@gotgenes/pi-subagents 6.0.0 → 6.1.0
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.
- package/CHANGELOG.md +23 -0
- package/docs/architecture/architecture.md +302 -210
- package/docs/plans/0098-extract-agent-record-state-machine.md +435 -0
- package/docs/plans/0102-consolidate-test-record-factory.md +176 -0
- package/docs/retro/0061-session-format-transcript.md +41 -0
- package/docs/retro/0102-consolidate-test-record-factory.md +30 -0
- package/package.json +1 -1
- package/src/agent-manager.ts +25 -50
- package/src/agent-record.ts +179 -0
- package/src/types.ts +3 -39
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
---
|
|
2
|
+
issue: 102
|
|
3
|
+
issue_title: "Consolidate test AgentRecord construction into a shared factory"
|
|
4
|
+
---
|
|
5
|
+
|
|
6
|
+
# Retro: #102 — Consolidate test AgentRecord construction into a shared factory
|
|
7
|
+
|
|
8
|
+
## Final Retrospective (2026-05-20T15:30:00Z)
|
|
9
|
+
|
|
10
|
+
### Session summary
|
|
11
|
+
|
|
12
|
+
Planned, implemented, and shipped a shared `createTestRecord()` factory in `test/helpers/make-record.ts`, migrating 7 test files from local factories and inline literals.
|
|
13
|
+
The session completed in three slash-command phases (plan → TDD → ship) with zero rework, zero test failures, and zero deviations from the plan.
|
|
14
|
+
Released as `pi-subagents-v6.0.1`.
|
|
15
|
+
|
|
16
|
+
### Observations
|
|
17
|
+
|
|
18
|
+
#### What went well
|
|
19
|
+
|
|
20
|
+
- The Explore subagent surveyed all 8 test files' factory patterns in parallel, producing a structured comparison table that directly informed the plan's default-value decisions.
|
|
21
|
+
- Before migrating `service-adapter.test.ts`, reading the `toSubagentRecord` source confirmed that `!== undefined` guards make absent-property vs. property-set-to-undefined semantically equivalent — avoiding a subtle test failure.
|
|
22
|
+
- All 4 TDD steps passed on first run, confirming the plan's migration strategy (preserve old defaults via overrides) was sound.
|
|
23
|
+
|
|
24
|
+
#### What caused friction (agent side)
|
|
25
|
+
|
|
26
|
+
No friction points — the mechanical nature of the refactoring and the well-specified plan eliminated ambiguity.
|
|
27
|
+
|
|
28
|
+
#### What caused friction (user side)
|
|
29
|
+
|
|
30
|
+
No friction points — the three-phase slash-command workflow required no manual intervention.
|
package/package.json
CHANGED
package/src/agent-manager.ts
CHANGED
|
@@ -9,10 +9,11 @@
|
|
|
9
9
|
import { randomUUID } from "node:crypto";
|
|
10
10
|
import type { Model } from "@earendil-works/pi-ai";
|
|
11
11
|
import type { AgentSession, ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
|
|
12
|
+
import { AgentRecord } from "./agent-record.js";
|
|
12
13
|
import type { AgentRunner, ToolActivity } from "./agent-runner.js";
|
|
13
14
|
import { debugLog } from "./debug.js";
|
|
14
15
|
import type { RunConfig } from "./runtime.js";
|
|
15
|
-
import type { AgentInvocation,
|
|
16
|
+
import type { AgentInvocation, IsolationMode, SubagentType, ThinkingLevel } from "./types.js";
|
|
16
17
|
import { addUsage } from "./usage.js";
|
|
17
18
|
import type { WorktreeManager } from "./worktree.js";
|
|
18
19
|
|
|
@@ -133,18 +134,15 @@ export class AgentManager {
|
|
|
133
134
|
): string {
|
|
134
135
|
const id = randomUUID().slice(0, 17);
|
|
135
136
|
const abortController = new AbortController();
|
|
136
|
-
const record
|
|
137
|
+
const record = new AgentRecord({
|
|
137
138
|
id,
|
|
138
139
|
type,
|
|
139
140
|
description: options.description,
|
|
140
141
|
status: options.isBackground ? "queued" : "running",
|
|
141
|
-
toolUses: 0,
|
|
142
142
|
startedAt: Date.now(),
|
|
143
143
|
abortController,
|
|
144
|
-
lifetimeUsage: { input: 0, output: 0, cacheWrite: 0 },
|
|
145
|
-
compactionCount: 0,
|
|
146
144
|
invocation: options.invocation,
|
|
147
|
-
};
|
|
145
|
+
});
|
|
148
146
|
this.agents.set(id, record);
|
|
149
147
|
|
|
150
148
|
const args: SpawnArgs = { pi, ctx, type, prompt, options };
|
|
@@ -184,8 +182,7 @@ export class AgentManager {
|
|
|
184
182
|
worktreeCwd = wt.path;
|
|
185
183
|
}
|
|
186
184
|
|
|
187
|
-
record.
|
|
188
|
-
record.startedAt = Date.now();
|
|
185
|
+
record.markRunning(Date.now());
|
|
189
186
|
if (options.isBackground) this.runningBackground++;
|
|
190
187
|
this.onStart?.(record);
|
|
191
188
|
|
|
@@ -244,27 +241,26 @@ export class AgentManager {
|
|
|
244
241
|
},
|
|
245
242
|
})
|
|
246
243
|
.then(({ responseText, session, aborted, steered, sessionFile }) => {
|
|
247
|
-
// Don't overwrite status if externally stopped via abort()
|
|
248
|
-
if (record.status !== "stopped") {
|
|
249
|
-
record.status = aborted ? "aborted" : steered ? "steered" : "completed";
|
|
250
|
-
}
|
|
251
|
-
record.result = responseText;
|
|
252
|
-
record.session = session;
|
|
253
|
-
record.completedAt ??= Date.now();
|
|
254
|
-
if (sessionFile) record.outputFile = sessionFile;
|
|
255
|
-
|
|
256
244
|
detach();
|
|
257
245
|
|
|
258
|
-
// Clean up worktree
|
|
246
|
+
// Clean up worktree before transition so the final result includes branch text
|
|
247
|
+
let finalResult = responseText;
|
|
259
248
|
if (record.worktree) {
|
|
260
249
|
const wtResult = this.worktrees.cleanup(record.worktree, options.description);
|
|
261
250
|
record.worktreeResult = wtResult;
|
|
262
251
|
if (wtResult.hasChanges && wtResult.branch) {
|
|
263
|
-
|
|
264
|
-
`\n\n---\nChanges saved to branch \`${wtResult.branch}\`. Merge with: \`git merge ${wtResult.branch}\``;
|
|
252
|
+
finalResult += `\n\n---\nChanges saved to branch \`${wtResult.branch}\`. Merge with: \`git merge ${wtResult.branch}\``;
|
|
265
253
|
}
|
|
266
254
|
}
|
|
267
255
|
|
|
256
|
+
// Transition — guards against overwriting externally-stopped status
|
|
257
|
+
if (aborted) record.markAborted(finalResult);
|
|
258
|
+
else if (steered) record.markSteered(finalResult);
|
|
259
|
+
else record.markCompleted(finalResult);
|
|
260
|
+
|
|
261
|
+
record.session = session;
|
|
262
|
+
if (sessionFile) record.outputFile = sessionFile;
|
|
263
|
+
|
|
268
264
|
if (options.isBackground) {
|
|
269
265
|
this.runningBackground--;
|
|
270
266
|
try { this.onComplete?.(record); } catch (err) { debugLog("onComplete callback", err); }
|
|
@@ -273,17 +269,10 @@ export class AgentManager {
|
|
|
273
269
|
return responseText;
|
|
274
270
|
})
|
|
275
271
|
.catch((err) => {
|
|
276
|
-
|
|
277
|
-
if (record.status !== "stopped") {
|
|
278
|
-
record.status = "error";
|
|
279
|
-
}
|
|
280
|
-
record.error = err instanceof Error ? err.message : String(err);
|
|
281
|
-
record.completedAt ??= Date.now();
|
|
272
|
+
record.markError(err);
|
|
282
273
|
|
|
283
274
|
detach();
|
|
284
275
|
|
|
285
|
-
|
|
286
|
-
|
|
287
276
|
// Best-effort worktree cleanup on error
|
|
288
277
|
if (record.worktree) {
|
|
289
278
|
try {
|
|
@@ -314,9 +303,7 @@ export class AgentManager {
|
|
|
314
303
|
} catch (err) {
|
|
315
304
|
// Late failure (e.g. strict worktree-isolation) — surface on the record
|
|
316
305
|
// so the user/agent can see it via /agents, then keep draining.
|
|
317
|
-
record.
|
|
318
|
-
record.error = err instanceof Error ? err.message : String(err);
|
|
319
|
-
record.completedAt = Date.now();
|
|
306
|
+
record.markError(err);
|
|
320
307
|
this.onComplete?.(record);
|
|
321
308
|
}
|
|
322
309
|
}
|
|
@@ -350,11 +337,7 @@ export class AgentManager {
|
|
|
350
337
|
const record = this.agents.get(id);
|
|
351
338
|
if (!record?.session) return undefined;
|
|
352
339
|
|
|
353
|
-
record.
|
|
354
|
-
record.startedAt = Date.now();
|
|
355
|
-
record.completedAt = undefined;
|
|
356
|
-
record.result = undefined;
|
|
357
|
-
record.error = undefined;
|
|
340
|
+
record.resetForResume(Date.now());
|
|
358
341
|
|
|
359
342
|
try {
|
|
360
343
|
const responseText = await this.runner.resume(record.session, prompt, {
|
|
@@ -370,13 +353,9 @@ export class AgentManager {
|
|
|
370
353
|
},
|
|
371
354
|
signal,
|
|
372
355
|
});
|
|
373
|
-
record.
|
|
374
|
-
record.result = responseText;
|
|
375
|
-
record.completedAt = Date.now();
|
|
356
|
+
record.markCompleted(responseText);
|
|
376
357
|
} catch (err) {
|
|
377
|
-
record.
|
|
378
|
-
record.error = err instanceof Error ? err.message : String(err);
|
|
379
|
-
record.completedAt = Date.now();
|
|
358
|
+
record.markError(err);
|
|
380
359
|
}
|
|
381
360
|
|
|
382
361
|
return record;
|
|
@@ -399,15 +378,13 @@ export class AgentManager {
|
|
|
399
378
|
// Remove from queue if queued
|
|
400
379
|
if (record.status === "queued") {
|
|
401
380
|
this.queue = this.queue.filter(q => q.id !== id);
|
|
402
|
-
record.
|
|
403
|
-
record.completedAt = Date.now();
|
|
381
|
+
record.markStopped();
|
|
404
382
|
return true;
|
|
405
383
|
}
|
|
406
384
|
|
|
407
385
|
if (record.status !== "running") return false;
|
|
408
386
|
record.abortController?.abort();
|
|
409
|
-
record.
|
|
410
|
-
record.completedAt = Date.now();
|
|
387
|
+
record.markStopped();
|
|
411
388
|
return true;
|
|
412
389
|
}
|
|
413
390
|
|
|
@@ -452,8 +429,7 @@ export class AgentManager {
|
|
|
452
429
|
for (const queued of this.queue) {
|
|
453
430
|
const record = this.agents.get(queued.id);
|
|
454
431
|
if (record) {
|
|
455
|
-
record.
|
|
456
|
-
record.completedAt = Date.now();
|
|
432
|
+
record.markStopped();
|
|
457
433
|
count++;
|
|
458
434
|
}
|
|
459
435
|
}
|
|
@@ -462,8 +438,7 @@ export class AgentManager {
|
|
|
462
438
|
for (const record of this.agents.values()) {
|
|
463
439
|
if (record.status === "running") {
|
|
464
440
|
record.abortController?.abort();
|
|
465
|
-
record.
|
|
466
|
-
record.completedAt = Date.now();
|
|
441
|
+
record.markStopped();
|
|
467
442
|
count++;
|
|
468
443
|
}
|
|
469
444
|
}
|
|
@@ -0,0 +1,179 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* agent-record.ts — AgentRecord class with encapsulated status-transition logic.
|
|
3
|
+
*
|
|
4
|
+
* Status transitions (status, result, error, startedAt, completedAt) are owned
|
|
5
|
+
* by the class and exposed via transition methods. External code reads these
|
|
6
|
+
* fields through public properties but cannot write them directly.
|
|
7
|
+
*
|
|
8
|
+
* Non-transition state (session, toolUses, lifetimeUsage, etc.) remains public.
|
|
9
|
+
*/
|
|
10
|
+
|
|
11
|
+
import type { AgentSession } from "@earendil-works/pi-coding-agent";
|
|
12
|
+
import type { AgentInvocation, SubagentType } from "./types.js";
|
|
13
|
+
import type { LifetimeUsage } from "./usage.js";
|
|
14
|
+
|
|
15
|
+
export type AgentRecordStatus =
|
|
16
|
+
| "queued"
|
|
17
|
+
| "running"
|
|
18
|
+
| "completed"
|
|
19
|
+
| "steered"
|
|
20
|
+
| "aborted"
|
|
21
|
+
| "stopped"
|
|
22
|
+
| "error";
|
|
23
|
+
|
|
24
|
+
export interface AgentRecordInit {
|
|
25
|
+
id: string;
|
|
26
|
+
type: SubagentType;
|
|
27
|
+
description: string;
|
|
28
|
+
status?: AgentRecordStatus;
|
|
29
|
+
startedAt?: number;
|
|
30
|
+
completedAt?: number;
|
|
31
|
+
result?: string;
|
|
32
|
+
error?: string;
|
|
33
|
+
toolUses?: number;
|
|
34
|
+
lifetimeUsage?: LifetimeUsage;
|
|
35
|
+
compactionCount?: number;
|
|
36
|
+
abortController?: AbortController;
|
|
37
|
+
invocation?: AgentInvocation;
|
|
38
|
+
session?: AgentSession;
|
|
39
|
+
promise?: Promise<string>;
|
|
40
|
+
resultConsumed?: boolean;
|
|
41
|
+
pendingSteers?: string[];
|
|
42
|
+
worktree?: { path: string; branch: string };
|
|
43
|
+
worktreeResult?: { hasChanges: boolean; branch?: string };
|
|
44
|
+
toolCallId?: string;
|
|
45
|
+
outputFile?: string;
|
|
46
|
+
}
|
|
47
|
+
|
|
48
|
+
export class AgentRecord {
|
|
49
|
+
// Identity — set once at construction
|
|
50
|
+
readonly id: string;
|
|
51
|
+
readonly type: SubagentType;
|
|
52
|
+
readonly description: string;
|
|
53
|
+
readonly invocation?: AgentInvocation;
|
|
54
|
+
|
|
55
|
+
// Transition state — encapsulated behind getters, mutated only via transition methods
|
|
56
|
+
private _status: AgentRecordStatus;
|
|
57
|
+
get status(): AgentRecordStatus { return this._status; }
|
|
58
|
+
|
|
59
|
+
private _result?: string;
|
|
60
|
+
get result(): string | undefined { return this._result; }
|
|
61
|
+
|
|
62
|
+
private _error?: string;
|
|
63
|
+
get error(): string | undefined { return this._error; }
|
|
64
|
+
|
|
65
|
+
private _startedAt: number;
|
|
66
|
+
get startedAt(): number { return this._startedAt; }
|
|
67
|
+
|
|
68
|
+
private _completedAt?: number;
|
|
69
|
+
get completedAt(): number | undefined { return this._completedAt; }
|
|
70
|
+
|
|
71
|
+
// Non-transition mutable state
|
|
72
|
+
toolUses: number;
|
|
73
|
+
lifetimeUsage: LifetimeUsage;
|
|
74
|
+
compactionCount: number;
|
|
75
|
+
session?: AgentSession;
|
|
76
|
+
abortController?: AbortController;
|
|
77
|
+
promise?: Promise<string>;
|
|
78
|
+
resultConsumed?: boolean;
|
|
79
|
+
pendingSteers?: string[];
|
|
80
|
+
worktree?: { path: string; branch: string };
|
|
81
|
+
worktreeResult?: { hasChanges: boolean; branch?: string };
|
|
82
|
+
toolCallId?: string;
|
|
83
|
+
outputFile?: string;
|
|
84
|
+
|
|
85
|
+
constructor(init: AgentRecordInit) {
|
|
86
|
+
this.id = init.id;
|
|
87
|
+
this.type = init.type;
|
|
88
|
+
this.description = init.description;
|
|
89
|
+
this.invocation = init.invocation;
|
|
90
|
+
|
|
91
|
+
this._status = init.status ?? "queued";
|
|
92
|
+
this._result = init.result;
|
|
93
|
+
this._error = init.error;
|
|
94
|
+
this._startedAt = init.startedAt ?? Date.now();
|
|
95
|
+
this._completedAt = init.completedAt;
|
|
96
|
+
|
|
97
|
+
this.toolUses = init.toolUses ?? 0;
|
|
98
|
+
this.lifetimeUsage = init.lifetimeUsage ?? { input: 0, output: 0, cacheWrite: 0 };
|
|
99
|
+
this.compactionCount = init.compactionCount ?? 0;
|
|
100
|
+
this.abortController = init.abortController;
|
|
101
|
+
this.session = init.session;
|
|
102
|
+
this.promise = init.promise;
|
|
103
|
+
this.resultConsumed = init.resultConsumed;
|
|
104
|
+
this.pendingSteers = init.pendingSteers;
|
|
105
|
+
this.worktree = init.worktree;
|
|
106
|
+
this.worktreeResult = init.worktreeResult;
|
|
107
|
+
this.toolCallId = init.toolCallId;
|
|
108
|
+
this.outputFile = init.outputFile;
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
/** Transition to running state. Sets status and startedAt. */
|
|
112
|
+
markRunning(startedAt: number): void {
|
|
113
|
+
this._status = "running";
|
|
114
|
+
this._startedAt = startedAt;
|
|
115
|
+
}
|
|
116
|
+
|
|
117
|
+
/**
|
|
118
|
+
* Transition to completed state.
|
|
119
|
+
* Always sets result and completedAt (??=). Only changes status if not stopped.
|
|
120
|
+
*/
|
|
121
|
+
markCompleted(result: string, completedAt?: number): void {
|
|
122
|
+
this._result = result;
|
|
123
|
+
this._completedAt ??= completedAt ?? Date.now();
|
|
124
|
+
if (this._status !== "stopped") {
|
|
125
|
+
this._status = "completed";
|
|
126
|
+
}
|
|
127
|
+
}
|
|
128
|
+
|
|
129
|
+
/**
|
|
130
|
+
* Transition to aborted state.
|
|
131
|
+
* Always sets result and completedAt (??=). Only changes status if not stopped.
|
|
132
|
+
*/
|
|
133
|
+
markAborted(result: string, completedAt?: number): void {
|
|
134
|
+
this._result = result;
|
|
135
|
+
this._completedAt ??= completedAt ?? Date.now();
|
|
136
|
+
if (this._status !== "stopped") {
|
|
137
|
+
this._status = "aborted";
|
|
138
|
+
}
|
|
139
|
+
}
|
|
140
|
+
|
|
141
|
+
/**
|
|
142
|
+
* Transition to steered state.
|
|
143
|
+
* Always sets result and completedAt (??=). Only changes status if not stopped.
|
|
144
|
+
*/
|
|
145
|
+
markSteered(result: string, completedAt?: number): void {
|
|
146
|
+
this._result = result;
|
|
147
|
+
this._completedAt ??= completedAt ?? Date.now();
|
|
148
|
+
if (this._status !== "stopped") {
|
|
149
|
+
this._status = "steered";
|
|
150
|
+
}
|
|
151
|
+
}
|
|
152
|
+
|
|
153
|
+
/**
|
|
154
|
+
* Transition to error state.
|
|
155
|
+
* Always sets error (formatted) and completedAt (??=). Only changes status if not stopped.
|
|
156
|
+
*/
|
|
157
|
+
markError(error: unknown, completedAt?: number): void {
|
|
158
|
+
this._error = error instanceof Error ? error.message : String(error);
|
|
159
|
+
this._completedAt ??= completedAt ?? Date.now();
|
|
160
|
+
if (this._status !== "stopped") {
|
|
161
|
+
this._status = "error";
|
|
162
|
+
}
|
|
163
|
+
}
|
|
164
|
+
|
|
165
|
+
/** Transition to stopped state. Always valid — no guard. */
|
|
166
|
+
markStopped(completedAt?: number): void {
|
|
167
|
+
this._status = "stopped";
|
|
168
|
+
this._completedAt = completedAt ?? Date.now();
|
|
169
|
+
}
|
|
170
|
+
|
|
171
|
+
/** Reset for resume: running status, new startedAt, clear completedAt/result/error. */
|
|
172
|
+
resetForResume(startedAt: number): void {
|
|
173
|
+
this._status = "running";
|
|
174
|
+
this._startedAt = startedAt;
|
|
175
|
+
this._completedAt = undefined;
|
|
176
|
+
this._result = undefined;
|
|
177
|
+
this._error = undefined;
|
|
178
|
+
}
|
|
179
|
+
}
|
package/src/types.ts
CHANGED
|
@@ -3,9 +3,10 @@
|
|
|
3
3
|
*/
|
|
4
4
|
|
|
5
5
|
import type { ThinkingLevel } from "@earendil-works/pi-ai";
|
|
6
|
-
import type { AgentSession } from "@earendil-works/pi-coding-agent";
|
|
7
|
-
import type { LifetimeUsage } from "./usage.js";
|
|
8
6
|
|
|
7
|
+
export type { AgentRecordInit, AgentRecordStatus } from "./agent-record.js";
|
|
8
|
+
|
|
9
|
+
export { AgentRecord } from "./agent-record.js";
|
|
9
10
|
export type { ThinkingLevel };
|
|
10
11
|
|
|
11
12
|
/** Agent type: any string name (built-in defaults or user-defined). */
|
|
@@ -55,43 +56,6 @@ export interface AgentConfig {
|
|
|
55
56
|
source?: "default" | "project" | "global";
|
|
56
57
|
}
|
|
57
58
|
|
|
58
|
-
export interface AgentRecord {
|
|
59
|
-
id: string;
|
|
60
|
-
type: SubagentType;
|
|
61
|
-
description: string;
|
|
62
|
-
status: "queued" | "running" | "completed" | "steered" | "aborted" | "stopped" | "error";
|
|
63
|
-
result?: string;
|
|
64
|
-
error?: string;
|
|
65
|
-
toolUses: number;
|
|
66
|
-
startedAt: number;
|
|
67
|
-
completedAt?: number;
|
|
68
|
-
session?: AgentSession;
|
|
69
|
-
abortController?: AbortController;
|
|
70
|
-
promise?: Promise<string>;
|
|
71
|
-
/** Set when result was already consumed via get_subagent_result — suppresses completion notification. */
|
|
72
|
-
resultConsumed?: boolean;
|
|
73
|
-
/** Steering messages queued before the session was ready. */
|
|
74
|
-
pendingSteers?: string[];
|
|
75
|
-
/** Worktree info if the agent is running in an isolated worktree. */
|
|
76
|
-
worktree?: { path: string; branch: string };
|
|
77
|
-
/** Worktree cleanup result after agent completion. */
|
|
78
|
-
worktreeResult?: { hasChanges: boolean; branch?: string };
|
|
79
|
-
/** The tool_use_id from the original Agent tool call. */
|
|
80
|
-
toolCallId?: string;
|
|
81
|
-
/** Path to the persisted session transcript file. */
|
|
82
|
-
outputFile?: string;
|
|
83
|
-
/**
|
|
84
|
-
* Lifetime usage breakdown, accumulated via `message_end` events. Survives
|
|
85
|
-
* compaction. Total = input + output + cacheWrite (cacheRead deliberately
|
|
86
|
-
* excluded — see issue #38). Initialized to zeros at spawn.
|
|
87
|
-
*/
|
|
88
|
-
lifetimeUsage: LifetimeUsage;
|
|
89
|
-
/** Number of times this agent's session has compacted. Initialized to 0 at spawn. */
|
|
90
|
-
compactionCount: number;
|
|
91
|
-
/** Resolved spawn params, captured for UI display. Fixed at spawn time. */
|
|
92
|
-
invocation?: AgentInvocation;
|
|
93
|
-
}
|
|
94
|
-
|
|
95
59
|
export interface AgentInvocation {
|
|
96
60
|
/** Short display name, e.g. "haiku" — only set when different from parent. */
|
|
97
61
|
modelName?: string;
|