@indigoai-us/hq-cli 5.345.73 → 5.345.74
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 +8 -0
- package/dist/commands/lanes-loop.js +23 -2
- package/dist/commands/lanes-resume.d.ts +2 -1
- package/dist/commands/lanes-resume.js +13 -2
- package/dist/commands/lanes-shared.d.ts +2 -0
- package/dist/commands/lanes.js +17 -1
- package/dist/lib/lanes/flag-keys.d.ts +2 -0
- package/dist/lib/lanes/flag-keys.js +2 -0
- package/dist/lib/lanes/loop.d.ts +2 -0
- package/dist/lib/lanes/loop.js +52 -14
- package/dist/lib/lanes/reconcile.js +8 -2
- package/dist/lib/lanes/spawn.d.ts +3 -1
- package/dist/lib/lanes/spawn.js +122 -2
- package/dist/lib/lanes/types.d.ts +4 -0
- package/dist/lib/lanes/watcher.d.ts +1 -12
- package/dist/lib/lanes/watcher.js +53 -5
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -2,6 +2,14 @@
|
|
|
2
2
|
|
|
3
3
|
## [Unreleased]
|
|
4
4
|
|
|
5
|
+
## [5.345.74] — 2026-10-10
|
|
6
|
+
|
|
7
|
+
- Lane envelope repair prompts now identify parser locations and schema fields,
|
|
8
|
+
include a minimal valid envelope, and keep `next_check_at` optional. Section 8
|
|
9
|
+
loop workers receive explicit handoff enums and one same-session repair turn;
|
|
10
|
+
failed handoff summaries retain the validation error. Both behaviors have
|
|
11
|
+
default-on hq-flags kill switches.
|
|
12
|
+
|
|
5
13
|
## [5.345.73] — 2026-10-10
|
|
6
14
|
- Fixed company resolution so a minimal entity liveness response is not treated as a full company record.
|
|
7
15
|
|
|
@@ -9,6 +9,8 @@ import { raiseLaneQuestion } from "../lib/lanes/questions.js";
|
|
|
9
9
|
import { isTerminalLaneState } from "../lib/lanes/types.js";
|
|
10
10
|
import { clearLoopControllerPid, loopAdmissionPath, runLoopLane, setLoopControllerPid } from "../lib/lanes/loop.js";
|
|
11
11
|
import { laneStories } from "../lib/lanes/stories.js";
|
|
12
|
+
import { readLanesRegistryFlagFailOpen } from "../lib/lanes/registry-flag.js";
|
|
13
|
+
import { LANE_LOOP_HANDOFF_REPAIR_FLAG } from "../lib/lanes/flag-keys.js";
|
|
12
14
|
const LOOP_RUNS_REL = path.join("workspace", "lanes-runs");
|
|
13
15
|
function loopRunDir(hqRoot, laneId) {
|
|
14
16
|
return path.join(hqRoot, LOOP_RUNS_REL, laneId);
|
|
@@ -34,6 +36,11 @@ export async function runLaneLoopController(ctx, laneId, deps) {
|
|
|
34
36
|
if (lane.loop_mode !== true)
|
|
35
37
|
throw new LanesError("lane_not_loop_mode", `Lane ${laneId} is not a loop lane`, { lane_id: laneId });
|
|
36
38
|
const runDir = loopRunDir(hqRoot, laneId);
|
|
39
|
+
const handoffRepairEnabled = await readLanesRegistryFlagFailOpen(LANE_LOOP_HANDOFF_REPAIR_FLAG, {
|
|
40
|
+
company: lane.company ?? lane.mesh_company,
|
|
41
|
+
defaultValue: true,
|
|
42
|
+
env: ctx.env,
|
|
43
|
+
});
|
|
37
44
|
let initialAdmission = ctx.env.HQ_LANE_LOOP_INITIAL === "1";
|
|
38
45
|
const stop = new AbortController();
|
|
39
46
|
const onSignal = () => stop.abort();
|
|
@@ -43,9 +50,10 @@ export async function runLaneLoopController(ctx, laneId, deps) {
|
|
|
43
50
|
const result = await runLoopLane({
|
|
44
51
|
lane,
|
|
45
52
|
runDir,
|
|
53
|
+
repairInvalidHandoff: handoffRepairEnabled,
|
|
46
54
|
signal: stop.signal,
|
|
47
55
|
idlePollMs: 250,
|
|
48
|
-
execute: async ({ envelope, signal, fresh, sessionId }) => {
|
|
56
|
+
execute: async ({ envelope, signal, fresh, sessionId, repairError }) => {
|
|
49
57
|
let current = readLane(hqRoot, laneId);
|
|
50
58
|
const data = envelope;
|
|
51
59
|
const storyId = typeof data.story_id === "string" && data.story_id.trim() ? data.story_id.trim() : current.story_id;
|
|
@@ -88,9 +96,21 @@ export async function runLaneLoopController(ctx, laneId, deps) {
|
|
|
88
96
|
const prompt = typeof data.prompt === "string" ? data.prompt :
|
|
89
97
|
`${phaseContext}\n\nPhase envelope:\n${JSON.stringify(envelope, null, 2)}`;
|
|
90
98
|
const briefPath = path.join(runDir, "brief.md");
|
|
99
|
+
const repairInstructions = repairError && data.schema === "hq-phase-envelope/v1"
|
|
100
|
+
? [
|
|
101
|
+
"HANDOFF REPAIR TURN: the previous handoff failed validation.",
|
|
102
|
+
`Exact validation error: ${repairError}`,
|
|
103
|
+
"Return a corrected hq-phase-handoff/v1 object. Preserve every valid supplied value and extension field.",
|
|
104
|
+
"Allowed status values: passed, failed, blocked. Allowed back_pressure values for tests, lint, typecheck, and build: pass, fail, skip.",
|
|
105
|
+
`Minimal valid shape: ${JSON.stringify({ schema: "hq-phase-handoff/v1", story_id: String(data.story_id), phase: String(data.phase), worker_id: String(data.worker_id), status: "passed", summary: "...", files_changed: [], commits: [], back_pressure: { tests: "skip", lint: "skip", typecheck: "skip", build: "skip" }, context_for_next: "..." }, null, 2)}`,
|
|
106
|
+
].join("\n")
|
|
107
|
+
: "";
|
|
91
108
|
fs.writeFileSync(briefPath, [
|
|
92
109
|
"Follow the lane worker instructions and complete this single bounded phase.",
|
|
93
|
-
|
|
110
|
+
handoffRepairEnabled
|
|
111
|
+
? "Return the standard hq-lane-envelope.v1 envelope when the phase is complete. For a section 8 phase envelope, put a complete hq-phase-handoff/v1 object in artifacts. The lane envelope decision uses done/ask/blocked; handoff status uses passed/failed/blocked. Allowed handoff status values are passed, failed, blocked. Allowed back_pressure values for tests, lint, typecheck, and build are pass, fail, skip. Include story_id, phase, worker_id, summary, files_changed, commits, back_pressure, and context_for_next."
|
|
112
|
+
: "Return the standard hq-lane-envelope.v1 envelope when the phase is complete. For a section 8 phase envelope, put a complete hq-phase-handoff/v1 object in artifacts; it must identify the requested story_id and phase and include worker_id, status, summary, files_changed, commits, back_pressure (tests/lint/typecheck/build), and context_for_next.",
|
|
113
|
+
repairInstructions,
|
|
94
114
|
"",
|
|
95
115
|
prompt,
|
|
96
116
|
].join("\n"));
|
|
@@ -137,6 +157,7 @@ export async function runLaneLoopController(ctx, laneId, deps) {
|
|
|
137
157
|
story: storyId,
|
|
138
158
|
freshThread: fresh,
|
|
139
159
|
...(sessionId ? { loopSessionId: sessionId } : {}),
|
|
160
|
+
...(repairError !== undefined ? { loopHandoffRepair: true } : {}),
|
|
140
161
|
workingDirectory: worktree,
|
|
141
162
|
});
|
|
142
163
|
}
|
|
@@ -1,4 +1,5 @@
|
|
|
1
|
-
import { type Ctx, type LaneBalanceMoveTargetOptions, type LaneResumeHandlerDeps, type LaneResumeOptions } from "./lanes-shared.js";
|
|
1
|
+
import { type Ctx, type LaneBalanceMoveTargetOptions, type LaneRecord, type LaneResumeHandlerDeps, type LaneResumeOptions } from "./lanes-shared.js";
|
|
2
|
+
export declare function isLoopHandoffRepairResume(lane: Pick<LaneRecord, "loop_mode" | "state" | "provider_session_id">, opts: Pick<LaneResumeOptions, "loopHandoffRepair" | "loopSessionId">): boolean;
|
|
2
3
|
export declare function createLaneResumeHandlers(deps: LaneResumeHandlerDeps): {
|
|
3
4
|
runClaudeBalanceMoveTarget: (ctx: Ctx, laneId: string, opts: LaneBalanceMoveTargetOptions) => Promise<void>;
|
|
4
5
|
runResume: (ctx: Ctx, laneId: string, opts: LaneResumeOptions) => Promise<void>;
|
|
@@ -10,6 +10,13 @@ import { markSpawnFailure, spawnLaneWorker } from "../lib/lanes/spawn.js";
|
|
|
10
10
|
import { appendProjectFocusOverride, removeProjectFocusOverride } from "../lib/lanes/story-ledger.js";
|
|
11
11
|
import { appendLaneFocusOverride, makeFocusOverride } from "../lib/lanes/focus-guard.js";
|
|
12
12
|
import { accountMoveSummary, accountRequestKeepsCurrentLane, assertCurrentResumeAccountAvailable, hasKeepAliveProgress, recordedResumeEffort, resetKeepAliveProgress, resolveResumeEffort } from "./lanes-shared.js";
|
|
13
|
+
export function isLoopHandoffRepairResume(lane, opts) {
|
|
14
|
+
return opts.loopHandoffRepair === true &&
|
|
15
|
+
lane.loop_mode === true &&
|
|
16
|
+
lane.state === "awaiting_input" &&
|
|
17
|
+
opts.loopSessionId !== undefined &&
|
|
18
|
+
opts.loopSessionId === lane.provider_session_id;
|
|
19
|
+
}
|
|
13
20
|
export function createLaneResumeHandlers(deps) {
|
|
14
21
|
const { archiveResumeMessage, assertLaneRunsHereForAdmission, assertLaneRunsHereForOperatorMutation, assessLaneAdmission, automaticBalanceMoveHistoryReason, authorizeFailedResume, authorizeResumeProfileRepin, casUpdateLane, dispatchOutpostLaneCommand, laneIsHostedHere, laneWithRecoveredProviderSession, mintMessageId, persistClaudeAccountSelection, prepareAccountMove, printJson, recordResumeSetup, resolveResumeProfile, selectAutomaticClaudeResumeAccount, validateResumeProfileModel, withResumeHireAdmission } = deps;
|
|
15
22
|
async function runClaudeBalanceMoveTarget(ctx, laneId, opts) {
|
|
@@ -156,6 +163,10 @@ export function createLaneResumeHandlers(deps) {
|
|
|
156
163
|
throw new LanesError("effort_resume_unsupported", `Provider ${lane.provider} cannot change effort on a resumed session`, { lane_id: laneId, provider: lane.provider });
|
|
157
164
|
}
|
|
158
165
|
lane = laneWithRecoveredProviderSession(hqRoot, lane);
|
|
166
|
+
const loopHandoffRepair = isLoopHandoffRepairResume(lane, opts);
|
|
167
|
+
if (opts.loopHandoffRepair === true && !loopHandoffRepair) {
|
|
168
|
+
throw new LanesError("loop_handoff_repair_session_mismatch", "Loop handoff repair requires the awaiting-input loop lane's current provider session", { lane_id: laneId, state: lane.state });
|
|
169
|
+
}
|
|
159
170
|
if (opts.loopSessionId !== undefined && opts.loopSessionId !== lane.provider_session_id) {
|
|
160
171
|
lane = casUpdateLane(hqRoot, laneId, (current) => ({ ...current, provider_session_id: opts.loopSessionId }));
|
|
161
172
|
}
|
|
@@ -192,7 +203,7 @@ export function createLaneResumeHandlers(deps) {
|
|
|
192
203
|
const pendingQuestions = lane.state === "awaiting_input"
|
|
193
204
|
? await deps.pendingQuestionsForLane(ctx, lane)
|
|
194
205
|
: [];
|
|
195
|
-
if (pendingQuestions.length > 0) {
|
|
206
|
+
if (pendingQuestions.length > 0 && !loopHandoffRepair) {
|
|
196
207
|
const companySlug = lane.company?.slug ?? lane.mesh_company ?? "<company-slug>";
|
|
197
208
|
const instructions = pendingQuestions.map(({ question_id, audience }) => `question ${question_id} awaits ${audience.kind}:${audience.id}; answer it with hq lanes questions answer ${question_id} --company ${companySlug} --freeform <answer>`);
|
|
198
209
|
const message = `Lane ${laneId} is awaiting input. ${instructions.join("; ")}`;
|
|
@@ -632,7 +643,7 @@ export function createLaneResumeHandlers(deps) {
|
|
|
632
643
|
kind: "resume",
|
|
633
644
|
from: resumedFrom,
|
|
634
645
|
...(opts.freshThread === true ? { fresh: true } : {}),
|
|
635
|
-
...(resumedFrom === "awaiting_input" && pendingQuestions.length === 0 ? { allowAwaitingInput: true } : {}),
|
|
646
|
+
...(resumedFrom === "awaiting_input" && (pendingQuestions.length === 0 || loopHandoffRepair) ? { allowAwaitingInput: true } : {}),
|
|
636
647
|
...(opts.fromFailed === true ? { fromFailed: true } : {}),
|
|
637
648
|
...(failedResumeExpectedSenior !== undefined
|
|
638
649
|
? { expectedSenior: failedResumeExpectedSenior }
|
|
@@ -81,6 +81,8 @@ export interface LaneResumeOptions {
|
|
|
81
81
|
workingDirectory?: string;
|
|
82
82
|
/** Internal loop runner session; selects this story's saved provider session. */
|
|
83
83
|
loopSessionId?: string;
|
|
84
|
+
/** Allows only a section 8 handoff repair to continue a loop lane with a pending question. */
|
|
85
|
+
loopHandoffRepair?: boolean;
|
|
84
86
|
meshCompany?: string;
|
|
85
87
|
profileRepinContext?: string;
|
|
86
88
|
grantContext?: string;
|
package/dist/commands/lanes.js
CHANGED
|
@@ -42,7 +42,7 @@ import { readLanesFlagLocalOverride, readLanesRegistryFlag, readLanesRegistryFla
|
|
|
42
42
|
import { readLaneStatusRowFields } from "../lib/lanes/status-fields.js";
|
|
43
43
|
import { appendProjectFocusOverride, readProjectStoryFocusData, removeProjectFocusOverride } from "../lib/lanes/story-ledger.js";
|
|
44
44
|
import { appendLaneFocusOverride, assertFocusGuardAllows, makeFocusOverride } from "../lib/lanes/focus-guard.js";
|
|
45
|
-
import { LANE_FOCUS_GUARD_FLAG, LANE_FOCUS_DASHBOARD_FLAG, LANE_HIRE_GRANT_FINISHED_LANES_FLAG, LANE_LIST_SKIP_REMOTE_REFRESH_FLAG, LANE_MESSAGE_ROUTE_OWNER_HOST_FLAG, LANE_ORPHANED_BARE_ID_DROPBOX_REPAIR_FLAG, LANE_PROFILE_PIN_EXHAUSTED_FALLBACK_FLAG, LANE_STATUS_ROW_FIELDS_FLAG, LANE_REAP_STOP_FLAG, LANE_REUSE_REFRESHES_PROFILE_FLAG, LANE_WORKER_CONTINUITY_FLAG, LANE_CONDUCT_LOOP_FLAG, } from "../lib/lanes/flag-keys.js";
|
|
45
|
+
import { LANE_FOCUS_GUARD_FLAG, LANE_FOCUS_DASHBOARD_FLAG, LANE_HIRE_GRANT_FINISHED_LANES_FLAG, LANE_LIST_SKIP_REMOTE_REFRESH_FLAG, LANE_MESSAGE_ROUTE_OWNER_HOST_FLAG, LANE_ORPHANED_BARE_ID_DROPBOX_REPAIR_FLAG, LANE_PROFILE_PIN_EXHAUSTED_FALLBACK_FLAG, LANE_STATUS_ROW_FIELDS_FLAG, LANE_REAP_STOP_FLAG, LANE_REUSE_REFRESHES_PROFILE_FLAG, LANE_WORKER_CONTINUITY_FLAG, LANE_CONDUCT_LOOP_FLAG, LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG, } from "../lib/lanes/flag-keys.js";
|
|
46
46
|
import { findWorkerContinuityLane, sameLaneHost, workerContinuityPrompt } from "../lib/lanes/worker-continuity.js";
|
|
47
47
|
import { enqueueLoopEnvelope, interruptLoopQueue, interruptLoopQueueForStory, loopControllerIsAlive, loopStatus, validateLoopEnvelope } from "../lib/lanes/loop.js";
|
|
48
48
|
import { runLaneLoopController, startLaneLoopController } from "./lanes-loop.js";
|
|
@@ -10021,8 +10021,24 @@ export function registerLanesCommand(parent, deps = {}) {
|
|
|
10021
10021
|
catch (err) {
|
|
10022
10022
|
throw new LanesError("repair_request_invalid", `Envelope repair request is invalid JSON: ${err instanceof Error ? err.message : String(err)}`, {});
|
|
10023
10023
|
}
|
|
10024
|
+
const laneId = payload && typeof payload === "object" && !Array.isArray(payload) && typeof payload.lane_id === "string"
|
|
10025
|
+
? payload.lane_id
|
|
10026
|
+
: undefined;
|
|
10027
|
+
const company = laneId ? (() => { try {
|
|
10028
|
+
const lane = readLane(ctx.hqRoot(), laneId);
|
|
10029
|
+
return lane.company ?? lane.mesh_company;
|
|
10030
|
+
}
|
|
10031
|
+
catch {
|
|
10032
|
+
return undefined;
|
|
10033
|
+
} })() : undefined;
|
|
10034
|
+
const envelopeRepairGuidanceEnabled = await readLanesRegistryFlagFailOpen(LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG, {
|
|
10035
|
+
company,
|
|
10036
|
+
defaultValue: true,
|
|
10037
|
+
env: ctx.env,
|
|
10038
|
+
});
|
|
10024
10039
|
printJson(spawnEnvelopeRepairRequest(ctx.hqRoot(), payload, {
|
|
10025
10040
|
env: ctx.env,
|
|
10041
|
+
envelopeRepairGuidanceEnabled,
|
|
10026
10042
|
hostname: ctx.hostname(),
|
|
10027
10043
|
existsSync: ctx.existsSync,
|
|
10028
10044
|
spawnProcess: ctx.spawnProcess,
|
|
@@ -44,4 +44,6 @@ export declare const LANE_REAP_STOP_FLAG = "lanes.reap-stop";
|
|
|
44
44
|
export declare const LANE_WORKER_CONTINUITY_FLAG = "lanes.worker-continuity";
|
|
45
45
|
/** Kill switch for conduct loop lanes; absence or read failure keeps it enabled. */
|
|
46
46
|
export declare const LANE_CONDUCT_LOOP_FLAG = "lanes.conduct-loop";
|
|
47
|
+
export declare const LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG = "lanes.envelope-repair-guidance";
|
|
48
|
+
export declare const LANE_LOOP_HANDOFF_REPAIR_FLAG = "lanes.loop-handoff-repair";
|
|
47
49
|
//# sourceMappingURL=flag-keys.d.ts.map
|
|
@@ -44,4 +44,6 @@ export const LANE_REAP_STOP_FLAG = "lanes.reap-stop";
|
|
|
44
44
|
export const LANE_WORKER_CONTINUITY_FLAG = "lanes.worker-continuity";
|
|
45
45
|
/** Kill switch for conduct loop lanes; absence or read failure keeps it enabled. */
|
|
46
46
|
export const LANE_CONDUCT_LOOP_FLAG = "lanes.conduct-loop";
|
|
47
|
+
export const LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG = "lanes.envelope-repair-guidance";
|
|
48
|
+
export const LANE_LOOP_HANDOFF_REPAIR_FLAG = "lanes.loop-handoff-repair";
|
|
47
49
|
//# sourceMappingURL=flag-keys.js.map
|
package/dist/lib/lanes/loop.d.ts
CHANGED
|
@@ -95,6 +95,7 @@ export interface LoopExecutionContext {
|
|
|
95
95
|
signal: AbortSignal;
|
|
96
96
|
sessionId?: string;
|
|
97
97
|
fresh: boolean;
|
|
98
|
+
repairError?: string;
|
|
98
99
|
}
|
|
99
100
|
export interface LoopExecutionResult {
|
|
100
101
|
output: string;
|
|
@@ -116,6 +117,7 @@ export interface LoopRunnerOptions {
|
|
|
116
117
|
now?: () => Date;
|
|
117
118
|
idlePollMs?: number;
|
|
118
119
|
signal?: AbortSignal;
|
|
120
|
+
repairInvalidHandoff?: boolean;
|
|
119
121
|
}
|
|
120
122
|
export interface LoopRunSummary {
|
|
121
123
|
state: "stopped" | "parked" | "aborted";
|
package/dist/lib/lanes/loop.js
CHANGED
|
@@ -750,7 +750,7 @@ export async function runLoopLane(options) {
|
|
|
750
750
|
if (envelope.kind === "stop") {
|
|
751
751
|
const stopped = { schema_version: LOOP_SCHEMA, envelope_id: record.envelope_id, status: "stopped", started_at: record.queued_at, finished_at: now().toISOString() };
|
|
752
752
|
try {
|
|
753
|
-
writeResultAt(options.runDir, file, record, stopped);
|
|
753
|
+
writeResultAt(options.runDir, file, record, stopped, options.repairInvalidHandoff !== false);
|
|
754
754
|
persistStatus(options.runDir, options.lane.lane_id, { state: "stopped", queue_depth: listLoopQueue(options.runDir).length, active_envelope_id: null, active_phase: null });
|
|
755
755
|
}
|
|
756
756
|
catch (error) {
|
|
@@ -811,7 +811,7 @@ export async function runLoopLane(options) {
|
|
|
811
811
|
if (timer.signal.aborted)
|
|
812
812
|
throw timer.signal.reason;
|
|
813
813
|
const finishedAt = now().toISOString();
|
|
814
|
-
|
|
814
|
+
let success = {
|
|
815
815
|
schema_version: LOOP_SCHEMA, envelope_id: record.envelope_id, status: "done", started_at: startedAt, finished_at: finishedAt,
|
|
816
816
|
output: result.output, ...(result.sessionId ? { session_id: result.sessionId } : {}),
|
|
817
817
|
...(previousSession && !fresh && !fallback ? { resumed: true } : { resumed: false }),
|
|
@@ -820,17 +820,55 @@ export async function runLoopLane(options) {
|
|
|
820
820
|
if (envelope.schema === PHASE_ENVELOPE_SCHEMA) {
|
|
821
821
|
let handoff;
|
|
822
822
|
try {
|
|
823
|
-
|
|
824
|
-
|
|
825
|
-
|
|
826
|
-
|
|
823
|
+
try {
|
|
824
|
+
handoff = JSON.parse(result.output);
|
|
825
|
+
}
|
|
826
|
+
catch {
|
|
827
|
+
throw new Error("section 8 phase result is not valid JSON");
|
|
828
|
+
}
|
|
829
|
+
const validated = validateSection8Handoff(handoff, envelope);
|
|
830
|
+
if (validated.worker_id !== envelope.worker_id) {
|
|
831
|
+
throw new Error("handoff worker_id must match the queued phase envelope");
|
|
832
|
+
}
|
|
827
833
|
}
|
|
828
|
-
|
|
829
|
-
|
|
830
|
-
|
|
834
|
+
catch (validationError) {
|
|
835
|
+
if (options.repairInvalidHandoff !== false && !timer.signal.aborted) {
|
|
836
|
+
const repairError = validationError instanceof Error ? validationError.message : String(validationError);
|
|
837
|
+
const repaired = await options.execute({
|
|
838
|
+
lane: options.lane,
|
|
839
|
+
envelope,
|
|
840
|
+
envelopeId: record.envelope_id,
|
|
841
|
+
signal: timer.signal,
|
|
842
|
+
fresh: false,
|
|
843
|
+
...(result.sessionId ? { sessionId: result.sessionId } : {}),
|
|
844
|
+
repairError,
|
|
845
|
+
});
|
|
846
|
+
if (timer.signal.aborted)
|
|
847
|
+
throw timer.signal.reason;
|
|
848
|
+
let repairedHandoff;
|
|
849
|
+
try {
|
|
850
|
+
repairedHandoff = JSON.parse(repaired.output);
|
|
851
|
+
}
|
|
852
|
+
catch {
|
|
853
|
+
throw new Error(`handoff repair returned invalid JSON: ${repairError}`);
|
|
854
|
+
}
|
|
855
|
+
const repairedValidated = validateSection8Handoff(repairedHandoff, envelope);
|
|
856
|
+
if (repairedValidated.worker_id !== envelope.worker_id) {
|
|
857
|
+
throw new Error(`handoff worker_id must match the queued phase envelope: ${repairError}`);
|
|
858
|
+
}
|
|
859
|
+
result = repaired;
|
|
860
|
+
success = {
|
|
861
|
+
...success,
|
|
862
|
+
output: repaired.output,
|
|
863
|
+
...(repaired.sessionId ? { session_id: repaired.sessionId } : {}),
|
|
864
|
+
};
|
|
865
|
+
}
|
|
866
|
+
else {
|
|
867
|
+
throw validationError;
|
|
868
|
+
}
|
|
831
869
|
}
|
|
832
870
|
}
|
|
833
|
-
writeResultAt(options.runDir, file, record, success);
|
|
871
|
+
writeResultAt(options.runDir, file, record, success, options.repairInvalidHandoff !== false);
|
|
834
872
|
if (result.sessionId) {
|
|
835
873
|
sessions.set(sessionKey, result.sessionId);
|
|
836
874
|
atomicJson(sessionsPath(options.runDir), Object.fromEntries(sessions));
|
|
@@ -853,7 +891,7 @@ export async function runLoopLane(options) {
|
|
|
853
891
|
}
|
|
854
892
|
if (!timer.signal.aborted) {
|
|
855
893
|
const failure = error instanceof Error ? error.message : String(error);
|
|
856
|
-
writeResultAt(options.runDir, file, record, { schema_version: LOOP_SCHEMA, envelope_id: record.envelope_id, status: "failed", started_at: startedAt, finished_at: now().toISOString(), error: failure });
|
|
894
|
+
writeResultAt(options.runDir, file, record, { schema_version: LOOP_SCHEMA, envelope_id: record.envelope_id, status: "failed", started_at: startedAt, finished_at: now().toISOString(), error: failure }, options.repairInvalidHandoff !== false);
|
|
857
895
|
completed += 1;
|
|
858
896
|
persistStatus(options.runDir, options.lane.lane_id, { state: "waiting", queue_depth: listLoopQueue(options.runDir).length, active_envelope_id: null, active_phase: null });
|
|
859
897
|
continue;
|
|
@@ -880,7 +918,7 @@ export async function runLoopLane(options) {
|
|
|
880
918
|
const message = error instanceof Error ? error.message : String(error);
|
|
881
919
|
const questionId = options.raiseQuestion ? randomUUID() : undefined;
|
|
882
920
|
const result = { schema_version: LOOP_SCHEMA, envelope_id: record.envelope_id, status: "stalled", started_at: startedAt, finished_at: now().toISOString(), error: message, ...(questionId ? { question_id: questionId } : {}) };
|
|
883
|
-
writeResultAt(options.runDir, file, record, result);
|
|
921
|
+
writeResultAt(options.runDir, file, record, result, options.repairInvalidHandoff !== false);
|
|
884
922
|
persistStatus(options.runDir, options.lane.lane_id, { state: "parked", queue_depth: listLoopQueue(options.runDir).length, active_envelope_id: null, active_phase: null });
|
|
885
923
|
completed += 1;
|
|
886
924
|
try {
|
|
@@ -921,7 +959,7 @@ function appendJournal(runDir, value) {
|
|
|
921
959
|
fs.mkdirSync(path.dirname(file), { recursive: true });
|
|
922
960
|
fs.appendFileSync(file, `${JSON.stringify(value)}\n`, { mode: 0o600 });
|
|
923
961
|
}
|
|
924
|
-
function writeResultAt(runDir, activeFile, record, result) {
|
|
962
|
+
function writeResultAt(runDir, activeFile, record, result, includeHandoffValidationError = true) {
|
|
925
963
|
const localPath = path.join(resultsDir(runDir), resultName(record.envelope_id));
|
|
926
964
|
atomicJson(localPath, result);
|
|
927
965
|
const envelope = record.envelope;
|
|
@@ -939,7 +977,7 @@ function writeResultAt(runDir, activeFile, record, result) {
|
|
|
939
977
|
schema: PHASE_HANDOFF_SCHEMA,
|
|
940
978
|
story_id: String(envelope.story_id), phase: String(envelope.phase),
|
|
941
979
|
worker_id: String(envelope.worker_id), status: stalled ? "blocked" : "failed",
|
|
942
|
-
summary: stalled ? "Phase stalled twice and needs operator direction." : "Phase execution failed.",
|
|
980
|
+
summary: stalled ? "Phase stalled twice and needs operator direction." : includeHandoffValidationError ? `Phase execution failed: ${result.error ?? "unknown error"}` : "Phase execution failed.",
|
|
943
981
|
files_changed: [], commits: [],
|
|
944
982
|
back_pressure: { tests: "skip", lint: "skip", typecheck: "skip", build: "skip" },
|
|
945
983
|
context_for_next: [result.error ?? result.status, ...(result.question_id ? [`Decision question: ${result.question_id}`] : [])].join("\n"),
|
|
@@ -24,7 +24,7 @@ import { ANSWERED_QUESTIONS_FILE, armLaneWatcher, envelopeRepairMode, isPidRecor
|
|
|
24
24
|
import { appendCappedHistory, lanesStoreRoot, parseLaneRecord, parseQuestionRecord, readLane, retryUpdateLane, readQuestion, listQuestionIds, updateLane, updateQuestion, } from "./store.js";
|
|
25
25
|
import { applyThrottleDecision, decideAdmission, defaultThrottleDeps, emptyThrottleState, evaluateThrottle, isCpuCheckEnabled, readAvailableMemMb, readThrottleState, resolveCpuBusyFraction, sampleCpuBusyObservation, recordCpuBusyObservation, throttleStatePath, writeThrottleState, } from "./throttle.js";
|
|
26
26
|
import { readLanesRegistryFlag, readLanesRegistryFlagFailOff, readLanesRegistryFlagFailOpen } from "./registry-flag.js";
|
|
27
|
-
import { LANE_BLOCKED_ESCALATION_FLAG, LANE_HIRE_GRANT_FINISHED_LANES_FLAG, LANE_KEEP_ALIVE_DUE_WAKE_PRECISE_FLAG, LANE_LEAD_ADMISSION_PRIORITY_FLAG, LANE_QUOTA_ACCOUNT_FAILOVER_FLAG, } from "./flag-keys.js";
|
|
27
|
+
import { LANE_BLOCKED_ESCALATION_FLAG, LANE_HIRE_GRANT_FINISHED_LANES_FLAG, LANE_KEEP_ALIVE_DUE_WAKE_PRECISE_FLAG, LANE_LEAD_ADMISSION_PRIORITY_FLAG, LANE_QUOTA_ACCOUNT_FAILOVER_FLAG, LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG, } from "./flag-keys.js";
|
|
28
28
|
import { checkLaneDiskSpace, configuredCodexRolloutHomes, laneDiskAccount, retainCodexRollouts } from "./disk-guard.js";
|
|
29
29
|
import { laneCompanyUidSource, resolveLaneCompany } from "./lane-company.js";
|
|
30
30
|
import { appendDropboxMessageOnce, appendDropboxMessageOnceAtPath } from "./dropbox.js";
|
|
@@ -1700,7 +1700,13 @@ async function heartbeatAndDeadHosts(hqRoot, lanes, summary, deps, now) {
|
|
|
1700
1700
|
}
|
|
1701
1701
|
const spawnRepair = deps.spawnRepair ??
|
|
1702
1702
|
((root, repairLane, input) => spawnEnvelopeRepair(root, repairLane, input, spawnDepsForHost(deps)));
|
|
1703
|
-
const
|
|
1703
|
+
const repairGuidanceFlagReader = deps.readRegistryFlag ?? readLanesRegistryFlagFailOpen;
|
|
1704
|
+
const repairGuidanceEnabled = await repairGuidanceFlagReader(LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG, {
|
|
1705
|
+
company: resolveLaneCompany(hqRoot, classified).company,
|
|
1706
|
+
defaultValue: true,
|
|
1707
|
+
env: deps.spawnLaneDeps?.env ?? deps.env ?? process.env,
|
|
1708
|
+
});
|
|
1709
|
+
const repair = decideEnvelopeRepair(hqRoot, lane.lane_id, repairCandidate, nowMs, spawnRepair, () => undefined, repairGuidanceEnabled);
|
|
1704
1710
|
if (repair.action === "spawned") {
|
|
1705
1711
|
armLaneWatcher(hqRoot, lane.lane_id, spawnDepsForHost(deps));
|
|
1706
1712
|
summary.heartbeats.push(`${lane.lane_id}:envelope-repair`);
|
|
@@ -94,6 +94,7 @@ export interface SpawnLaneDeps {
|
|
|
94
94
|
locateTranscript?: (home: string, sessionId: string) => string | null;
|
|
95
95
|
/** Internal loop-lane override for the single provider turn's working directory. */
|
|
96
96
|
workingDirectory?: string;
|
|
97
|
+
envelopeRepairGuidanceEnabled?: boolean;
|
|
97
98
|
}
|
|
98
99
|
export interface ProviderLaunch {
|
|
99
100
|
command: string;
|
|
@@ -210,12 +211,12 @@ export declare function resumePrompt(input: {
|
|
|
210
211
|
second_error?: string;
|
|
211
212
|
};
|
|
212
213
|
}): string;
|
|
213
|
-
/** Spec §4: the one repair turn. Schema, parser error, raw text, JSON only. */
|
|
214
214
|
export declare function envelopeRepairPrompt(input: {
|
|
215
215
|
envelopePath: string;
|
|
216
216
|
error: string;
|
|
217
217
|
rawText: string;
|
|
218
218
|
missing?: boolean;
|
|
219
|
+
detailedGuidance?: boolean;
|
|
219
220
|
}): string;
|
|
220
221
|
export interface ProviderResumeInput {
|
|
221
222
|
session_id: string;
|
|
@@ -319,6 +320,7 @@ export interface EnvelopeRepairSpawnInput {
|
|
|
319
320
|
rawText: string;
|
|
320
321
|
/** True when this run never wrote its expected envelope file. */
|
|
321
322
|
missing?: boolean;
|
|
323
|
+
detailedGuidance?: boolean;
|
|
322
324
|
/** Repair-only warning hook for profile metadata ignored during recovery. */
|
|
323
325
|
onProfileWarning?: (fields: readonly string[]) => void;
|
|
324
326
|
}
|
package/dist/lib/lanes/spawn.js
CHANGED
|
@@ -1375,6 +1375,42 @@ export function resumePrompt(input) {
|
|
|
1375
1375
|
return parts.join("\n\n");
|
|
1376
1376
|
}
|
|
1377
1377
|
/** Spec §4: the one repair turn. Schema, parser error, raw text, JSON only. */
|
|
1378
|
+
function recoverRawJsonArrayField(rawText, field) {
|
|
1379
|
+
const match = new RegExp(`"${field}"\\s*:\\s*\\[`).exec(rawText);
|
|
1380
|
+
if (!match)
|
|
1381
|
+
return undefined;
|
|
1382
|
+
const start = match.index + match[0].lastIndexOf("[");
|
|
1383
|
+
let depth = 0;
|
|
1384
|
+
let inString = false;
|
|
1385
|
+
let escaped = false;
|
|
1386
|
+
for (let index = start; index < rawText.length; index += 1) {
|
|
1387
|
+
const char = rawText[index];
|
|
1388
|
+
if (inString) {
|
|
1389
|
+
if (escaped)
|
|
1390
|
+
escaped = false;
|
|
1391
|
+
else if (char === "\\")
|
|
1392
|
+
escaped = true;
|
|
1393
|
+
else if (char === '"')
|
|
1394
|
+
inString = false;
|
|
1395
|
+
continue;
|
|
1396
|
+
}
|
|
1397
|
+
if (char === '"')
|
|
1398
|
+
inString = true;
|
|
1399
|
+
else if (char === "[")
|
|
1400
|
+
depth += 1;
|
|
1401
|
+
else if (char === "]" && --depth === 0) {
|
|
1402
|
+
try {
|
|
1403
|
+
const parsed = JSON.parse(rawText.slice(start, index + 1));
|
|
1404
|
+
return Array.isArray(parsed) ? parsed : undefined;
|
|
1405
|
+
}
|
|
1406
|
+
catch {
|
|
1407
|
+
process.stderr.write("hq lanes: envelope repair could not recover the supplied JSON array\n");
|
|
1408
|
+
return undefined;
|
|
1409
|
+
}
|
|
1410
|
+
}
|
|
1411
|
+
}
|
|
1412
|
+
return undefined;
|
|
1413
|
+
}
|
|
1378
1414
|
export function envelopeRepairPrompt(input) {
|
|
1379
1415
|
const maxRawTextChars = 24_000;
|
|
1380
1416
|
const rawText = input.rawText.length <= maxRawTextChars
|
|
@@ -1393,11 +1429,94 @@ export function envelopeRepairPrompt(input) {
|
|
|
1393
1429
|
`Output only that JSON. Do not add keys the schema does not name, and do not resume the task.`,
|
|
1394
1430
|
].join("\n\n");
|
|
1395
1431
|
}
|
|
1432
|
+
if (input.detailedGuidance === false) {
|
|
1433
|
+
return [
|
|
1434
|
+
`The file you wrote at ${input.envelopePath} is not a valid hq-lane-envelope.v1. This is the single repair turn: rewrite that file so it validates, and do nothing else.`,
|
|
1435
|
+
`Parser error: ${input.error}`,
|
|
1436
|
+
"For an ask question, include non-empty id, question, and context fields.",
|
|
1437
|
+
"Schema:", JSON.stringify(LANE_ENVELOPE_JSON_SCHEMA, null, 2),
|
|
1438
|
+
"Your previous output was:", rawText,
|
|
1439
|
+
`Write the corrected JSON object to ${input.envelopePath}, overwriting it. Output only that JSON. Do not add keys the schema does not name, and do not resume the task.`,
|
|
1440
|
+
].join("\n\n");
|
|
1441
|
+
}
|
|
1442
|
+
const location = input.error.match(/position (\d+) \(line (\d+) column (\d+)\)/i);
|
|
1443
|
+
const lineColumn = input.error.match(/line (\d+) column (\d+)/i);
|
|
1444
|
+
const offendingToken = location
|
|
1445
|
+
? input.rawText[Number(location[1])] ?? "<end of file>"
|
|
1446
|
+
: undefined;
|
|
1447
|
+
const preciseError = location || lineColumn
|
|
1448
|
+
? `${input.error}\nFault location: line ${location?.[2] ?? lineColumn?.[1]}, column ${location?.[3] ?? lineColumn?.[2]}.${offendingToken === undefined ? "" : ` Offending token: ${JSON.stringify(offendingToken)}.`}`
|
|
1449
|
+
: input.error;
|
|
1450
|
+
const syntaxFault = /unexpected (?:token|end)|property name.*in JSON at position|JSON at position \d+|position \d+/i.test(input.error);
|
|
1451
|
+
const originalDecision = input.rawText.match(/"decision"\s*:\s*"(done|ask|blocked)"/)?.[1] ?? "done";
|
|
1452
|
+
const originalCompleteMatch = input.rawText.match(/"story_complete"\s*:\s*(true|false)/)?.[1];
|
|
1453
|
+
const originalComplete = originalCompleteMatch === undefined ? undefined : originalCompleteMatch === "true";
|
|
1454
|
+
const originalStatus = input.rawText.match(/"story_status"\s*:\s*"(queued|in_progress|review|done)"/)?.[1];
|
|
1455
|
+
let originalObject = {};
|
|
1456
|
+
try {
|
|
1457
|
+
const parsed = JSON.parse(input.rawText);
|
|
1458
|
+
if (typeof parsed === "object" && parsed !== null && !Array.isArray(parsed))
|
|
1459
|
+
originalObject = parsed;
|
|
1460
|
+
}
|
|
1461
|
+
catch {
|
|
1462
|
+
process.stderr.write("hq lanes: envelope repair is recovering values from invalid JSON\n");
|
|
1463
|
+
}
|
|
1464
|
+
const rawStringField = (field) => {
|
|
1465
|
+
const token = input.rawText.match(new RegExp(`"${field}"\\s*:\\s*("(?:\\\\.|[^"\\\\])*")`))?.[1];
|
|
1466
|
+
if (!token)
|
|
1467
|
+
return undefined;
|
|
1468
|
+
try {
|
|
1469
|
+
return JSON.parse(token);
|
|
1470
|
+
}
|
|
1471
|
+
catch {
|
|
1472
|
+
process.stderr.write(`hq lanes: envelope repair could not recover the supplied ${field} string\n`);
|
|
1473
|
+
return undefined;
|
|
1474
|
+
}
|
|
1475
|
+
};
|
|
1476
|
+
const minimalEnvelope = {
|
|
1477
|
+
schema_version: "hq-lane-envelope.v1",
|
|
1478
|
+
decision: originalDecision,
|
|
1479
|
+
summary: rawStringField("summary") || "Describe the completed work or remaining issue.",
|
|
1480
|
+
story_status: originalComplete === true ? "done" : originalComplete === false && originalStatus === "done" ? "review" : originalStatus ?? "review",
|
|
1481
|
+
artifacts: [],
|
|
1482
|
+
};
|
|
1483
|
+
const originalCompleteReason = rawStringField("story_complete_reason");
|
|
1484
|
+
if (originalComplete !== undefined)
|
|
1485
|
+
minimalEnvelope.story_complete = originalComplete;
|
|
1486
|
+
if (originalComplete !== undefined || originalCompleteReason !== undefined) {
|
|
1487
|
+
minimalEnvelope.story_complete_reason = originalCompleteReason || (originalComplete === true ? "State what was delivered and how it was verified." : "State what remains missing.");
|
|
1488
|
+
}
|
|
1489
|
+
const originalArtifacts = Array.isArray(originalObject.artifacts)
|
|
1490
|
+
? originalObject.artifacts
|
|
1491
|
+
: recoverRawJsonArrayField(input.rawText, "artifacts");
|
|
1492
|
+
if (originalArtifacts)
|
|
1493
|
+
minimalEnvelope.artifacts = originalArtifacts;
|
|
1494
|
+
if (originalDecision === "ask") {
|
|
1495
|
+
const originalQuestion = Array.isArray(originalObject.questions) && originalObject.questions[0] && typeof originalObject.questions[0] === "object"
|
|
1496
|
+
? originalObject.questions[0]
|
|
1497
|
+
: {};
|
|
1498
|
+
const options = Array.isArray(originalQuestion.options) && originalQuestion.options.every((item) => typeof item === "string")
|
|
1499
|
+
? originalQuestion.options
|
|
1500
|
+
: ["Option A", "Option B"];
|
|
1501
|
+
minimalEnvelope.questions = [{
|
|
1502
|
+
id: typeof originalQuestion.id === "string" && originalQuestion.id.trim() ? originalQuestion.id : "repair-question",
|
|
1503
|
+
question: typeof originalQuestion.question === "string" && originalQuestion.question.trim() ? originalQuestion.question : rawStringField("question") || "What decision is needed?",
|
|
1504
|
+
context: typeof originalQuestion.context === "string" && originalQuestion.context.trim() ? originalQuestion.context : rawStringField("context") || "State the unresolved choice.",
|
|
1505
|
+
options,
|
|
1506
|
+
}];
|
|
1507
|
+
}
|
|
1508
|
+
if (originalDecision === "blocked")
|
|
1509
|
+
minimalEnvelope.blocked_reason = rawStringField("blocked_reason") || "State the external blocker.";
|
|
1396
1510
|
return [
|
|
1397
1511
|
`The file you wrote at ${input.envelopePath} is not a valid hq-lane-envelope.v1. ` +
|
|
1398
1512
|
`This is the single repair turn: rewrite that file so it validates, and do nothing else.`,
|
|
1399
|
-
`
|
|
1400
|
-
|
|
1513
|
+
`Exact validation error: ${preciseError}`,
|
|
1514
|
+
syntaxFault
|
|
1515
|
+
? "This is a JSON syntax error. Fix the exact reported line and column and remove the offending token or delimiter. Do not add trailing commas."
|
|
1516
|
+
: "This is a schema validation error. Correct the named field path and use only the allowed value(s) stated in the error.",
|
|
1517
|
+
"Keep every supplied value that is already valid. Do not invent next_check_at; it is optional. For decision=ask include non-empty id, question, and context fields; for decision=blocked include blocked_reason.",
|
|
1518
|
+
"Minimal valid envelope shape (retain your original decision and values when valid):",
|
|
1519
|
+
JSON.stringify(minimalEnvelope, null, 2),
|
|
1401
1520
|
"Schema:",
|
|
1402
1521
|
JSON.stringify(LANE_ENVELOPE_JSON_SCHEMA, null, 2),
|
|
1403
1522
|
"Your previous output was:",
|
|
@@ -2894,6 +3013,7 @@ export function spawnEnvelopeRepair(hqRoot, lane, input, deps = {}) {
|
|
|
2894
3013
|
error: input.error,
|
|
2895
3014
|
rawText: input.rawText,
|
|
2896
3015
|
missing: input.missing,
|
|
3016
|
+
detailedGuidance: input.detailedGuidance ?? deps.envelopeRepairGuidanceEnabled,
|
|
2897
3017
|
});
|
|
2898
3018
|
const command = buildProviderCommand(lane.provider, {
|
|
2899
3019
|
hqRoot,
|
|
@@ -800,6 +800,10 @@ export interface LaneEnvelopeRepair {
|
|
|
800
800
|
skipped_reason?: string;
|
|
801
801
|
/** Envelope slot the repair was asked to rewrite. */
|
|
802
802
|
envelope_ref?: string;
|
|
803
|
+
/** Absolute raw envelope retained when the repair output also fails JSON parsing. */
|
|
804
|
+
raw_envelope_path?: string;
|
|
805
|
+
/** Recoverable artifacts from malformed raw JSON, surfaced by lanes show --json. */
|
|
806
|
+
artifacts?: unknown[];
|
|
803
807
|
at: string;
|
|
804
808
|
}
|
|
805
809
|
export interface AccountHistoryEntry {
|
|
@@ -247,18 +247,7 @@ export type EnvelopeRepairDecision = {
|
|
|
247
247
|
action: "give_up";
|
|
248
248
|
errors: string[];
|
|
249
249
|
};
|
|
250
|
-
|
|
251
|
-
* Spec §4 repair decision for an exited worker whose expected envelope is
|
|
252
|
-
* invalid or missing. First failure on a resumable provider with a recorded
|
|
253
|
-
* session id: record `envelope_repair` on the lane and spawn the repair turn.
|
|
254
|
-
* A provider with no session resume gets the same single attempt as a fresh
|
|
255
|
-
* turn for invalid files. Anything else (already attempted, no session id on a
|
|
256
|
-
* resumable provider, non-running lane, spawn failure): give up with every
|
|
257
|
-
* error collected. The caller preserves the normal interrupted classification
|
|
258
|
-
* for a missing file and uses `envelope_invalid` only for an existing invalid
|
|
259
|
-
* file.
|
|
260
|
-
*/
|
|
261
|
-
export declare function decideEnvelopeRepair(hqRoot: string, laneId: string, latest: LatestEnvelopeFile, nowMs: number, spawnRepair: NonNullable<WatcherLoopDeps["spawnRepair"]>, logDecision: (record: Record<string, unknown>) => void): EnvelopeRepairDecision;
|
|
250
|
+
export declare function decideEnvelopeRepair(hqRoot: string, laneId: string, latest: LatestEnvelopeFile, nowMs: number, spawnRepair: NonNullable<WatcherLoopDeps["spawnRepair"]>, logDecision: (record: Record<string, unknown>) => void, repairGuidanceEnabled?: boolean): EnvelopeRepairDecision;
|
|
262
251
|
/** Exported so tests can prove the watcher reads the live record, not spawn env. */
|
|
263
252
|
export declare function laneIdentityFromRecord(hqRoot: string, laneId: string): {
|
|
264
253
|
lane_id: string;
|
|
@@ -49,13 +49,13 @@ import { loadLanesConfig, providerProgressEventMinIntervalMs, providerQuotaDefau
|
|
|
49
49
|
import { resolveAdmissionHost as resolveProcessAdmissionHost } from "./admission-host.js";
|
|
50
50
|
import { createLaneMesh, emitLaneDecisionEvents, emitLaneProgressEvent, emitLaneTerminalStateEvent, laneStoryStatusSuppressionKey, progressEventDue, } from "./mesh-events.js";
|
|
51
51
|
import { envelopeQuestionId } from "./questions.js";
|
|
52
|
-
import { readLanesRegistryFlag } from "./registry-flag.js";
|
|
52
|
+
import { readLanesRegistryFlag, readLanesRegistryFlagFailOpen } from "./registry-flag.js";
|
|
53
53
|
import { laneCompanyUidSource, resolveLaneCompany } from "./lane-company.js";
|
|
54
54
|
import { blockerFreshnessFingerprint, resolveBlockerFreshnessForLane, } from "./blocker-freshness.js";
|
|
55
55
|
import { krFreshnessFingerprint, resolveKrFreshnessForLane, } from "./kr-freshness.js";
|
|
56
56
|
import { managerStatusWasWritten, readStatusSnapshot, resolveManagerStatusTarget, } from "./manager-status-gate.js";
|
|
57
57
|
import { LEAD_LANE_KEEP_ALIVE_FIRST_RESUME_SETTLE_MS, LEAD_LANE_KEEP_ALIVE_FLAG, LEAD_LANE_KEEP_ALIVE_IMMEDIATE_FIRST_RESUME_FLAG, LEAD_LANE_KEEP_ALIVE_TERMINAL_TRIGGER_FLAG, queueLaneTerminalResume, queueManagerStatusGateResume, } from "./keep-alive.js";
|
|
58
|
-
import { LANE_ENVELOPE_REPAIR_CONTINUE_FLAG } from "./flag-keys.js";
|
|
58
|
+
import { LANE_ENVELOPE_REPAIR_CONTINUE_FLAG, LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG } from "./flag-keys.js";
|
|
59
59
|
import { monitorSessionDir } from "../monitor.js";
|
|
60
60
|
import { laneRunDir, missingPidProbe, probePidRecord, readPidRecord, readProcStarttime, readSpawnManifest, envelopeRepairMode, spawnEnvelopeRepair, writePidRecord, retryLaneOnAlternateAccount, } from "./spawn.js";
|
|
61
61
|
/** Spec §5.7: no signal for 20 minutes while the process is still alive. */
|
|
@@ -1574,7 +1574,13 @@ async function runWatcherLoopOwned(hqRoot, laneId, deps) {
|
|
|
1574
1574
|
const missingRepairAllowed = !repairCandidate.missing ||
|
|
1575
1575
|
(lane !== null && envelopeRepairMode(lane.provider) === "resume");
|
|
1576
1576
|
if (missingRepairAllowed) {
|
|
1577
|
-
const
|
|
1577
|
+
const repairLane = readLaneSoft(hqRoot, laneId);
|
|
1578
|
+
const repairGuidanceEnabled = await readLanesRegistryFlagFailOpen(LANE_ENVELOPE_REPAIR_GUIDANCE_FLAG, {
|
|
1579
|
+
company: repairLane?.company ?? repairLane?.mesh_company,
|
|
1580
|
+
defaultValue: true,
|
|
1581
|
+
env: process.env,
|
|
1582
|
+
});
|
|
1583
|
+
const repair = decideEnvelopeRepair(hqRoot, laneId, repairCandidate, now(), spawnRepair, logDecision, repairGuidanceEnabled);
|
|
1578
1584
|
if (repair.action === "spawned") {
|
|
1579
1585
|
lastSignalMs = now();
|
|
1580
1586
|
lastAlive = null;
|
|
@@ -1941,7 +1947,43 @@ async function runWatcherLoopOwned(hqRoot, laneId, deps) {
|
|
|
1941
1947
|
* for a missing file and uses `envelope_invalid` only for an existing invalid
|
|
1942
1948
|
* file.
|
|
1943
1949
|
*/
|
|
1944
|
-
|
|
1950
|
+
function recoverRawEnvelopeArtifacts(rawText) {
|
|
1951
|
+
const match = /"artifacts"\s*:\s*\[/.exec(rawText);
|
|
1952
|
+
if (!match)
|
|
1953
|
+
return undefined;
|
|
1954
|
+
const start = match.index + match[0].lastIndexOf("[");
|
|
1955
|
+
let depth = 0;
|
|
1956
|
+
let inString = false;
|
|
1957
|
+
let escaped = false;
|
|
1958
|
+
for (let index = start; index < rawText.length; index += 1) {
|
|
1959
|
+
const char = rawText[index];
|
|
1960
|
+
if (inString) {
|
|
1961
|
+
if (escaped)
|
|
1962
|
+
escaped = false;
|
|
1963
|
+
else if (char === "\\")
|
|
1964
|
+
escaped = true;
|
|
1965
|
+
else if (char === '"')
|
|
1966
|
+
inString = false;
|
|
1967
|
+
continue;
|
|
1968
|
+
}
|
|
1969
|
+
if (char === '"')
|
|
1970
|
+
inString = true;
|
|
1971
|
+
else if (char === "[")
|
|
1972
|
+
depth += 1;
|
|
1973
|
+
else if (char === "]" && --depth === 0) {
|
|
1974
|
+
try {
|
|
1975
|
+
const parsed = JSON.parse(rawText.slice(start, index + 1));
|
|
1976
|
+
return Array.isArray(parsed) ? parsed : undefined;
|
|
1977
|
+
}
|
|
1978
|
+
catch {
|
|
1979
|
+
process.stderr.write("hq lanes watcher: could not recover artifacts from the invalid envelope\n");
|
|
1980
|
+
return undefined;
|
|
1981
|
+
}
|
|
1982
|
+
}
|
|
1983
|
+
}
|
|
1984
|
+
return undefined;
|
|
1985
|
+
}
|
|
1986
|
+
export function decideEnvelopeRepair(hqRoot, laneId, latest, nowMs, spawnRepair, logDecision, repairGuidanceEnabled = true) {
|
|
1945
1987
|
const at = new Date(nowMs).toISOString();
|
|
1946
1988
|
const error = latest.error ?? "unparseable envelope";
|
|
1947
1989
|
const lane = readLaneSoft(hqRoot, laneId);
|
|
@@ -1964,10 +2006,14 @@ export function decideEnvelopeRepair(hqRoot, laneId, latest, nowMs, spawnRepair,
|
|
|
1964
2006
|
// The repair turn ran and its output does not parse either.
|
|
1965
2007
|
if (prior.attempted) {
|
|
1966
2008
|
const errors = [prior.first_error, error];
|
|
2009
|
+
let artifacts;
|
|
2010
|
+
if (repairGuidanceEnabled) {
|
|
2011
|
+
artifacts = recoverRawEnvelopeArtifacts(latest.text);
|
|
2012
|
+
}
|
|
1967
2013
|
try {
|
|
1968
2014
|
retryUpdateLane(hqRoot, laneId, "watcher", (current) => ({
|
|
1969
2015
|
...current,
|
|
1970
|
-
envelope_repair: { ...prior, succeeded: false, second_error: error },
|
|
2016
|
+
envelope_repair: { ...prior, succeeded: false, second_error: error, ...(repairGuidanceEnabled ? { raw_envelope_path: latest.path } : {}), ...(artifacts ? { artifacts } : {}) },
|
|
1971
2017
|
}));
|
|
1972
2018
|
}
|
|
1973
2019
|
catch (err) {
|
|
@@ -1993,6 +2039,7 @@ export function decideEnvelopeRepair(hqRoot, laneId, latest, nowMs, spawnRepair,
|
|
|
1993
2039
|
const base = {
|
|
1994
2040
|
first_error: error,
|
|
1995
2041
|
envelope_ref: repairRef,
|
|
2042
|
+
...(repairGuidanceEnabled ? { raw_envelope_path: latest.path } : {}),
|
|
1996
2043
|
at,
|
|
1997
2044
|
};
|
|
1998
2045
|
if (skipped !== undefined) {
|
|
@@ -2008,6 +2055,7 @@ export function decideEnvelopeRepair(hqRoot, laneId, latest, nowMs, spawnRepair,
|
|
|
2008
2055
|
error,
|
|
2009
2056
|
rawText: latest.text,
|
|
2010
2057
|
missing: latest.missing,
|
|
2058
|
+
detailedGuidance: repairGuidanceEnabled,
|
|
2011
2059
|
onProfileWarning: (fields) => {
|
|
2012
2060
|
logDecision({
|
|
2013
2061
|
event: "envelope_repair",
|