claude-bot/EJClaw
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Recover Codex compaction failures in paired owner flow

Browse Source
This commit is contained in:
ejclaw
2026-04-15 03:11:14 +09:00
parent c76902a84b
commit 2ab3bddc0e
20 changed files with 449 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/db/base-schema.ts
View File

@@ -118,6 +118,7 @@ export function applyBaseSchema(database: Database): void {
plan_notes TEXT,
review_requested_at TEXT,
round_trip_count INTEGER NOT NULL DEFAULT 0,
owner_failure_count INTEGER NOT NULL DEFAULT 0,
status TEXT NOT NULL DEFAULT 'active',
arbiter_verdict TEXT,
arbiter_requested_at TEXT,

1
src/db/bootstrap.test.ts
View File

@@ -35,6 +35,7 @@ function getExpectedSchemaMigrations(): Array<{
{ version: 8, name: 'paired_task_schema_cleanup' },
{ version: 9, name: 'paired_workspace_project_schema_cleanup' },
{ version: 10, name: 'paired_turn_provenance_upgrade' },
{ version: 11, name: 'owner_failure_count' },
];
}

23
src/db/migrations/011_owner-failure-count.ts Normal file
View File

@@ -0,0 +1,23 @@
import type { Database } from 'bun:sqlite';
import { tableHasColumn } from './helpers.js';
import type { SchemaMigrationDefinition } from './types.js';
export const OWNER_FAILURE_COUNT_MIGRATION: SchemaMigrationDefinition = {
version: 11,
name: 'owner_failure_count',
apply(database: Database) {
if (!tableHasColumn(database, 'paired_tasks', 'owner_failure_count')) {
database.exec(`
ALTER TABLE paired_tasks
ADD COLUMN owner_failure_count INTEGER NOT NULL DEFAULT 0
`);
}
database.exec(`
UPDATE paired_tasks
SET owner_failure_count = 0
WHERE owner_failure_count IS NULL
`);
},
};

2
src/db/migrations/index.ts
View File

@@ -10,6 +10,7 @@ import { RUNTIME_SERVICE_METADATA_MIGRATION } from './007_runtime-service-metada
import { PAIRED_TASK_SCHEMA_CLEANUP_MIGRATION } from './008_paired-task-schema-cleanup.js';
import { PAIRED_WORKSPACE_PROJECT_SCHEMA_CLEANUP_MIGRATION } from './009_paired-workspace-project-schema-cleanup.js';
import { PAIRED_TURN_PROVENANCE_UPGRADE_MIGRATION } from './010_paired-turn-provenance-upgrade.js';
import { OWNER_FAILURE_COUNT_MIGRATION } from './011_owner-failure-count.js';
import type {
SchemaMigrationArgs,
SchemaMigrationDefinition,
@@ -28,6 +29,7 @@ const ORDERED_SCHEMA_MIGRATIONS: readonly SchemaMigrationDefinition[] = [
PAIRED_TASK_SCHEMA_CLEANUP_MIGRATION,
PAIRED_WORKSPACE_PROJECT_SCHEMA_CLEANUP_MIGRATION,
PAIRED_TURN_PROVENANCE_UPGRADE_MIGRATION,
OWNER_FAILURE_COUNT_MIGRATION,
];
function ensureSchemaMigrationsTable(database: Database): void {

13
src/db/paired-state.ts
View File

@@ -50,6 +50,7 @@ export type PairedTaskUpdates = Partial<
| 'plan_notes'
| 'review_requested_at'
| 'round_trip_count'
| 'owner_failure_count'
| 'status'
| 'arbiter_verdict'
| 'arbiter_requested_at'
@@ -171,6 +172,7 @@ export function createPairedTaskInDatabase(
plan_notes,
review_requested_at,
round_trip_count,
owner_failure_count,
status,
arbiter_verdict,
arbiter_requested_at,
@@ -178,7 +180,7 @@ export function createPairedTaskInDatabase(
created_at,
updated_at
)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
`,
)
.run(
@@ -195,6 +197,7 @@ export function createPairedTaskInDatabase(
task.plan_notes,
task.review_requested_at,
task.round_trip_count,
task.owner_failure_count ?? 0,
task.status,
task.arbiter_verdict,
task.arbiter_requested_at,
@@ -279,6 +282,10 @@ export function updatePairedTaskInDatabase(
fields.push('round_trip_count = ?');
values.push(updates.round_trip_count);
}
if (updates.owner_failure_count !== undefined) {
fields.push('owner_failure_count = ?');
values.push(updates.owner_failure_count);
}
if (updates.status !== undefined) {
fields.push('status = ?');
values.push(updates.status);
@@ -338,6 +345,10 @@ export function updatePairedTaskIfUnchangedInDatabase(
fields.push('round_trip_count = ?');
values.push(updates.round_trip_count);
}
if (updates.owner_failure_count !== undefined) {
fields.push('owner_failure_count = ?');
values.push(updates.owner_failure_count);
}
if (updates.status !== undefined) {
fields.push('status = ?');
values.push(updates.status);

16
src/message-agent-executor-attempt-runner.ts
View File

@@ -5,7 +5,10 @@ import type { AttemptStreamedTrigger } from './agent-attempt-retry.js';
import { runAgentProcess, type AgentOutput } from './agent-runner.js';
import { getCodexAccountCount } from './codex-token-rotation.js';
import type { PreparedPairedExecutionContext } from './paired-execution-context.js';
import { shouldResetSessionOnAgentFailure } from './session-recovery.js';
import {
shouldResetCodexSessionOnAgentFailure,
shouldResetSessionOnAgentFailure,
} from './session-recovery.js';
import type { AgentType, RegisteredGroup, RoomRoleContext } from './types.js';
export interface MessageAgentAttempt {
@@ -148,6 +151,13 @@ export async function runMessageAgentAttempt(args: {
) {
resetSessionRequested = true;
}
if (
!isClaudeCodeAgent &&
provider === 'codex' &&
shouldResetCodexSessionOnAgentFailure(output)
) {
resetSessionRequested = true;
}
if (
output.newSessionId &&
!resetSessionRequested &&
@@ -197,7 +207,9 @@ export async function runMessageAgentAttempt(args: {
? outputText.slice(0, 160)
: output.error?.slice(0, 160),
},
'Suppressed retryable Claude session failure from chat output',
provider === 'claude'
? 'Suppressed retryable Claude session failure from chat output'
: 'Suppressed retryable Codex session failure from chat output',
);
return;
}

77
src/message-agent-executor-lifecycle.ts
View File

@@ -12,7 +12,11 @@ import type {
} from './agent-error-detection.js';
import type { PairedExecutionLifecycle } from './message-agent-executor-paired.js';
import type { MessageAgentAttempt } from './message-agent-executor-attempt-runner.js';
import { shouldResetSessionOnAgentFailure } from './session-recovery.js';
import {
shouldResetCodexSessionOnAgentFailure,
shouldResetSessionOnAgentFailure,
shouldRetryFreshCodexSessionOnAgentFailure,
} from './session-recovery.js';
import { getErrorMessage } from './utils.js';
type AttemptResult = 'success' | 'error';
@@ -208,6 +212,57 @@ export async function executeMessageAgentAttemptLifecycle(args: {
};
};
const isRetryableCodexSessionFailure = (
attempt: MessageAgentAttempt,
): boolean => {
if (provider !== 'codex' || attempt.sawOutput) {
return false;
}
if (attempt.retryableSessionFailureDetected === true) {
return true;
}
if (attempt.error == null) {
return false;
}
return shouldRetryFreshCodexSessionOnAgentFailure({
result: null,
error: getErrorMessage(attempt.error),
});
};
const recoverRetryableCodexSessionFailure = async (
attempt: MessageAgentAttempt,
): Promise<{
attempt: MessageAgentAttempt;
resolved: AttemptResult | null;
}> => {
if (!isRetryableCodexSessionFailure(attempt)) {
return { attempt, resolved: null };
}
clearStoredSession();
clearRoleSdkSessions();
log.warn(
'Cleared poisoned Codex session before visible output, retrying fresh session',
);
const freshAttempt = await runTrackedAttempt('codex');
if (!isRetryableCodexSessionFailure(freshAttempt)) {
return { attempt: freshAttempt, resolved: null };
}
clearStoredSession();
log.warn('Fresh Codex retry also hit a retryable session failure');
log.error('Retryable Codex session failure persisted after fresh retry');
return {
attempt: freshAttempt,
resolved: 'error',
};
};
const handlePrimaryAttemptFailure = async (
attempt: MessageAgentAttempt,
rotationMessage: string,
@@ -294,6 +349,19 @@ export async function executeMessageAgentAttemptLifecycle(args: {
);
}
if (
!isClaudeCodeAgent &&
provider === 'codex' &&
(resetSessionRequested || shouldResetCodexSessionOnAgentFailure(output))
) {
clearStoredSession();
clearRoleSdkSessions();
log.warn(
{ sessionFolder },
'Cleared poisoned Codex session after unrecoverable error',
);
}
if (output.status === 'error') {
return handlePrimaryAttemptFailure(
attempt,
@@ -354,6 +422,13 @@ export async function executeMessageAgentAttemptLifecycle(args: {
}
primaryAttempt = recoveredSessionAttempt.attempt;
const recoveredCodexSessionAttempt =
await recoverRetryableCodexSessionFailure(primaryAttempt);
if (recoveredCodexSessionAttempt.resolved) {
return recoveredCodexSessionAttempt.resolved;
}
primaryAttempt = recoveredCodexSessionAttempt.attempt;
if (primaryAttempt.error) {
return handlePrimaryAttemptFailure(
primaryAttempt,

11
src/message-agent-executor-rules.test.ts
View File

@@ -172,6 +172,17 @@ describe('message-agent-executor-rules', () => {
).toBe('pending');
});
it('resolves pending arbiter follow-up requeue after repeated owner execution failures', () => {
expect(
resolvePairedFollowUpQueueAction({
completedRole: 'owner',
executionStatus: 'failed',
sawOutput: false,
taskStatus: 'arbiter_requested',
}),
).toBe('pending');
});
it('does not request an executor-side follow-up after successful owner output moved the task to review_ready', () => {
expect(
resolvePairedFollowUpQueueAction({

66
src/message-agent-executor.test.ts
View File

@@ -196,7 +196,9 @@ vi.mock('./agent-error-detection.js', async (importOriginal) => {
});
vi.mock('./session-recovery.js', () => ({
shouldResetCodexSessionOnAgentFailure: vi.fn(() => false),
shouldResetSessionOnAgentFailure: vi.fn(() => false),
shouldRetryFreshCodexSessionOnAgentFailure: vi.fn(() => false),
shouldRetryFreshSessionOnAgentFailure: vi.fn(() => false),
}));
@@ -3330,6 +3332,70 @@ describe('runAgentForGroup Claude rotation', () => {
expect(deps.clearSession).toHaveBeenCalledWith('test-claude');
});
it('drops a poisoned Codex session id before retrying a fresh session after remote compaction failure', async () => {
const group = {
...makeGroup(),
folder: 'test-codex',
agentType: 'codex' as const,
};
const deps = {
...makeDeps(),
getSessions: () => ({ 'test-codex': 'stale-codex-session-id' }),
};
vi.mocked(serviceRouting.getEffectiveChannelLease).mockReturnValue({
chat_jid: 'group@test',
owner_agent_type: 'codex',
reviewer_agent_type: null,
arbiter_agent_type: null,
owner_service_id: 'codex-main',
reviewer_service_id: null,
arbiter_service_id: null,
activated_at: null,
reason: null,
explicit: false,
});
vi.mocked(sessionRecovery.shouldRetryFreshCodexSessionOnAgentFailure)
.mockReturnValueOnce(true)
.mockReturnValueOnce(false);
vi.mocked(sessionRecovery.shouldResetCodexSessionOnAgentFailure)
.mockReturnValueOnce(true)
.mockReturnValueOnce(false);
vi.mocked(agentRunner.runAgentProcess)
.mockImplementationOnce(async (_group, input) => {
expect(input.sessionId).toBe('stale-codex-session-id');
throw new Error(
"Error running remote compact task: Unknown parameter: 'prompt_cache_retention'",
);
})
.mockImplementationOnce(async (_group, input, _onProcess, onOutput) => {
expect(input.sessionId).toBeUndefined();
await onOutput?.({
status: 'success',
phase: 'final',
result: 'fresh Codex retry success',
});
return {
status: 'success',
result: 'fresh Codex retry success',
newSessionId: 'fresh-codex-session-id',
};
});
const result = await runAgentForGroup(deps, {
group,
prompt: 'hello',
chatJid: 'group@test',
runId: 'run-stale-codex-session-id-retry',
onOutput: async () => {},
});
expect(result).toBe('success');
expect(agentRunner.runAgentProcess).toHaveBeenCalledTimes(2);
expect(deps.clearSession).toHaveBeenCalledWith('test-codex');
});
it('suppresses a usage-exhausted banner even when Claude already emitted progress text', async () => {
const outputs: string[] = [];

26
src/message-runtime-rules.test.ts
View File

@@ -133,7 +133,7 @@ describe('message-runtime-rules', () => {
).toEqual({ kind: 'none' });
});
it('dispatches owner delivery success through a paired follow-up enqueue only for reviewer turns', () => {
it('dispatches owner delivery success through a paired follow-up enqueue for reviewer or arbiter turns', () => {
expect(
resolveFollowUpDispatch({
source: 'owner-delivery-success',
@@ -143,6 +143,15 @@ describe('message-runtime-rules', () => {
kind: 'enqueue',
queueKind: 'paired-follow-up',
});
expect(
resolveFollowUpDispatch({
source: 'owner-delivery-success',
nextTurnAction: { kind: 'arbiter-turn' },
}),
).toEqual({
kind: 'enqueue',
queueKind: 'paired-follow-up',
});
expect(
resolveFollowUpDispatch({
source: 'owner-delivery-success',
@@ -151,6 +160,21 @@ describe('message-runtime-rules', () => {
).toEqual({ kind: 'none' });
});
it('dispatches failed owner recovery through a paired follow-up when execution escalated to arbiter', () => {
expect(
resolveFollowUpDispatch({
source: 'executor-recovery',
nextTurnAction: { kind: 'arbiter-turn' },
completedRole: 'owner',
executionStatus: 'failed',
sawOutput: false,
}),
).toEqual({
kind: 'enqueue',
queueKind: 'paired-follow-up',
});
});
it('dispatches reviewer and arbiter delivery success through paired follow-up enqueue when a handoff is pending', () => {
expect(
resolveFollowUpDispatch({

4
src/message-runtime-rules.ts
View File

@@ -183,7 +183,8 @@ export function resolveFollowUpDispatch(args: {
args.source === 'owner-delivery-success' ||
args.completedRole === 'owner'
) {
return args.nextTurnAction.kind === 'reviewer-turn'
return args.nextTurnAction.kind === 'reviewer-turn' ||
args.nextTurnAction.kind === 'arbiter-turn'
? { kind: 'enqueue', queueKind: 'paired-follow-up' }
: { kind: 'none' };
}
@@ -211,6 +212,7 @@ export function resolveFollowUpDispatch(args: {
return { kind: 'none' };
}
if (
args.completedRole !== 'owner' &&
args.completedRole !== 'reviewer' &&
args.completedRole !== 'arbiter'
) {

69
src/paired-execution-context-owner.ts
View File

@@ -21,26 +21,72 @@ import {
import type { PairedTask } from './types.js';
type OwnerFinalizeOutcome = 'stop' | 're_review';
const OWNER_FAILURE_ESCALATION_THRESHOLD = 2;
export function handleFailedOwnerExecution(args: {
task: PairedTask;
taskId: string;
summary?: string | null;
}): void {
const { task, taskId } = args;
const { task, taskId, summary } = args;
const now = new Date().toISOString();
const nextFailureCount = (task.owner_failure_count ?? 0) + 1;
if (nextFailureCount >= OWNER_FAILURE_ESCALATION_THRESHOLD) {
requestArbiterOrEscalate({
taskId,
currentStatus: task.status,
expectedUpdatedAt: task.updated_at,
now,
arbiterLogMessage:
'Owner failed repeatedly without a visible verdict — requesting arbiter',
escalateLogMessage:
'Owner failed repeatedly without a visible verdict — escalating to user',
logContext: {
taskId,
role: 'owner',
previousStatus: task.status,
ownerFailureCount: nextFailureCount,
summary: summary?.slice(0, 160),
},
patch: {
owner_failure_count: nextFailureCount,
},
});
return;
}
if (task.status !== 'active') {
const now = new Date().toISOString();
transitionPairedTaskStatus({
taskId,
currentStatus: task.status,
nextStatus: 'active',
expectedUpdatedAt: task.updated_at,
updatedAt: now,
patch: {
owner_failure_count: nextFailureCount,
},
});
} else {
applyPairedTaskPatch({
taskId,
expectedUpdatedAt: task.updated_at,
updatedAt: now,
patch: {
owner_failure_count: nextFailureCount,
},
});
logger.info(
{ taskId, role: 'owner', previousStatus: task.status },
'Reset task to active after failed execution',
);
}
logger.info(
{
taskId,
role: 'owner',
previousStatus: task.status,
ownerFailureCount: nextFailureCount,
summary: summary?.slice(0, 160),
},
'Reset task to active after failed owner execution',
);
}
function handleOwnerFinalizeCompletion(args: {
@@ -90,6 +136,9 @@ function handleOwnerFinalizeCompletion(args: {
hasNewChanges,
summary: summary?.slice(0, 100),
},
patch: {
owner_failure_count: 0,
},
});
return 'stop';
}
@@ -102,6 +151,9 @@ function handleOwnerFinalizeCompletion(args: {
nextStatus: 'active',
expectedUpdatedAt: task.updated_at,
updatedAt: now,
patch: {
owner_failure_count: 0,
},
});
}
logger.info(
@@ -126,6 +178,7 @@ function handleOwnerFinalizeCompletion(args: {
updatedAt: now,
patch: {
completion_reason: 'done',
owner_failure_count: 0,
},
});
logger.info(
@@ -166,6 +219,7 @@ function maybeAutoTriggerReviewerAfterOwnerCompletion(args: {
updatedAt: now,
patch: {
round_trip_count: task.round_trip_count + 1,
owner_failure_count: 0,
},
});
if (hasActiveCiWatcherForChat(task.chat_jid)) {
@@ -229,6 +283,9 @@ export function handleOwnerCompletion(args: {
ownerVerdict,
summary: summary?.slice(0, 100),
},
patch: {
owner_failure_count: 0,
},
});
return;
}

15
src/paired-execution-context-shared.ts
View File

@@ -247,6 +247,7 @@ export function transitionPairedTaskStatus(args: {
plan_notes?: string | null;
review_requested_at?: string | null;
round_trip_count?: number;
owner_failure_count?: number;
arbiter_verdict?: string | null;
arbiter_requested_at?: string | null;
completion_reason?: string | null;
@@ -290,6 +291,7 @@ export function applyPairedTaskPatch(args: {
plan_notes?: string | null;
review_requested_at?: string | null;
round_trip_count?: number;
owner_failure_count?: number;
status?: PairedTaskStatus;
arbiter_verdict?: string | null;
arbiter_requested_at?: string | null;
@@ -324,6 +326,17 @@ export function requestArbiterOrEscalate(args: {
arbiterLogMessage: string;
escalateLogMessage: string;
logContext?: Record<string, unknown>;
patch?: {
title?: string | null;
source_ref?: string | null;
plan_notes?: string | null;
review_requested_at?: string | null;
round_trip_count?: number;
owner_failure_count?: number;
arbiter_verdict?: string | null;
arbiter_requested_at?: string | null;
completion_reason?: string | null;
};
}): boolean {
const {
taskId,
@@ -342,6 +355,7 @@ export function requestArbiterOrEscalate(args: {
expectedUpdatedAt,
updatedAt: now,
patch: {
...args.patch,
arbiter_requested_at: now,
},
});
@@ -358,6 +372,7 @@ export function requestArbiterOrEscalate(args: {
expectedUpdatedAt,
updatedAt: now,
patch: {
...args.patch,
completion_reason: 'escalated',
},
});

30
src/paired-execution-context.test.ts
View File

@@ -139,6 +139,7 @@ function buildPairedTask(overrides: Partial<PairedTask> = {}): PairedTask {
plan_notes: null,
review_requested_at: null,
round_trip_count: 0,
owner_failure_count: 0,
status: 'active',
arbiter_verdict: null,
arbiter_requested_at: null,
@@ -980,7 +981,7 @@ describe('paired execution context', () => {
);
});
it('still resets owner tasks to active after failed execution', () => {
it('increments owner failure count and resets owner tasks to active after failed execution', () => {
vi.mocked(db.getPairedTaskById).mockReturnValue(
buildPairedTask({
status: 'merge_ready',
@@ -998,11 +999,38 @@ describe('paired execution context', () => {
'task-1',
expect.objectContaining({
status: 'active',
owner_failure_count: 1,
updated_at: expect.any(String),
}),
);
});
it('requests arbiter after repeated owner execution failures without a visible verdict', () => {
vi.spyOn(config, 'isArbiterEnabled').mockReturnValue(true);
vi.mocked(db.getPairedTaskById).mockReturnValue(
buildPairedTask({
status: 'active',
owner_failure_count: 1,
}),
);
completePairedExecutionContext({
taskId: 'task-1',
role: 'owner',
status: 'failed',
summary: "Error running remote compact task: Unknown parameter: 'prompt_cache_retention'",
});
expect(db.updatePairedTask).toHaveBeenCalledWith(
'task-1',
expect.objectContaining({
status: 'arbiter_requested',
owner_failure_count: 2,
arbiter_requested_at: expect.any(String),
}),
);
});
it('releases the execution lease even when a completion handler throws', () => {
const transitionError = new Error('transition failed');
vi.mocked(db.getPairedTaskById).mockReturnValue(

7
src/paired-execution-context.ts
View File

@@ -137,6 +137,7 @@ function createActiveTaskForRoom(args: {
plan_notes: null,
review_requested_at: null,
round_trip_count: 0,
owner_failure_count: 0,
status: 'active',
arbiter_verdict: null,
arbiter_requested_at: null,
@@ -369,7 +370,7 @@ export function preparePairedExecutionContext(args: {
expectedUpdatedAt: latestTask.updated_at,
updatedAt: now,
patch: {
...(hasHuman ? { round_trip_count: 0 } : {}),
...(hasHuman ? { round_trip_count: 0, owner_failure_count: 0 } : {}),
},
});
} else {
@@ -378,7 +379,7 @@ export function preparePairedExecutionContext(args: {
expectedUpdatedAt: latestTask.updated_at,
updatedAt: now,
patch: {
...(hasHuman ? { round_trip_count: 0 } : {}),
...(hasHuman ? { round_trip_count: 0, owner_failure_count: 0 } : {}),
},
});
}
@@ -551,7 +552,7 @@ export function completePairedExecutionContext(args: {
handleFailedArbiterExecution({ task, taskId });
return;
}
handleFailedOwnerExecution({ task, taskId });
handleFailedOwnerExecution({ task, taskId, summary: args.summary });
return;
}

35
src/session-recovery.test.ts
View File

@@ -1,7 +1,9 @@
import { describe, expect, it } from 'vitest';
import {
shouldResetCodexSessionOnAgentFailure,
shouldResetSessionOnAgentFailure,
shouldRetryFreshCodexSessionOnAgentFailure,
shouldRetryFreshSessionOnAgentFailure,
} from './session-recovery.js';
@@ -99,3 +101,36 @@ describe('shouldRetryFreshSessionOnAgentFailure', () => {
).toBe(false);
});
});
describe('shouldResetCodexSessionOnAgentFailure', () => {
it('matches remote compact task failures that mention prompt_cache_retention', () => {
expect(
shouldResetCodexSessionOnAgentFailure({
result: null,
error:
"Error running remote compact task: Unknown parameter: 'prompt_cache_retention'",
}),
).toBe(true);
});
it('does not match unrelated Codex failures', () => {
expect(
shouldResetCodexSessionOnAgentFailure({
result: null,
error: 'Codex process exited with code 1',
}),
).toBe(false);
});
});
describe('shouldRetryFreshCodexSessionOnAgentFailure', () => {
it('retries the same remote compact task failure with a fresh session', () => {
expect(
shouldRetryFreshCodexSessionOnAgentFailure({
result: null,
error:
"Error running remote compact task: Unknown parameter: 'prompt_cache_retention'",
}),
).toBe(true);
});
});

23
src/session-recovery.ts
View File

@@ -17,6 +17,11 @@ const SESSION_RETRY_PATTERNS = [
/invalid[^\n]*signature[^\n]*thinking/i,
];
const CODEX_SESSION_RESET_PATTERNS = [
/Error running remote compact task/i,
/prompt_cache_retention/i,
];
function toText(value: string | object | null | undefined): string[] {
if (!value) return [];
if (typeof value === 'string') return [value];
@@ -51,3 +56,21 @@ export function shouldRetryFreshSessionOnAgentFailure(
SESSION_RETRY_PATTERNS.some((pattern) => pattern.test(text)),
);
}
export function shouldResetCodexSessionOnAgentFailure(
output: Pick<AgentOutput, 'result' | 'output' | 'error'>,
): boolean {
const texts = [
...toText(getAgentOutputText(output)),
...toText(output.error),
];
return texts.some((text) =>
CODEX_SESSION_RESET_PATTERNS.some((pattern) => pattern.test(text)),
);
}
export function shouldRetryFreshCodexSessionOnAgentFailure(
output: Pick<AgentOutput, 'result' | 'output' | 'error'>,
): boolean {
return shouldResetCodexSessionOnAgentFailure(output);
}

29
src/streamed-output-evaluator.test.ts
View File

@@ -21,6 +21,7 @@ vi.mock('./codex-token-rotation.js', () => ({
}));
vi.mock('./session-recovery.js', () => ({
shouldRetryFreshCodexSessionOnAgentFailure: vi.fn(() => false),
shouldRetryFreshSessionOnAgentFailure: vi.fn(() => false),
}));
@@ -34,7 +35,10 @@ import {
isClaudeAuthError,
} from './agent-error-detection.js';
import { detectCodexRotationTrigger } from './codex-token-rotation.js';
import { shouldRetryFreshSessionOnAgentFailure } from './session-recovery.js';
import {
shouldRetryFreshCodexSessionOnAgentFailure,
shouldRetryFreshSessionOnAgentFailure,
} from './session-recovery.js';
import {
evaluateStreamedOutput,
@@ -91,6 +95,7 @@ beforeEach(() => {
shouldRotate: false,
reason: '',
});
vi.mocked(shouldRetryFreshCodexSessionOnAgentFailure).mockReturnValue(false);
vi.mocked(shouldRetryFreshSessionOnAgentFailure).mockReturnValue(false);
});
@@ -252,6 +257,28 @@ describe('evaluateStreamedOutput', () => {
});
});
describe('Codex retryable session failure suppression', () => {
it('suppresses remote compact failures before any visible output', () => {
vi.mocked(shouldRetryFreshCodexSessionOnAgentFailure).mockReturnValue(
true,
);
const result = evaluateStreamedOutput(
successOutput(
"Error running remote compact task: Unknown parameter: 'prompt_cache_retention'",
'intermediate',
),
freshState(),
codexOpts,
);
expect(result.shouldForwardOutput).toBe(false);
expect(result.suppressedRetryableSessionFailure).toBe(true);
expect(result.state.retryableSessionFailureDetected).toBe(true);
expect(result.state.sawOutput).toBe(false);
});
});
describe('Claude auth-expired banner', () => {
it('suppresses output and returns newTrigger', () => {
vi.mocked(isClaudeAuthExpiredMessage).mockReturnValue(true);

18
src/streamed-output-evaluator.ts
View File

@@ -15,7 +15,10 @@ import {
} from './agent-error-detection.js';
import type { AgentOutput } from './agent-runner.js';
import { detectCodexRotationTrigger } from './codex-token-rotation.js';
import { shouldRetryFreshSessionOnAgentFailure } from './session-recovery.js';
import {
shouldRetryFreshCodexSessionOnAgentFailure,
shouldRetryFreshSessionOnAgentFailure,
} from './session-recovery.js';
export interface StreamedTriggerReason {
reason: AgentTriggerReason;
@@ -74,6 +77,19 @@ export function evaluateStreamedOutput(
};
}
if (
isPrimaryCodex &&
!state.sawOutput &&
shouldRetryFreshCodexSessionOnAgentFailure(output)
) {
nextState.retryableSessionFailureDetected = true;
return {
state: nextState,
shouldForwardOutput: false,
suppressedRetryableSessionFailure: true,
};
}
if (
isPrimaryClaude &&
output.status === 'success' &&

1
src/types.ts
View File

@@ -82,6 +82,7 @@ export interface PairedTask {
plan_notes: string | null;
review_requested_at: string | null;
round_trip_count: number;
owner_failure_count?: number | null;
status: PairedTaskStatus;
arbiter_verdict: string | null;
arbiter_requested_at: string | null;