js +2 -1
  44. package/dist/workflows/dag/dynamic-runtime/map.js +1 -0
  45. package/dist/workflows/dag/init-hybrid.js +3 -3
  46. package/dist/workflows/dag/skills.js +3 -3
  47. package/dist/workflows/dag/types.js +2 -0
  48. package/dist/workflows/dynamic/compile.js +11 -0
  49. package/dist/workflows/dynamic/spec.js +1 -0
  50. package/docs/README.md +72 -65
  51. package/docs/agent-dag-recovery-playbook.md +184 -184
  52. package/docs/agent-dag-runner.md +42 -40
  53. package/docs/architecture/runtime-boundaries.md +147 -147
  54. package/docs/cursor-executor-usage.md +25 -25
  55. package/docs/decisions/README.md +3 -3
  56. package/docs/design/README.md +36 -36
  57. package/docs/development-principles.md +73 -71
  58. package/docs/dynamic-workflow-dag-engine-roadmap.md +1749 -1749
  59. package/docs/exec-plans/README.md +6 -6
  60. package/docs/exec-plans/active/README.md +7 -10
  61. package/docs/exec-plans/completed/README.md +19 -9
  62. package/docs/feature-workflow.md +186 -186
  63. package/docs/harness-methodology-debugging.md +153 -153
  64. package/docs/harness-methodology-tdd.md +130 -130
  65. package/docs/harness-methodology-verification.md +27 -27
  66. package/docs/init-surface.manifest.json +175 -0
  67. package/docs/loop-agent-harness.md +42 -42
  68. package/docs/production-readiness.md +96 -96
  69. package/docs/progress/README.md +3 -3
  70. package/docs/reports/README.md +5 -5
  71. package/docs/skills/README.md +6 -0
  72. package/docs/skills/vetted-skill-registry.md +26 -0
  73. package/docs/templates/adr.md +60 -60
  74. package/docs/templates/agent-dag-authority-surface-audit.prompt.md +94 -94
  75. package/docs/templates/agent-dag-decision-envelope.schema.json +213 -213
  76. package/docs/templates/agent-dag-decision-gate-dogfood-report.md +117 -117
  77. package/docs/templates/agent-dag-decision-gate.prompt.md +246 -246
  78. package/docs/templates/agent-dag-process-supervisor.prompt.md +98 -98
  79. package/docs/templates/agent-dag-report.schema.json +454 -454
  80. package/docs/templates/agent-dag-review-verdict.prompt.md +68 -68
  81. package/docs/templates/agent-dag.base.json +195 -195
  82. package/docs/templates/agent-dag.final-verification.json +190 -190
  83. package/docs/templates/agent-dag.schema.json +316 -316
  84. package/docs/templates/agent-dag.supervised-implementation.json +500 -500
  85. package/docs/templates/exec-plan.md +64 -64
  86. package/docs/templates/feature-spec.md +53 -53
  87. package/docs/templates/hybrid-dag.json +193 -193
  88. package/docs/templates/init-evolution-review.md +33 -0
  89. package/docs/templates/production-readiness-checklist.md +57 -57
  90. package/docs/templates/progress-log.md +17 -17
  91. package/docs/templates/project-start-checklist.md +9 -9
  92. package/docs/templates/qa-report.md +48 -48
  93. package/docs/templates/sprint-contract.md +29 -29
  94. package/docs/verification-matrix.md +41 -41
  95. package/examples/decision-gate-agent-dag.json +123 -123
  96. package/examples/example-dag.json +51 -51
  97. package/examples/hybrid-loop-agent-dag.json +194 -194
  98. package/harness.json +94 -92
  99. package/package.json +66 -62
  100. package/skills/ai-engineering-context/SKILL.md +48 -48
  101. package/skills/code-review-core/SKILL.md +20 -0
  102. package/skills/codebase-scout/SKILL.md +19 -0
  103. package/skills/init-capability-evolution/SKILL.md +69 -0
  104. package/skills/loop-agent/SKILL.md +147 -145
  105. package/skills/loop-agent/references/README.md +67 -67
  106. package/skills/loop-agent/references/command-reference.md +403 -357
  107. package/skills/loop-agent/references/harness-policy.md +259 -258
  108. package/skills/loop-agent/references/hybrid-dag.md +216 -216
  109. package/skills/loop-agent/references/learned/README.md +21 -21
  110. package/skills/loop-agent/references/long-running-loop.md +59 -59
  111. package/skills/loop-agent/references/model-routing.md +36 -36
  112. package/skills/loop-agent/references/multi-worktree.md +54 -54
  113. package/skills/loop-agent/references/one-shot-runs.md +85 -85
  114. package/skills/loop-agent/references/orchestrator-and-interventions.md +169 -169
  115. package/skills/loop-agent/references/pi-prompt.md +23 -23
  116. package/skills/loop-agent/references/pi-subagent-assisted-mode.md +81 -81
  117. package/skills/loop-agent/references/post-implementation-and-patterns.md +44 -44
  118. package/skills/loop-agent/references/task-workflow.md +84 -84
  119. package/skills/loop-agent/references/verification-and-failure-handling.md +128 -128
  120. package/skills/requesting-code-review/SKILL.md +101 -101
  121. package/skills/requesting-code-review/code-reviewer.md +168 -168
  122. package/skills/systematic-debugging/CREATION-LOG.md +119 -119
  123. package/skills/systematic-debugging/SKILL.md +296 -296
  124. package/skills/systematic-debugging/condition-based-waiting-example.ts +158 -158
  125. package/skills/systematic-debugging/condition-based-waiting.md +115 -115
  126. package/skills/systematic-debugging/defense-in-depth.md +122 -122
  127. package/skills/systematic-debugging/find-polluter.sh +63 -63
  128. package/skills/systematic-debugging/root-cause-tracing.md +169 -169
  129. package/skills/systematic-debugging/test-academic.md +14 -14
  130. package/skills/systematic-debugging/test-pressure-1.md +58 -58
  131. package/skills/systematic-debugging/test-pressure-2.md +68 -68
  132. package/skills/systematic-debugging/test-pressure-3.md +69 -69
  133. package/skills/test-driven-development/SKILL.md +20 -0
  134. package/skills/verification-before-completion/SKILL.md +154 -154
  135. package/skills/webapp-testing/SKILL.md +19 -0
@@ -195,280 +195,280 @@ export function createDagCanvasObserver(opts) {
195
195
  flush: () => writer.flush(),
196
196
  };
197
197
  }
198
- const CANVAS_HEADER = `/* AUTO-GENERATED by . run-dag canvas observer. Derived view only — canonical state lives in .harness/dag-runs/. */
199
- import {
200
- Card,
201
- CardBody,
202
- CardHeader,
203
- Divider,
204
- Grid,
205
- H1,
206
- H2,
207
- Pill,
208
- Row,
209
- Stack,
210
- Stat,
211
- Text,
212
- computeDAGLayout,
213
- useHostTheme,
214
- } from 'cursor/canvas';
215
- import { useMemo } from 'react';
216
-
217
- type TaskStatus = 'PENDING' | 'RUNNING' | 'FINISHED' | 'ERROR' | 'SKIPPED';
218
- type Complexity = 'HIGH' | 'MED' | 'LOW';
219
- type Executor = 'cursor' | 'pi' | 'shell' | 'static';
220
-
221
- interface TaskState {
222
- id: string;
223
- depends_on: string[];
224
- complexity: Complexity;
225
- executor: Executor;
226
- subtask_prompt: string;
227
- status: TaskStatus;
228
- model: string;
229
- startedAt?: number;
230
- finishedAt?: number;
231
- resultText?: string;
232
- errorMessage?: string;
233
- inputTokens?: number;
234
- outputTokens?: number;
235
- durationMs?: number;
236
- skippedReason?: string;
237
- }
238
-
239
- interface RunState {
240
- title: string;
241
- runId: string;
242
- startedAt: number;
243
- finishedAt?: number;
244
- runOutcome?: 'SUCCESS' | 'FAILED' | 'INTERRUPTED' | 'PAUSED';
245
- runMessage?: string;
246
- tasks: TaskState[];
198
+ const CANVAS_HEADER = `/* AUTO-GENERATED by . run-dag canvas observer. Derived view only — canonical state lives in .harness/dag-runs/. */
199
+ import {
200
+ Card,
201
+ CardBody,
202
+ CardHeader,
203
+ Divider,
204
+ Grid,
205
+ H1,
206
+ H2,
207
+ Pill,
208
+ Row,
209
+ Stack,
210
+ Stat,
211
+ Text,
212
+ computeDAGLayout,
213
+ useHostTheme,
214
+ } from 'cursor/canvas';
215
+ import { useMemo } from 'react';
216
+
217
+ type TaskStatus = 'PENDING' | 'RUNNING' | 'FINISHED' | 'ERROR' | 'SKIPPED';
218
+ type Complexity = 'HIGH' | 'MED' | 'LOW';
219
+ type Executor = 'cursor' | 'pi' | 'shell' | 'static';
220
+
221
+ interface TaskState {
222
+ id: string;
223
+ depends_on: string[];
224
+ complexity: Complexity;
225
+ executor: Executor;
226
+ subtask_prompt: string;
227
+ status: TaskStatus;
228
+ model: string;
229
+ startedAt?: number;
230
+ finishedAt?: number;
231
+ resultText?: string;
232
+ errorMessage?: string;
233
+ inputTokens?: number;
234
+ outputTokens?: number;
235
+ durationMs?: number;
236
+ skippedReason?: string;
237
+ }
238
+
239
+ interface RunState {
240
+ title: string;
241
+ runId: string;
242
+ startedAt: number;
243
+ finishedAt?: number;
244
+ runOutcome?: 'SUCCESS' | 'FAILED' | 'INTERRUPTED' | 'PAUSED';
245
+ runMessage?: string;
246
+ tasks: TaskState[];
247
247
  }`;
248
- const CANVAS_BODY = String.raw `const NODE_W = 200;
249
- const NODE_H = 72;
250
-
251
- function pillToneFor(status: TaskStatus): 'neutral' | 'info' | 'success' | 'warning' {
252
- switch (status) {
253
- case 'PENDING':
254
- return 'neutral';
255
- case 'RUNNING':
256
- return 'info';
257
- case 'FINISHED':
258
- return 'success';
259
- case 'ERROR':
260
- return 'warning';
261
- case 'SKIPPED':
262
- return 'neutral';
263
- }
264
- }
265
-
266
- function formatDuration(ms?: number): string {
267
- if (ms === undefined) return '—';
268
- if (ms < 1000) return ms + 'ms';
269
- const s = ms / 1000;
270
- if (s < 60) return s.toFixed(1) + 's';
271
- const m = Math.floor(s / 60);
272
- const rem = Math.round(s - m * 60);
273
- return m + 'm ' + rem + 's';
274
- }
275
-
276
- function elapsed(state: RunState): number {
277
- const end = state.finishedAt ?? Date.now();
278
- return end - state.startedAt;
279
- }
280
-
281
- function DAGGraph({ state }: { state: RunState }): JSX.Element {
282
- const theme = useHostTheme();
283
- const layout = computeDAGLayout({
284
- nodes: state.tasks.map((t) => ({ id: t.id })),
285
- edges: state.tasks.flatMap((t) =>
286
- t.depends_on.map((d) => ({ from: d, to: t.id })),
287
- ),
288
- direction: 'vertical',
289
- nodeWidth: NODE_W,
290
- nodeHeight: NODE_H,
291
- rankGap: 72,
292
- nodeGap: 40,
293
- padding: 24,
294
- });
295
- const byId = new Map(state.tasks.map((t) => [t.id, t]));
296
-
297
- return (
298
- <svg width={layout.width} height={layout.height} style={{ display: 'block', maxWidth: '100%' }}>
299
- {layout.edges.map((e, i) => (
300
- <line
301
- key={'e' + i}
302
- x1={e.sourceX}
303
- y1={e.sourceY}
304
- x2={e.targetX}
305
- y2={e.targetY}
306
- stroke={theme.stroke.secondary}
307
- strokeWidth={1.25}
308
- markerEnd="url(#dag-arrow)"
309
- />
310
- ))}
311
- <defs>
312
- <marker id="dag-arrow" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="6" markerHeight="6" orient="auto-start-reverse">
313
- <path d="M 0 0 L 10 5 L 0 10 z" fill={theme.stroke.secondary} />
314
- </marker>
315
- </defs>
316
- {layout.nodes.map((n) => {
317
- const t = byId.get(n.id);
318
- if (!t) return null;
319
- return (
320
- <g key={n.id} transform={'translate(' + n.x + ',' + n.y + ')'}>
321
- <rect
322
- width={NODE_W}
323
- height={NODE_H}
324
- rx={8}
325
- ry={8}
326
- fill={theme.fill.secondary}
327
- stroke={theme.stroke.secondary}
328
- strokeWidth={t.status === 'RUNNING' ? 2 : 1}
329
- />
330
- <text x={12} y={22} fontSize={13} fontWeight={600} fill={theme.text.primary}>
331
- {t.id.length > 22 ? t.id.slice(0, 21) + '…' : t.id}
332
- </text>
333
- <text x={12} y={42} fontSize={10.5} fill={theme.text.secondary}>
334
- {t.executor} · {t.complexity} · {t.status}
335
- </text>
336
- <text x={12} y={58} fontSize={10.5} fill={theme.text.tertiary}>
337
- {formatDuration(t.durationMs)}
338
- </text>
339
- </g>
340
- );
341
- })}
342
- </svg>
343
- );
344
- }
345
-
346
- export default function DagRun(): JSX.Element {
347
- const counts = useMemo(
348
- () =>
349
- STATE.tasks.reduce(
350
- (acc, t) => {
351
- acc.total += 1;
352
- const key = t.status.toLowerCase() as 'pending' | 'running' | 'finished' | 'error' | 'skipped';
353
- if (key in acc) acc[key] += 1;
354
- return acc;
355
- },
356
- { total: 0, pending: 0, running: 0, finished: 0, error: 0, skipped: 0 },
357
- ),
358
- [],
359
- );
360
- const isFinal = STATE.finishedAt !== undefined;
361
- const statusLabel =
362
- STATE.runOutcome === 'PAUSED'
363
- ? 'PAUSED'
364
- : STATE.runOutcome === 'FAILED'
365
- ? 'FAILED'
366
- : isFinal
367
- ? 'COMPLETE'
368
- : 'RUNNING';
369
- const statusTone =
370
- STATE.runOutcome === 'FAILED'
371
- ? 'danger'
372
- : STATE.runOutcome === 'PAUSED'
373
- ? 'warning'
374
- : isFinal
375
- ? 'success'
376
- : 'info';
377
-
378
- return (
379
- <Stack gap={20}>
380
- <Stack gap={6}>
381
- <H1>{STATE.title}</H1>
382
- <Text tone="secondary" size="small">
383
- run {STATE.runId}
384
- </Text>
385
- <Row gap={10} align="center">
386
- <Pill tone={statusTone} active size="sm">
387
- {statusLabel}
388
- </Pill>
389
- <Text tone="secondary" size="small">
390
- {counts.total} tasks · elapsed {formatDuration(elapsed(STATE))}
391
- </Text>
392
- </Row>
393
- {STATE.runMessage ? (
394
- <Text tone="secondary" size="small">
395
- {STATE.runMessage}
396
- </Text>
397
- ) : null}
398
- </Stack>
399
-
400
- <Grid columns={6} gap={12}>
401
- <Stat value={String(counts.total)} label="Total" />
402
- <Stat value={String(counts.pending)} label="Pending" />
403
- <Stat value={String(counts.running)} label="Running" tone={counts.running > 0 ? 'info' : undefined} />
404
- <Stat value={String(counts.finished)} label="Finished" tone={counts.finished > 0 ? 'success' : undefined} />
405
- <Stat value={String(counts.error)} label="Errored" tone={counts.error > 0 ? 'danger' : undefined} />
406
- <Stat value={String(counts.skipped)} label="Skipped" />
407
- </Grid>
408
-
409
- <Divider />
410
-
411
- <Stack gap={12}>
412
- <H2>Graph</H2>
413
- <DAGGraph state={STATE} />
414
- </Stack>
415
-
416
- <Divider />
417
-
418
- <Stack gap={12}>
419
- <H2>Tasks</H2>
420
- <Stack gap={10}>
421
- {STATE.tasks.map((t) => (
422
- <Card key={t.id} collapsible defaultOpen={t.status !== 'PENDING'}>
423
- <CardHeader
424
- trailing={
425
- <Row gap={6} align="center">
426
- <Pill tone="neutral" size="sm">
427
- {t.executor}
428
- </Pill>
429
- <Pill tone={pillToneFor(t.status)} active={t.status !== 'PENDING'} size="sm">
430
- {t.status}
431
- </Pill>
432
- </Row>
433
- }
434
- >
435
- {t.id}
436
- </CardHeader>
437
- <CardBody>
438
- <Stack gap={8}>
439
- <Text tone="secondary" size="small">
440
- Model {t.model}
441
- {t.depends_on.length > 0 ? ' · depends on ' + t.depends_on.join(', ') : ''}
442
- {t.durationMs !== undefined ? ' · ' + formatDuration(t.durationMs) : ''}
443
- </Text>
444
- <Text size="small">
445
- <Text as="span" weight="semibold">
446
- Prompt:{' '}
447
- </Text>
448
- {t.subtask_prompt}
449
- </Text>
450
- {t.resultText ? (
451
- <pre style={{ margin: 0, whiteSpace: 'pre-wrap', wordBreak: 'break-word' }}>
452
- {t.resultText}
453
- {t.status === 'RUNNING' ? '\u2588' : ''}
454
- </pre>
455
- ) : null}
456
- {t.skippedReason ? (
457
- <Text size="small" tone="secondary">
458
- Skipped: {t.skippedReason}
459
- </Text>
460
- ) : null}
461
- {t.errorMessage ? (
462
- <Text size="small" tone="secondary">
463
- {t.errorMessage}
464
- </Text>
465
- ) : null}
466
- </Stack>
467
- </CardBody>
468
- </Card>
469
- ))}
470
- </Stack>
471
- </Stack>
472
- </Stack>
473
- );
248
+ const CANVAS_BODY = String.raw `const NODE_W = 200;
249
+ const NODE_H = 72;
250
+
251
+ function pillToneFor(status: TaskStatus): 'neutral' | 'info' | 'success' | 'warning' {
252
+ switch (status) {
253
+ case 'PENDING':
254
+ return 'neutral';
255
+ case 'RUNNING':
256
+ return 'info';
257
+ case 'FINISHED':
258
+ return 'success';
259
+ case 'ERROR':
260
+ return 'warning';
261
+ case 'SKIPPED':
262
+ return 'neutral';
263
+ }
264
+ }
265
+
266
+ function formatDuration(ms?: number): string {
267
+ if (ms === undefined) return '—';
268
+ if (ms < 1000) return ms + 'ms';
269
+ const s = ms / 1000;
270
+ if (s < 60) return s.toFixed(1) + 's';
271
+ const m = Math.floor(s / 60);
272
+ const rem = Math.round(s - m * 60);
273
+ return m + 'm ' + rem + 's';
274
+ }
275
+
276
+ function elapsed(state: RunState): number {
277
+ const end = state.finishedAt ?? Date.now();
278
+ return end - state.startedAt;
279
+ }
280
+
281
+ function DAGGraph({ state }: { state: RunState }): JSX.Element {
282
+ const theme = useHostTheme();
283
+ const layout = computeDAGLayout({
284
+ nodes: state.tasks.map((t) => ({ id: t.id })),
285
+ edges: state.tasks.flatMap((t) =>
286
+ t.depends_on.map((d) => ({ from: d, to: t.id })),
287
+ ),
288
+ direction: 'vertical',
289
+ nodeWidth: NODE_W,
290
+ nodeHeight: NODE_H,
291
+ rankGap: 72,
292
+ nodeGap: 40,
293
+ padding: 24,
294
+ });
295
+ const byId = new Map(state.tasks.map((t) => [t.id, t]));
296
+
297
+ return (
298
+ <svg width={layout.width} height={layout.height} style={{ display: 'block', maxWidth: '100%' }}>
299
+ {layout.edges.map((e, i) => (
300
+ <line
301
+ key={'e' + i}
302
+ x1={e.sourceX}
303
+ y1={e.sourceY}
304
+ x2={e.targetX}
305
+ y2={e.targetY}
306
+ stroke={theme.stroke.secondary}
307
+ strokeWidth={1.25}
308
+ markerEnd="url(#dag-arrow)"
309
+ />
310
+ ))}
311
+ <defs>
312
+ <marker id="dag-arrow" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="6" markerHeight="6" orient="auto-start-reverse">
313
+ <path d="M 0 0 L 10 5 L 0 10 z" fill={theme.stroke.secondary} />
314
+ </marker>
315
+ </defs>
316
+ {layout.nodes.map((n) => {
317
+ const t = byId.get(n.id);
318
+ if (!t) return null;
319
+ return (
320
+ <g key={n.id} transform={'translate(' + n.x + ',' + n.y + ')'}>
321
+ <rect
322
+ width={NODE_W}
323
+ height={NODE_H}
324
+ rx={8}
325
+ ry={8}
326
+ fill={theme.fill.secondary}
327
+ stroke={theme.stroke.secondary}
328
+ strokeWidth={t.status === 'RUNNING' ? 2 : 1}
329
+ />
330
+ <text x={12} y={22} fontSize={13} fontWeight={600} fill={theme.text.primary}>
331
+ {t.id.length > 22 ? t.id.slice(0, 21) + '…' : t.id}
332
+ </text>
333
+ <text x={12} y={42} fontSize={10.5} fill={theme.text.secondary}>
334
+ {t.executor} · {t.complexity} · {t.status}
335
+ </text>
336
+ <text x={12} y={58} fontSize={10.5} fill={theme.text.tertiary}>
337
+ {formatDuration(t.durationMs)}
338
+ </text>
339
+ </g>
340
+ );
341
+ })}
342
+ </svg>
343
+ );
344
+ }
345
+
346
+ export default function DagRun(): JSX.Element {
347
+ const counts = useMemo(
348
+ () =>
349
+ STATE.tasks.reduce(
350
+ (acc, t) => {
351
+ acc.total += 1;
352
+ const key = t.status.toLowerCase() as 'pending' | 'running' | 'finished' | 'error' | 'skipped';
353
+ if (key in acc) acc[key] += 1;
354
+ return acc;
355
+ },
356
+ { total: 0, pending: 0, running: 0, finished: 0, error: 0, skipped: 0 },
357
+ ),
358
+ [],
359
+ );
360
+ const isFinal = STATE.finishedAt !== undefined;
361
+ const statusLabel =
362
+ STATE.runOutcome === 'PAUSED'
363
+ ? 'PAUSED'
364
+ : STATE.runOutcome === 'FAILED'
365
+ ? 'FAILED'
366
+ : isFinal
367
+ ? 'COMPLETE'
368
+ : 'RUNNING';
369
+ const statusTone =
370
+ STATE.runOutcome === 'FAILED'
371
+ ? 'danger'
372
+ : STATE.runOutcome === 'PAUSED'
373
+ ? 'warning'
374
+ : isFinal
375
+ ? 'success'
376
+ : 'info';
377
+
378
+ return (
379
+ <Stack gap={20}>
380
+ <Stack gap={6}>
381
+ <H1>{STATE.title}</H1>
382
+ <Text tone="secondary" size="small">
383
+ run {STATE.runId}
384
+ </Text>
385
+ <Row gap={10} align="center">
386
+ <Pill tone={statusTone} active size="sm">
387
+ {statusLabel}
388
+ </Pill>
389
+ <Text tone="secondary" size="small">
390
+ {counts.total} tasks · elapsed {formatDuration(elapsed(STATE))}
391
+ </Text>
392
+ </Row>
393
+ {STATE.runMessage ? (
394
+ <Text tone="secondary" size="small">
395
+ {STATE.runMessage}
396
+ </Text>
397
+ ) : null}
398
+ </Stack>
399
+
400
+ <Grid columns={6} gap={12}>
401
+ <Stat value={String(counts.total)} label="Total" />
402
+ <Stat value={String(counts.pending)} label="Pending" />
403
+ <Stat value={String(counts.running)} label="Running" tone={counts.running > 0 ? 'info' : undefined} />
404
+ <Stat value={String(counts.finished)} label="Finished" tone={counts.finished > 0 ? 'success' : undefined} />
405
+ <Stat value={String(counts.error)} label="Errored" tone={counts.error > 0 ? 'danger' : undefined} />
406
+ <Stat value={String(counts.skipped)} label="Skipped" />
407
+ </Grid>
408
+
409
+ <Divider />
410
+
411
+ <Stack gap={12}>
412
+ <H2>Graph</H2>
413
+ <DAGGraph state={STATE} />
414
+ </Stack>
415
+
416
+ <Divider />
417
+
418
+ <Stack gap={12}>
419
+ <H2>Tasks</H2>
420
+ <Stack gap={10}>
421
+ {STATE.tasks.map((t) => (
422
+ <Card key={t.id} collapsible defaultOpen={t.status !== 'PENDING'}>
423
+ <CardHeader
424
+ trailing={
425
+ <Row gap={6} align="center">
426
+ <Pill tone="neutral" size="sm">
427
+ {t.executor}
428
+ </Pill>
429
+ <Pill tone={pillToneFor(t.status)} active={t.status !== 'PENDING'} size="sm">
430
+ {t.status}
431
+ </Pill>
432
+ </Row>
433
+ }
434
+ >
435
+ {t.id}
436
+ </CardHeader>
437
+ <CardBody>
438
+ <Stack gap={8}>
439
+ <Text tone="secondary" size="small">
440
+ Model {t.model}
441
+ {t.depends_on.length > 0 ? ' · depends on ' + t.depends_on.join(', ') : ''}
442
+ {t.durationMs !== undefined ? ' · ' + formatDuration(t.durationMs) : ''}
443
+ </Text>
444
+ <Text size="small">
445
+ <Text as="span" weight="semibold">
446
+ Prompt:{' '}
447
+ </Text>
448
+ {t.subtask_prompt}
449
+ </Text>
450
+ {t.resultText ? (
451
+ <pre style={{ margin: 0, whiteSpace: 'pre-wrap', wordBreak: 'break-word' }}>
452
+ {t.resultText}
453
+ {t.status === 'RUNNING' ? '\u2588' : ''}
454
+ </pre>
455
+ ) : null}
456
+ {t.skippedReason ? (
457
+ <Text size="small" tone="secondary">
458
+ Skipped: {t.skippedReason}
459
+ </Text>
460
+ ) : null}
461
+ {t.errorMessage ? (
462
+ <Text size="small" tone="secondary">
463
+ {t.errorMessage}
464
+ </Text>
465
+ ) : null}
466
+ </Stack>
467
+ </CardBody>
468
+ </Card>
469
+ ))}
470
+ </Stack>
471
+ </Stack>
472
+ </Stack>
473
+ );
474
474
  }`;
@@ -27,7 +27,7 @@ function renderLoopTemplate(template, iteration) {
27
27
  .replace(/\{\{\s*iteration\s*\}\}/g, String(iteration))
28
28
  .replace(/\{\{\s*iterationStatus\s*\}\}/g, iterationStatus);
29
29
  }
30
- function buildLoopBodyChildTask(input) {
30
+ export function buildLoopBodyChildTask(input) {
31
31
  const { parent, bodyTask, iteration, nodeId, bodyIdMap } = input;
32
32
  const mappedDepends = bodyTask.dependsOn.map((depId) => bodyIdMap.get(depId) ?? depId);
33
33
  return {
@@ -37,6 +37,7 @@ function buildLoopBodyChildTask(input) {
37
37
  subtask_prompt: renderLoopTemplate(bodyTask.subtaskPromptTemplate, iteration),
38
38
  executor: bodyTask.executor,
39
39
  role: bodyTask.role,
40
+ skills: bodyTask.skills,
40
41
  writePolicy: bodyTask.writePolicy,
41
42
  allowedPaths: bodyTask.allowedPaths,
42
43
  forbiddenPaths: bodyTask.forbiddenPaths,
@@ -18,6 +18,7 @@ export function buildExpandedChildTask(input) {
18
18
  subtask_prompt: subtaskPrompt,
19
19
  executor: child.executor,
20
20
  role: child.role,
21
+ skills: child.skills,
21
22
  writePolicy: child.writePolicy,
22
23
  allowedPaths: renderDynamicPatternList(child.allowedPaths, item, index, expansion.itemName) ?? [],
23
24
  forbiddenPaths: renderDynamicPatternList(child.forbiddenPaths, item, index, expansion.itemName) ?? [],
@@ -22,9 +22,9 @@ const HYBRID_DEFAULTS = {
22
22
  };
23
23
  const HYBRID_SKILLS_BY_ROLE = {
24
24
  planner: ["loop-agent"],
25
- scout: ["ai-engineering-context"],
26
- implementer: ["verification-before-completion"],
27
- reviewer: ["requesting-code-review"],
25
+ scout: ["ai-engineering-context", "codebase-scout"],
26
+ implementer: ["test-driven-development", "verification-before-completion"],
27
+ reviewer: ["requesting-code-review", "code-review-core"],
28
28
  verifier: ["verification-before-completion", "systematic-debugging"],
29
29
  closeout: ["loop-agent", "verification-before-completion"],
30
30
  };
@@ -4,10 +4,10 @@
4
4
  */
5
5
  export const DEFAULT_SKILLS_BY_ROLE = {
6
6
  planner: ["loop-agent"],
7
- scout: ["ai-engineering-context"],
8
- implementer: ["verification-before-completion"],
7
+ scout: ["ai-engineering-context", "codebase-scout"],
8
+ implementer: ["test-driven-development", "verification-before-completion"],
9
9
  verifier: ["verification-before-completion", "systematic-debugging"],
10
- reviewer: ["requesting-code-review"],
10
+ reviewer: ["requesting-code-review", "code-review-core"],
11
11
  supervisor: ["verification-before-completion", "loop-agent"],
12
12
  closeout: ["loop-agent", "verification-before-completion"],
13
13
  };
@@ -94,6 +94,7 @@ export const dagStaticConfigSchema = z.object({
94
94
  export const dagDynamicExpansionChildTaskSchema = z.object({
95
95
  executor: dagNodeExecutorSchema.default("pi"),
96
96
  role: dagRoleSchema.optional(),
97
+ skills: z.array(z.string()).optional(),
97
98
  complexity: dagComplexitySchema.default("LOW"),
98
99
  subtaskPromptTemplate: z.string().min(1),
99
100
  outputContract: z.string().optional(),
@@ -153,6 +154,7 @@ export const dagDynamicLoopBodyTaskSchema = z.object({
153
154
  dependsOn: z.array(z.string()).default([]),
154
155
  executor: dagNodeExecutorSchema.default("pi"),
155
156
  role: dagRoleSchema.optional(),
157
+ skills: z.array(z.string()).optional(),
156
158
  complexity: dagComplexitySchema.default("LOW"),
157
159
  subtaskPromptTemplate: z.string().min(1),
158
160
  outputContract: z.string().optional(),