@osolmaz/pi-workflows 0.1.0 → 0.3.0

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 (171) hide show
  1. package/README.md +107 -26
  2. package/dist/builtins/monitor.workflow.d.ts +71 -0
  3. package/dist/builtins/monitor.workflow.js +234 -0
  4. package/dist/builtins/monitor.workflow.js.map +1 -0
  5. package/dist/controllers/conditions.d.ts +6 -0
  6. package/dist/controllers/conditions.js +68 -0
  7. package/dist/controllers/conditions.js.map +1 -0
  8. package/dist/controllers/definition.d.ts +6 -0
  9. package/dist/controllers/definition.js +45 -0
  10. package/dist/controllers/definition.js.map +1 -0
  11. package/dist/controllers/effects.d.ts +14 -0
  12. package/dist/controllers/effects.js +104 -0
  13. package/dist/controllers/effects.js.map +1 -0
  14. package/dist/controllers/errors.d.ts +12 -0
  15. package/dist/controllers/errors.js +25 -0
  16. package/dist/controllers/errors.js.map +1 -0
  17. package/dist/controllers/index.d.ts +13 -0
  18. package/dist/controllers/index.js +13 -0
  19. package/dist/controllers/index.js.map +1 -0
  20. package/dist/controllers/json.d.ts +5 -0
  21. package/dist/controllers/json.js +57 -0
  22. package/dist/controllers/json.js.map +1 -0
  23. package/dist/controllers/loader.d.ts +23 -0
  24. package/dist/controllers/loader.js +74 -0
  25. package/dist/controllers/loader.js.map +1 -0
  26. package/dist/controllers/manager.d.ts +58 -0
  27. package/dist/controllers/manager.js +399 -0
  28. package/dist/controllers/manager.js.map +1 -0
  29. package/dist/controllers/results.d.ts +5 -0
  30. package/dist/controllers/results.js +32 -0
  31. package/dist/controllers/results.js.map +1 -0
  32. package/dist/controllers/sqlite.d.ts +212 -0
  33. package/dist/controllers/sqlite.js +1009 -0
  34. package/dist/controllers/sqlite.js.map +1 -0
  35. package/dist/controllers/store.d.ts +112 -0
  36. package/dist/controllers/store.js +32 -0
  37. package/dist/controllers/store.js.map +1 -0
  38. package/dist/controllers/types.d.ts +159 -0
  39. package/dist/controllers/types.js +2 -0
  40. package/dist/controllers/types.js.map +1 -0
  41. package/dist/controllers/workflow-engine-scheduler.d.ts +25 -0
  42. package/dist/controllers/workflow-engine-scheduler.js +93 -0
  43. package/dist/controllers/workflow-engine-scheduler.js.map +1 -0
  44. package/dist/controllers/workflows.d.ts +27 -0
  45. package/dist/controllers/workflows.js +109 -0
  46. package/dist/controllers/workflows.js.map +1 -0
  47. package/dist/extension/controller-host.d.ts +47 -0
  48. package/dist/extension/controller-host.js +110 -0
  49. package/dist/extension/controller-host.js.map +1 -0
  50. package/dist/extension/executor.d.ts +14 -1
  51. package/dist/extension/executor.js +12 -2
  52. package/dist/extension/executor.js.map +1 -1
  53. package/dist/extension/index.d.ts +7 -0
  54. package/dist/extension/index.js +1118 -89
  55. package/dist/extension/index.js.map +1 -1
  56. package/dist/extension/recorder.d.ts +91 -0
  57. package/dist/extension/recorder.js +537 -0
  58. package/dist/extension/recorder.js.map +1 -0
  59. package/dist/extension/session-events.d.ts +134 -0
  60. package/dist/extension/session-events.js +60 -0
  61. package/dist/extension/session-events.js.map +1 -0
  62. package/dist/extension/widget.js +25 -24
  63. package/dist/extension/widget.js.map +1 -1
  64. package/dist/extension/workflow-tool.d.ts +28 -0
  65. package/dist/extension/workflow-tool.js +33 -0
  66. package/dist/extension/workflow-tool.js.map +1 -0
  67. package/dist/host/processes.d.ts +24 -0
  68. package/dist/host/processes.js +114 -0
  69. package/dist/host/processes.js.map +1 -0
  70. package/dist/host/rpc-bridge.d.ts +9 -0
  71. package/dist/host/rpc-bridge.js +39 -0
  72. package/dist/host/rpc-bridge.js.map +1 -0
  73. package/dist/host/rpc-executor.d.ts +38 -0
  74. package/dist/host/rpc-executor.js +254 -0
  75. package/dist/host/rpc-executor.js.map +1 -0
  76. package/dist/host/runner.d.ts +49 -0
  77. package/dist/host/runner.js +350 -0
  78. package/dist/host/runner.js.map +1 -0
  79. package/dist/render/canvas.d.ts +1 -1
  80. package/dist/render/canvas.js +5 -0
  81. package/dist/render/canvas.js.map +1 -1
  82. package/dist/render/graph-render.d.ts +5 -0
  83. package/dist/render/graph-render.js +211 -48
  84. package/dist/render/graph-render.js.map +1 -1
  85. package/dist/viewer/cli.d.ts +7 -3
  86. package/dist/viewer/cli.js +150 -19
  87. package/dist/viewer/cli.js.map +1 -1
  88. package/dist/viewer/render.js +19 -3
  89. package/dist/viewer/render.js.map +1 -1
  90. package/dist/viewer/session-reducer.d.ts +45 -0
  91. package/dist/viewer/session-reducer.js +266 -0
  92. package/dist/viewer/session-reducer.js.map +1 -0
  93. package/dist/workflows/artifacts.d.ts +40 -0
  94. package/dist/workflows/artifacts.js +155 -0
  95. package/dist/workflows/artifacts.js.map +1 -0
  96. package/dist/workflows/engine.d.ts +38 -0
  97. package/dist/workflows/engine.js +281 -20
  98. package/dist/workflows/engine.js.map +1 -1
  99. package/dist/workflows/errors.d.ts +23 -0
  100. package/dist/workflows/errors.js +38 -0
  101. package/dist/workflows/errors.js.map +1 -1
  102. package/dist/workflows/graph.js +0 -5
  103. package/dist/workflows/graph.js.map +1 -1
  104. package/dist/workflows/index.d.ts +3 -2
  105. package/dist/workflows/index.js +2 -1
  106. package/dist/workflows/index.js.map +1 -1
  107. package/dist/workflows/loader.d.ts +5 -3
  108. package/dist/workflows/loader.js +10 -1
  109. package/dist/workflows/loader.js.map +1 -1
  110. package/dist/workflows/schema.js +1 -1
  111. package/dist/workflows/schema.js.map +1 -1
  112. package/dist/workflows/store.d.ts +98 -10
  113. package/dist/workflows/store.js +921 -46
  114. package/dist/workflows/store.js.map +1 -1
  115. package/dist/workflows/types.d.ts +136 -3
  116. package/docs/CONTROLLERS.md +215 -0
  117. package/docs/development.md +51 -24
  118. package/docs/live-replay-protocol.md +155 -0
  119. package/docs/plans/2026-08-04-controller-runtime-plan.md +169 -0
  120. package/docs/plans/2026-08-05-always-on-workflows-plan.md +125 -0
  121. package/docs/plans/2026-08-10-agent-managed-monitor-workflows-plan.md +184 -0
  122. package/docs/plans/piw-viewer-experience-implementation-plan.md +674 -0
  123. package/docs/plans/replayable-run-bundles-implementation-plan.md +65 -0
  124. package/docs/plans/session-event-replay-implementation-plan.md +494 -0
  125. package/docs/plans/tui-viewer-implementation-plan.md +64 -0
  126. package/docs/run-bundles.md +359 -55
  127. package/docs/session-event-journal.md +470 -0
  128. package/docs/tui-viewer.md +218 -0
  129. package/docs/workflows.md +131 -11
  130. package/examples/controllers/pull-request.controller.ts +215 -0
  131. package/package.json +11 -2
  132. package/src/builtins/monitor.workflow.ts +278 -0
  133. package/src/controllers/conditions.ts +110 -0
  134. package/src/controllers/definition.ts +65 -0
  135. package/src/controllers/effects.ts +123 -0
  136. package/src/controllers/errors.ts +27 -0
  137. package/src/controllers/index.ts +90 -0
  138. package/src/controllers/json.ts +62 -0
  139. package/src/controllers/loader.ts +104 -0
  140. package/src/controllers/manager.ts +533 -0
  141. package/src/controllers/results.ts +46 -0
  142. package/src/controllers/sqlite.ts +1427 -0
  143. package/src/controllers/store.ts +160 -0
  144. package/src/controllers/types.ts +183 -0
  145. package/src/controllers/workflow-engine-scheduler.ts +145 -0
  146. package/src/controllers/workflows.ts +152 -0
  147. package/src/extension/controller-host.ts +163 -0
  148. package/src/extension/executor.ts +29 -2
  149. package/src/extension/index.ts +1315 -109
  150. package/src/extension/recorder.ts +662 -0
  151. package/src/extension/session-events.ts +119 -0
  152. package/src/extension/widget.ts +26 -24
  153. package/src/extension/workflow-tool.ts +59 -0
  154. package/src/host/processes.ts +119 -0
  155. package/src/host/rpc-bridge.ts +44 -0
  156. package/src/host/rpc-executor.ts +299 -0
  157. package/src/host/runner.ts +406 -0
  158. package/src/render/canvas.ts +19 -1
  159. package/src/render/graph-render.ts +277 -44
  160. package/src/viewer/cli.ts +167 -21
  161. package/src/viewer/render.ts +21 -3
  162. package/src/viewer/session-reducer.ts +347 -0
  163. package/src/workflows/artifacts.ts +188 -0
  164. package/src/workflows/engine.ts +365 -19
  165. package/src/workflows/errors.ts +45 -0
  166. package/src/workflows/graph.ts +0 -5
  167. package/src/workflows/index.ts +15 -0
  168. package/src/workflows/loader.ts +13 -3
  169. package/src/workflows/schema.ts +1 -1
  170. package/src/workflows/store.ts +1157 -48
  171. package/src/workflows/types.ts +151 -3
@@ -1,15 +1,26 @@
1
1
  import { randomUUID } from "node:crypto";
2
2
  import { isDeepStrictEqual } from "node:util";
3
- import { CancelledError, errorMessage, isAbortLikeError, TimeoutError } from "./errors.js";
3
+ import { resolveArtifacts } from "./artifacts.js";
4
+ import {
5
+ CancelledError,
6
+ errorMessage,
7
+ isAbortLikeError,
8
+ isClaimLostError,
9
+ isRunParkedError,
10
+ RunParkedError,
11
+ TimeoutError,
12
+ WorkflowSourceChangedError,
13
+ } from "./errors.js";
4
14
  import { resolveNext, resolveNextForOutcome, validateWorkflowDefinition } from "./graph.js";
5
15
  import { extractJsonValue } from "./json.js";
6
16
  import { runShellAction, shellResultFromError } from "./shell.js";
7
- import { WorkflowRunStore, createRunId } from "./store.js";
17
+ import { RUN_STATE_SCHEMA, WorkflowRunStore, createRunId, readRunBundle } from "./store.js";
8
18
  import type {
9
19
  AgentNodeDefinition,
10
20
  AgentStepExecutor,
11
21
  ActionNodeDefinition,
12
22
  CheckpointNodeDefinition,
23
+ ConversationRange,
13
24
  ShellActionNodeDefinition,
14
25
  ShellActionResult,
15
26
  WorkflowActionReceipt,
@@ -35,6 +46,7 @@ type NodeExecution = {
35
46
  output: unknown;
36
47
  promptText: string | null;
37
48
  action?: WorkflowActionReceipt;
49
+ conversation?: ConversationRange;
38
50
  };
39
51
 
40
52
  /**
@@ -64,17 +76,22 @@ export class WorkflowEngine {
64
76
  private readonly defaultNodeTimeoutMs: number;
65
77
  private readonly maxSteps: number;
66
78
  private readonly onEvent?: WorkflowEngineOptions["onEvent"];
79
+ private readonly onRunStarted?: WorkflowEngineOptions["onRunStarted"];
80
+ private readonly onRunFinishing?: WorkflowEngineOptions["onRunFinishing"];
67
81
  private activeAbort: AbortController | null = null;
68
82
  private cancelled = false;
83
+ private parked = false;
69
84
  private paused = false;
70
85
  private wakePause: (() => void) | null = null;
71
86
 
72
87
  constructor(options: WorkflowEngineOptions) {
73
88
  this.executor = options.executor;
74
- this.store = new WorkflowRunStore(options.outputRoot);
89
+ this.store = options.store ?? new WorkflowRunStore(options.outputRoot);
75
90
  this.defaultNodeTimeoutMs = options.defaultNodeTimeoutMs ?? DEFAULT_NODE_TIMEOUT_MS;
76
91
  this.maxSteps = options.maxSteps ?? DEFAULT_MAX_STEPS;
77
92
  this.onEvent = options.onEvent;
93
+ this.onRunStarted = options.onRunStarted;
94
+ this.onRunFinishing = options.onRunFinishing;
78
95
  }
79
96
 
80
97
  get outputRoot(): string {
@@ -90,6 +107,17 @@ export class WorkflowEngine {
90
107
  this.wakePause?.();
91
108
  }
92
109
 
110
+ /**
111
+ * Stop without a terminal event so another runner can claim and resume
112
+ * the run. The active node aborts; its partial attempt is never recorded,
113
+ * so resume reruns that node from its last persisted boundary.
114
+ */
115
+ park(): void {
116
+ this.parked = true;
117
+ this.activeAbort?.abort(new CancelledError());
118
+ this.wakePause?.();
119
+ }
120
+
93
121
  /**
94
122
  * Request a pause. The current step finishes normally; the engine then
95
123
  * holds before dispatching the next node until `resume` (or `cancel`).
@@ -112,17 +140,27 @@ export class WorkflowEngine {
112
140
  async run(
113
141
  workflow: WorkflowDefinition,
114
142
  input: unknown,
115
- options: { workflowPath?: string } = {},
143
+ options: { workflowPath?: string; workflowHash?: string; runId?: string } = {},
116
144
  ): Promise<WorkflowRunResult> {
117
145
  validateWorkflowDefinition(workflow);
118
146
  // Fail before any bundle exists so bad input cannot leave a partial run
119
147
  // on disk or silently change shape when state.json round-trips.
120
148
  const normalizedInput = input === undefined ? null : input;
121
149
  assertJsonSerializable(normalizedInput, "Workflow run input");
150
+ if (options.runId !== undefined && !/^[A-Za-z0-9][A-Za-z0-9._-]{0,199}$/.test(options.runId)) {
151
+ throw new Error(`Invalid workflow run id: ${JSON.stringify(options.runId)}`);
152
+ }
122
153
  this.cancelled = false;
123
154
  this.paused = false;
155
+ this.parked = false;
124
156
 
125
- const state = await this.createRunState(workflow, normalizedInput, options.workflowPath);
157
+ const state = await this.createRunState(
158
+ workflow,
159
+ normalizedInput,
160
+ options.workflowPath,
161
+ options.workflowHash,
162
+ options.runId,
163
+ );
126
164
  const runDir = await this.store.initializeRunBundle(workflow, state);
127
165
  await this.persist(runDir, state, {
128
166
  scope: "run",
@@ -130,19 +168,284 @@ export class WorkflowEngine {
130
168
  payload: {
131
169
  workflowName: workflow.name,
132
170
  ...(state.runTitle ? { runTitle: state.runTitle } : {}),
171
+ input: state.input,
133
172
  },
134
173
  });
174
+ // Awaited so anything the hook writes (e.g. a session binding and its
175
+ // `session_bound` event) lands before node events and can never trail
176
+ // the terminal event of a fast run.
177
+ await this.onRunStarted?.(runDir, state);
135
178
 
136
179
  try {
137
180
  await this.executeGraph(workflow, state, runDir);
138
181
  } catch (error) {
139
- const cancelled = this.cancelled || isAbortLikeError(error);
182
+ if (isRunParkedError(error) || this.parked) {
183
+ return { runDir, state };
184
+ }
185
+ await this.finishAfterError(runDir, state, error);
186
+ return { runDir, state };
187
+ }
188
+ return { runDir, state };
189
+ }
190
+
191
+ /**
192
+ * Resume an interrupted run at the node it stopped on. The caller must
193
+ * hold the run's queue claim. Completed nodes replay from the recorded
194
+ * state; only the interrupted node and everything downstream rerun.
195
+ */
196
+ async resumeRun(
197
+ workflow: WorkflowDefinition,
198
+ runId: string,
199
+ options: { workflowHash?: string; force?: boolean } = {},
200
+ ): Promise<WorkflowRunResult> {
201
+ validateWorkflowDefinition(workflow);
202
+ // Reset before any await: a park or cancel landing during preparation
203
+ // must survive, or a host drain would hang while the run executes.
204
+ this.cancelled = false;
205
+ this.paused = false;
206
+ this.parked = false;
207
+ const bundle = await this.store.prepareRunResume(runId);
208
+ const { runDir } = bundle;
209
+ const state = bundle.state;
210
+ const hashMismatch =
211
+ state.workflowHash !== undefined &&
212
+ options.workflowHash !== undefined &&
213
+ state.workflowHash !== options.workflowHash;
214
+ if (hashMismatch && options.force !== true) {
215
+ throw new WorkflowSourceChangedError(runId);
216
+ }
217
+
218
+ const point = this.resumePointFor(workflow, state, "wait");
219
+ // A resumed run starts unpaused; the operator can pause again. The
220
+ // interrupted node's stale in-flight markers go away before the resume
221
+ // event so the projection matches what the engine is about to do.
222
+ delete state.paused;
223
+ delete state.currentNode;
224
+ delete state.currentAttemptId;
225
+ delete state.currentNodeStartedAt;
226
+ delete state.statusDetail;
227
+ await this.persist(runDir, state, {
228
+ scope: "run",
229
+ type: "run_resumed",
230
+ payload: {
231
+ ...(point.nodeId !== null ? { resumeAt: point.nodeId } : {}),
232
+ replayedSteps: state.steps.length,
233
+ ...(hashMismatch ? { workflowHashMismatch: true, forced: true } : {}),
234
+ },
235
+ });
236
+ await this.onRunStarted?.(runDir, state);
237
+
238
+ if (point.nodeId === null) {
239
+ // The last recorded transition already finished the graph; the crash
240
+ // happened before the terminal event was written. A finished
241
+ // checkpoint restores its waiting gate rather than completing.
242
+ if (point.waitingOn !== undefined) {
243
+ await this.finishRun(runDir, state, "waiting", {
244
+ waitingOn: point.waitingOn,
245
+ finalOutput: point.lastOutput,
246
+ });
247
+ } else if (point.failedResult === undefined) {
248
+ await this.finishRun(runDir, state, "completed", { finalOutput: point.lastOutput });
249
+ } else {
250
+ const timedOut = point.failedResult.outcome === "timed_out";
251
+ await this.finishRun(runDir, state, timedOut ? "timed_out" : "failed", {
252
+ error: point.failedResult.error ?? `Workflow node failed: ${point.failedResult.nodeId}`,
253
+ });
254
+ }
255
+ return { runDir, state };
256
+ }
257
+
258
+ try {
259
+ await this.executeGraph(
260
+ workflow,
261
+ state,
262
+ runDir,
263
+ point.nodeId,
264
+ state.steps.length,
265
+ point.lastOutput,
266
+ );
267
+ } catch (error) {
268
+ if (isRunParkedError(error) || this.parked) {
269
+ return { runDir, state };
270
+ }
271
+ await this.finishAfterError(runDir, state, error);
272
+ return { runDir, state };
273
+ }
274
+ return { runDir, state };
275
+ }
276
+
277
+ /**
278
+ * Start a continuation run from a checkpointed parent. The new run gets a
279
+ * fresh bundle and trace, carries forward the parent's outputs, results,
280
+ * and step accounting, and continues routing after the checkpoint.
281
+ */
282
+ async continueRun(
283
+ workflow: WorkflowDefinition,
284
+ parentRunId: string,
285
+ input: unknown,
286
+ options: { workflowPath?: string; workflowHash?: string; runId?: string; force?: boolean } = {},
287
+ ): Promise<WorkflowRunResult> {
288
+ validateWorkflowDefinition(workflow);
289
+ this.cancelled = false;
290
+ this.paused = false;
291
+ this.parked = false;
292
+ const parent = await readRunBundle(this.store.runDirFor(parentRunId));
293
+ if (parent === null) {
294
+ throw new Error(`Cannot continue from unreadable workflow run: ${parentRunId}`);
295
+ }
296
+ if (parent.state.status !== "waiting" || parent.state.waitingOn === undefined) {
297
+ throw new Error(
298
+ `Cannot continue workflow run ${parentRunId} with status ${parent.state.status}`,
299
+ );
300
+ }
301
+ const hashMismatch =
302
+ parent.state.workflowHash !== undefined &&
303
+ options.workflowHash !== undefined &&
304
+ parent.state.workflowHash !== options.workflowHash;
305
+ if (hashMismatch && options.force !== true) {
306
+ throw new WorkflowSourceChangedError(parentRunId);
307
+ }
308
+
309
+ const normalizedInput = input === undefined ? null : input;
310
+ assertJsonSerializable(normalizedInput, "Workflow run input");
311
+ if (options.runId !== undefined && !/^[A-Za-z0-9][A-Za-z0-9._-]{0,199}$/.test(options.runId)) {
312
+ throw new Error(`Invalid workflow run id: ${JSON.stringify(options.runId)}`);
313
+ }
314
+
315
+ const state = await this.createRunState(
316
+ workflow,
317
+ normalizedInput,
318
+ options.workflowPath,
319
+ options.workflowHash,
320
+ options.runId,
321
+ );
322
+ state.parentRunId = parentRunId;
323
+ // Artifact references point into the parent's bundle, so carried values
324
+ // are fully resolved here and re-externalized into the new bundle.
325
+ state.outputs = (await resolveArtifacts(
326
+ parent.state.outputs,
327
+ parent.runDir,
328
+ )) as WorkflowRunState["outputs"];
329
+ state.results = (await resolveArtifacts(
330
+ parent.state.results,
331
+ parent.runDir,
332
+ )) as WorkflowRunState["results"];
333
+ state.steps = (await resolveArtifacts(
334
+ parent.state.steps,
335
+ parent.runDir,
336
+ )) as WorkflowRunState["steps"];
337
+ state.carriedStepCount = state.steps.length;
338
+
339
+ const runDir = await this.store.initializeRunBundle(workflow, state);
340
+ await this.persist(runDir, state, {
341
+ scope: "run",
342
+ type: "run_started",
343
+ payload: {
344
+ workflowName: workflow.name,
345
+ ...(state.runTitle ? { runTitle: state.runTitle } : {}),
346
+ input: state.input,
347
+ continuedFrom: parentRunId,
348
+ checkpoint: parent.state.waitingOn,
349
+ carriedSteps: state.steps.length,
350
+ },
351
+ });
352
+ await this.onRunStarted?.(runDir, state);
353
+
354
+ const point = this.resumePointFor(workflow, state, "continue");
355
+ if (point.nodeId === null) {
356
+ // The checkpoint was the final node; the answer completes the chain.
357
+ await this.finishRun(runDir, state, "completed", { finalOutput: point.lastOutput });
358
+ return { runDir, state };
359
+ }
360
+ try {
361
+ await this.executeGraph(
362
+ workflow,
363
+ state,
364
+ runDir,
365
+ point.nodeId,
366
+ state.steps.length,
367
+ point.lastOutput,
368
+ );
369
+ } catch (error) {
370
+ if (isRunParkedError(error) || this.parked) {
371
+ return { runDir, state };
372
+ }
373
+ await this.finishAfterError(runDir, state, error);
374
+ return { runDir, state };
375
+ }
376
+ return { runDir, state };
377
+ }
378
+
379
+ /**
380
+ * Find where a resumed run continues. An in-flight node reruns; otherwise
381
+ * routing continues from the last recorded step. A null nodeId means the
382
+ * graph was already done when the crash hit.
383
+ */
384
+ private resumePointFor(
385
+ workflow: WorkflowDefinition,
386
+ state: WorkflowRunState,
387
+ checkpointBehavior: "wait" | "continue",
388
+ ): {
389
+ nodeId: string | null;
390
+ lastOutput?: unknown;
391
+ failedResult?: WorkflowNodeResult;
392
+ waitingOn?: string;
393
+ } {
394
+ if (state.currentNode !== undefined) {
395
+ if (workflow.nodes[state.currentNode] === undefined) {
396
+ throw new Error(`Resume node is missing from the workflow: ${state.currentNode}`);
397
+ }
398
+ return { nodeId: state.currentNode };
399
+ }
400
+ const lastStep = state.steps.at(-1);
401
+ if (lastStep === undefined) {
402
+ return { nodeId: workflow.startAt };
403
+ }
404
+ const result = state.results[lastStep.nodeId];
405
+ if (result === undefined) {
406
+ return { nodeId: lastStep.nodeId };
407
+ }
408
+ if (result.outcome === "ok") {
409
+ // A recorded checkpoint means the run should be waiting; a crash
410
+ // before the run_waiting persist restores the gate instead of
411
+ // routing past it. The gate applies to this run's own checkpoint
412
+ // only: a continuation's carried steps end with the parent's
413
+ // already-answered checkpoint, and routing must continue from it.
414
+ const isCarriedStep = state.steps.length <= (state.carriedStepCount ?? 0);
415
+ if (
416
+ checkpointBehavior === "wait" &&
417
+ !isCarriedStep &&
418
+ workflow.nodes[lastStep.nodeId]?.nodeType === "checkpoint"
419
+ ) {
420
+ return { nodeId: null, waitingOn: lastStep.nodeId, lastOutput: result.output };
421
+ }
422
+ const next = resolveNext(workflow.edges, lastStep.nodeId, result.output, result);
423
+ return next === null
424
+ ? { nodeId: null, lastOutput: result.output }
425
+ : { nodeId: next, lastOutput: result.output };
426
+ }
427
+ const next = resolveNextForOutcome(workflow.edges, lastStep.nodeId, result);
428
+ return next === null ? { nodeId: null, failedResult: result } : { nodeId: next };
429
+ }
430
+
431
+ private async finishAfterError(
432
+ runDir: string,
433
+ state: WorkflowRunState,
434
+ error: unknown,
435
+ ): Promise<void> {
436
+ const cancelled = this.cancelled || isAbortLikeError(error);
437
+ try {
140
438
  await this.finishRun(runDir, state, cancelled ? "cancelled" : "failed", {
141
439
  error: errorMessage(error),
142
440
  });
143
- return { runDir, state };
441
+ } catch (finishError) {
442
+ // A fenced-out runner must not touch the bundle, including terminal
443
+ // projections. Propagate the claim loss instead of the node error.
444
+ if (isClaimLostError(finishError)) {
445
+ throw finishError;
446
+ }
447
+ throw error;
144
448
  }
145
- return { runDir, state };
146
449
  }
147
450
 
148
451
  /**
@@ -175,13 +478,18 @@ export class WorkflowEngine {
175
478
  workflow: WorkflowDefinition,
176
479
  input: unknown,
177
480
  workflowPath: string | undefined,
481
+ workflowHash: string | undefined,
482
+ runId: string | undefined,
178
483
  ): Promise<WorkflowRunState> {
179
484
  const now = new Date().toISOString();
180
485
  return {
181
- runId: createRunId(workflow.name),
486
+ schema: RUN_STATE_SCHEMA,
487
+ traceSeq: 0,
488
+ runId: runId ?? createRunId(workflow.name),
182
489
  workflowName: workflow.name,
183
490
  ...(await this.resolveTitleBounded(workflow, input)),
184
491
  ...(workflowPath !== undefined ? { workflowPath } : {}),
492
+ ...(workflowHash !== undefined ? { workflowHash } : {}),
185
493
  startedAt: now,
186
494
  updatedAt: now,
187
495
  status: "running",
@@ -196,11 +504,14 @@ export class WorkflowEngine {
196
504
  workflow: WorkflowDefinition,
197
505
  state: WorkflowRunState,
198
506
  runDir: string,
507
+ startNodeId: string | null = workflow.startAt,
508
+ executedStepsBase = 0,
509
+ initialLastOutput?: unknown,
199
510
  ): Promise<void> {
200
511
  const maxSteps = workflow.maxSteps ?? this.maxSteps;
201
- let currentNodeId: string | null = workflow.startAt;
202
- let executedSteps = 0;
203
- let lastOutput: unknown;
512
+ let currentNodeId: string | null = startNodeId;
513
+ let executedSteps = executedStepsBase;
514
+ let lastOutput: unknown = initialLastOutput;
204
515
 
205
516
  while (currentNodeId !== null) {
206
517
  await this.holdWhilePaused(state, runDir);
@@ -217,7 +528,14 @@ export class WorkflowEngine {
217
528
  }
218
529
 
219
530
  const attempt = await this.executeNode(workflow, state, runDir, currentNodeId, node);
531
+ if (this.parked) {
532
+ // Do not record the aborted attempt: the projection keeps the node
533
+ // as in-flight, and resume reruns it with a fresh attempt.
534
+ throw new RunParkedError();
535
+ }
220
536
  this.recordAttempt(state, attempt);
537
+ // The terminal node event carries the output, receipt, and conversation
538
+ // linkage so the trace alone is sufficient to reconstruct the run.
221
539
  await this.persist(runDir, state, {
222
540
  scope: "node",
223
541
  type: attempt.result.outcome === "ok" ? "node_finished" : "node_failed",
@@ -226,7 +544,12 @@ export class WorkflowEngine {
226
544
  payload: {
227
545
  outcome: attempt.result.outcome,
228
546
  durationMs: attempt.result.durationMs,
547
+ ...(attempt.result.outcome === "ok" ? { output: attempt.result.output ?? null } : {}),
229
548
  ...(attempt.result.error !== undefined ? { error: attempt.result.error } : {}),
549
+ ...(attempt.execution?.action !== undefined ? { action: attempt.execution.action } : {}),
550
+ ...(attempt.execution?.conversation !== undefined
551
+ ? { conversation: attempt.execution.conversation }
552
+ : {}),
230
553
  },
231
554
  });
232
555
 
@@ -259,6 +582,9 @@ export class WorkflowEngine {
259
582
  * never interrupts a node mid-flight; it only delays the next dispatch.
260
583
  */
261
584
  private async holdWhilePaused(state: WorkflowRunState, runDir: string): Promise<void> {
585
+ if (this.parked) {
586
+ throw new RunParkedError();
587
+ }
262
588
  if (this.cancelled) {
263
589
  throw new CancelledError();
264
590
  }
@@ -267,12 +593,15 @@ export class WorkflowEngine {
267
593
  }
268
594
  state.paused = true;
269
595
  await this.persist(runDir, state, { scope: "run", type: "run_paused", payload: {} });
270
- while (this.paused && !this.cancelled) {
596
+ while (this.paused && !this.cancelled && !this.parked) {
271
597
  await new Promise<void>((resolve) => {
272
598
  this.wakePause = resolve;
273
599
  });
274
600
  }
275
601
  this.wakePause = null;
602
+ if (this.parked) {
603
+ throw new RunParkedError();
604
+ }
276
605
  delete state.paused;
277
606
  if (this.cancelled) {
278
607
  throw new CancelledError();
@@ -316,16 +645,18 @@ export class WorkflowEngine {
316
645
  outcome: attempt.result.outcome,
317
646
  startedAt: attempt.result.startedAt,
318
647
  finishedAt: attempt.result.finishedAt,
319
- promptText: attempt.execution?.promptText ?? null,
648
+ prompt: attempt.execution?.promptText ?? null,
320
649
  // `undefined` would drop the required field during JSON serialization.
321
650
  output: attempt.result.output ?? null,
322
651
  ...(attempt.result.error !== undefined ? { error: attempt.result.error } : {}),
323
652
  ...(attempt.execution?.action !== undefined ? { action: attempt.execution.action } : {}),
653
+ ...(attempt.execution?.conversation !== undefined
654
+ ? { conversation: attempt.execution.conversation }
655
+ : {}),
324
656
  };
325
657
  state.steps.push(step);
326
658
  delete state.currentNode;
327
659
  delete state.currentAttemptId;
328
- delete state.currentNodeType;
329
660
  delete state.currentNodeStartedAt;
330
661
  delete state.statusDetail;
331
662
  }
@@ -341,7 +672,6 @@ export class WorkflowEngine {
341
672
  const startedAt = new Date().toISOString();
342
673
  state.currentNode = nodeId;
343
674
  state.currentAttemptId = attemptId;
344
- state.currentNodeType = node.nodeType;
345
675
  state.currentNodeStartedAt = startedAt;
346
676
  if (node.statusDetail !== undefined) {
347
677
  state.statusDetail = node.statusDetail;
@@ -431,6 +761,11 @@ export class WorkflowEngine {
431
761
  const timeoutMs = node.timeoutMs ?? this.defaultNodeTimeoutMs;
432
762
  const abort = new AbortController();
433
763
  this.activeAbort = abort;
764
+ if (this.parked) {
765
+ // A park that landed during the node_started persist must not let the
766
+ // node dispatch: its discarded side effects would rerun on resume.
767
+ throw new RunParkedError();
768
+ }
434
769
  if (this.cancelled) {
435
770
  throw new CancelledError();
436
771
  }
@@ -571,7 +906,11 @@ export class WorkflowEngine {
571
906
  },
572
907
  signal,
573
908
  );
574
- return { output: submission.output, promptText: prompt };
909
+ return {
910
+ output: submission.output,
911
+ promptText: prompt,
912
+ ...(submission.conversation !== undefined ? { conversation: submission.conversation } : {}),
913
+ };
575
914
  }
576
915
 
577
916
  private async acceptSubmission(
@@ -629,6 +968,13 @@ export class WorkflowEngine {
629
968
  if (status === "failed" && state.status === "timed_out") {
630
969
  status = "timed_out";
631
970
  }
971
+ // Let observers (e.g. the session recorder) stop and drain before the
972
+ // terminal event exists, so the bundle is immutable from that point on.
973
+ try {
974
+ await this.onRunFinishing?.(runDir, state);
975
+ } catch {
976
+ // Finishing the run wins over observer failures.
977
+ }
632
978
  state.status = status;
633
979
  state.finishedAt = new Date().toISOString();
634
980
  if (fields.error !== undefined) {
@@ -642,7 +988,6 @@ export class WorkflowEngine {
642
988
  }
643
989
  delete state.currentNode;
644
990
  delete state.currentAttemptId;
645
- delete state.currentNodeType;
646
991
  delete state.currentNodeStartedAt;
647
992
  await this.persist(runDir, state, {
648
993
  scope: "run",
@@ -651,6 +996,7 @@ export class WorkflowEngine {
651
996
  status,
652
997
  ...(fields.error !== undefined ? { error: fields.error } : {}),
653
998
  ...(fields.waitingOn !== undefined ? { waitingOn: fields.waitingOn } : {}),
999
+ ...(fields.finalOutput !== undefined ? { finalOutput: fields.finalOutput } : {}),
654
1000
  },
655
1001
  });
656
1002
  }
@@ -771,7 +1117,7 @@ export function appendStepContract(
771
1117
  `Workflow step contract (workflow: ${workflowName}, step: ${nodeId}, attempt: ${attemptId})`,
772
1118
  "",
773
1119
  "Complete this step by calling the `workflow` tool exactly once with:",
774
- `{"step": ${JSON.stringify(nodeId)}, "attempt": ${JSON.stringify(attemptId)}, "output": <your result>}`,
1120
+ `{"action": "submit", "step": ${JSON.stringify(nodeId)}, "attempt": ${JSON.stringify(attemptId)}, "output": <your result>}`,
775
1121
  `Expected output: ${expectedOutput ?? "a JSON object with your result"}`,
776
1122
  "The step is complete only after the workflow tool accepts the output.",
777
1123
  "If the tool reports a validation error, correct the output and call it again.",
@@ -15,6 +15,51 @@ export class CancelledError extends Error {
15
15
  }
16
16
  }
17
17
 
18
+ /**
19
+ * Thrown when a runner writes to a run bundle after losing its queue claim.
20
+ * The current claim holder owns the run from that point on, so the fenced
21
+ * writer must stop touching the bundle immediately.
22
+ */
23
+ export class ClaimLostError extends Error {
24
+ readonly runId: string;
25
+
26
+ constructor(runId: string) {
27
+ super(`Workflow run claim lost: ${runId}`);
28
+ this.name = "ClaimLostError";
29
+ this.runId = runId;
30
+ }
31
+ }
32
+
33
+ /**
34
+ * Internal stop signal for close-to-park: the engine halts without writing
35
+ * a terminal event, leaving a resumable bundle for the next claim holder.
36
+ */
37
+ export class RunParkedError extends Error {
38
+ constructor() {
39
+ super("Workflow run was parked");
40
+ this.name = "RunParkedError";
41
+ }
42
+ }
43
+
44
+ export function isRunParkedError(error: unknown): error is RunParkedError {
45
+ return error instanceof RunParkedError;
46
+ }
47
+
48
+ /** The workflow source changed after the run started; resume needs force. */
49
+ export class WorkflowSourceChangedError extends Error {
50
+ readonly runId: string;
51
+
52
+ constructor(runId: string) {
53
+ super(`Workflow source changed since run ${runId} started; pass force to resume anyway`);
54
+ this.name = "WorkflowSourceChangedError";
55
+ this.runId = runId;
56
+ }
57
+ }
58
+
59
+ export function isClaimLostError(error: unknown): error is ClaimLostError {
60
+ return error instanceof ClaimLostError;
61
+ }
62
+
18
63
  export function isAbortLikeError(error: unknown): boolean {
19
64
  return error instanceof CancelledError || (error instanceof Error && error.name === "AbortError");
20
65
  }
@@ -58,11 +58,6 @@ function validateWorkflowEdge(
58
58
  outgoingEdges: Set<string>,
59
59
  ): void {
60
60
  assertKnownNode(workflow, edge.from, "Workflow edge references unknown from-node");
61
- if (workflow.nodes[edge.from]?.nodeType === "checkpoint") {
62
- // A checkpoint terminates the run as `waiting` and nothing resumes it,
63
- // so an outgoing edge would make its targets silently unreachable.
64
- throw new Error(`Workflow checkpoint node must not declare an outgoing edge: ${edge.from}`);
65
- }
66
61
  if (outgoingEdges.has(edge.from)) {
67
62
  throw new Error(`Workflow node must not declare multiple outgoing edges: ${edge.from}`);
68
63
  }
@@ -28,9 +28,19 @@ export {
28
28
  } from "./loader.js";
29
29
  export { renderShellCommand, runShellAction } from "./shell.js";
30
30
  export { sanitizeText, stripAnsi } from "./text.js";
31
+ export {
32
+ ARTIFACT_THRESHOLD_BYTES,
33
+ ArtifactWriter,
34
+ decodeValueWith,
35
+ encodeValue,
36
+ isArtifactValue,
37
+ resolveArtifacts,
38
+ } from "./artifacts.js";
31
39
  export {
32
40
  DEFINITION_SNAPSHOT_SCHEMA,
33
41
  RUN_BUNDLE_SCHEMA,
42
+ RUN_STATE_SCHEMA,
43
+ SESSION_BINDING_SCHEMA,
34
44
  TRACE_EVENT_SCHEMA,
35
45
  WorkflowRunStore,
36
46
  createDefinitionSnapshot,
@@ -47,7 +57,10 @@ export type {
47
57
  AgentStepRequest,
48
58
  AgentStepSubmission,
49
59
  ActionNodeDefinition,
60
+ ArtifactRef,
61
+ ArtifactValue,
50
62
  CheckpointNodeDefinition,
63
+ ConversationRange,
51
64
  ComputeNodeDefinition,
52
65
  FunctionActionNodeDefinition,
53
66
  MaybePromise,
@@ -70,6 +83,8 @@ export type {
70
83
  WorkflowRunResult,
71
84
  WorkflowRunState,
72
85
  WorkflowRunStatus,
86
+ WorkflowSessionBinding,
87
+ WorkflowSessionEntryRecord,
73
88
  WorkflowStepRecord,
74
89
  WorkflowTraceEvent,
75
90
  WorkflowTraceEventDraft,