Deduplicate paired reviewer handoffs

This commit is contained in:
ejclaw
2026-04-07 08:07:25 +09:00
parent b902493deb
commit 1c22d6e56b
8 changed files with 330 additions and 35 deletions

View File

@@ -172,7 +172,7 @@ describe('message-agent-executor-rules', () => {
).toBe('pending'); ).toBe('pending');
}); });
it('requests a generic follow-up after successful owner output moved the task to review_ready', () => { it('does not request an executor-side follow-up after successful owner output moved the task to review_ready', () => {
expect( expect(
resolvePairedFollowUpQueueAction({ resolvePairedFollowUpQueueAction({
completedRole: 'owner', completedRole: 'owner',
@@ -180,10 +180,10 @@ describe('message-agent-executor-rules', () => {
sawOutput: true, sawOutput: true,
taskStatus: 'review_ready', taskStatus: 'review_ready',
}), }),
).toBe('generic'); ).toBe('none');
}); });
it('skips generic requeue when reviewer success moved the task to merge_ready for inline finalize', () => { it('does not request an executor-side follow-up when reviewer success moved the task to merge_ready', () => {
expect( expect(
resolvePairedFollowUpQueueAction({ resolvePairedFollowUpQueueAction({
completedRole: 'reviewer', completedRole: 'reviewer',
@@ -191,7 +191,7 @@ describe('message-agent-executor-rules', () => {
sawOutput: true, sawOutput: true,
taskStatus: 'merge_ready', taskStatus: 'merge_ready',
}), }),
).toBe('skip-inline-finalize'); ).toBe('none');
}); });
it('does not request a generic follow-up after successful reviewer output moved the task back to active', () => { it('does not request a generic follow-up after successful reviewer output moved the task back to active', () => {

View File

@@ -11,9 +11,7 @@ export {
} from './agent-attempt-retry.js'; } from './agent-attempt-retry.js';
export type PairedFollowUpQueueAction = export type PairedFollowUpQueueAction =
| 'generic'
| 'pending' | 'pending'
| 'skip-inline-finalize'
| 'none'; | 'none';
export function resolvePairedFollowUpQueueAction(args: { export function resolvePairedFollowUpQueueAction(args: {
@@ -31,14 +29,7 @@ export function resolvePairedFollowUpQueueAction(args: {
}); });
if (args.executionStatus === 'succeeded' && args.sawOutput) { if (args.executionStatus === 'succeeded' && args.sawOutput) {
if (nextTurnAction.kind === 'reviewer-turn') { return 'none';
return 'generic';
}
return args.completedRole === 'reviewer' &&
nextTurnAction.kind === 'finalize-owner-turn'
? 'skip-inline-finalize'
: 'none';
} }
const shouldRequeuePendingPairedTurn = const shouldRequeuePendingPairedTurn =

View File

@@ -1238,7 +1238,7 @@ describe('runAgentForGroup room memory', () => {
expect(deps.queue.enqueueMessageCheck).not.toHaveBeenCalled(); expect(deps.queue.enqueueMessageCheck).not.toHaveBeenCalled();
}); });
it('enqueues a generic follow-up when owner output moved the task to review_ready', async () => { it('does not enqueue an executor-side follow-up when owner output moved the task to review_ready', async () => {
const group = { ...makeGroup(), folder: 'test-group' }; const group = { ...makeGroup(), folder: 'test-group' };
const deps = makeDeps(); const deps = makeDeps();
@@ -1311,8 +1311,7 @@ describe('runAgentForGroup room memory', () => {
}); });
expect(result).toBe('success'); expect(result).toBe('success');
expect(deps.queue.enqueueMessageCheck).toHaveBeenCalledTimes(1); expect(deps.queue.enqueueMessageCheck).not.toHaveBeenCalled();
expect(deps.queue.enqueueMessageCheck).toHaveBeenCalledWith('group@test');
}); });
it('stores reviewer turn output before transitioning the paired task back to active', async () => { it('stores reviewer turn output before transitioning the paired task back to active', async () => {
@@ -1419,6 +1418,110 @@ describe('runAgentForGroup room memory', () => {
); );
}); });
it('stores reviewer approval output before transitioning the paired task to merge_ready', async () => {
const group = { ...makeGroup(), folder: 'test-group' };
const deps = makeDeps();
const onOutput = vi.fn(async () => {});
vi.mocked(serviceRouting.getEffectiveChannelLease).mockReturnValue({
chat_jid: 'group@test',
owner_agent_type: 'codex',
reviewer_agent_type: 'claude-code',
arbiter_agent_type: null,
owner_service_id: 'codex-main',
reviewer_service_id: 'claude',
arbiter_service_id: null,
activated_at: null,
reason: null,
explicit: false,
});
vi.mocked(
pairedExecutionContext.preparePairedExecutionContext,
).mockReturnValue({
task: {
id: 'paired-task-reviewer-approval-order',
chat_jid: 'group@test',
group_folder: 'test-group',
owner_service_id: 'codex-main',
reviewer_service_id: 'claude',
title: null,
source_ref: 'HEAD',
plan_notes: null,
round_trip_count: 1,
review_requested_at: '2026-03-31T00:00:00.000Z',
status: 'in_review',
arbiter_verdict: null,
arbiter_requested_at: null,
completion_reason: null,
created_at: '2026-03-31T00:00:00.000Z',
updated_at: '2026-03-31T00:00:00.000Z',
},
workspace: null,
envOverrides: {},
});
vi.mocked(db.getPairedTaskById).mockReturnValue({
id: 'paired-task-reviewer-approval-order',
chat_jid: 'group@test',
group_folder: 'test-group',
owner_service_id: 'codex-main',
reviewer_service_id: 'claude',
title: null,
source_ref: 'HEAD',
plan_notes: null,
round_trip_count: 1,
review_requested_at: '2026-03-31T00:00:00.000Z',
status: 'merge_ready',
arbiter_verdict: null,
arbiter_requested_at: null,
completion_reason: null,
created_at: '2026-03-31T00:00:00.000Z',
updated_at: '2026-03-31T00:00:00.000Z',
});
vi.mocked(agentRunner.runAgentProcess).mockImplementation(
async (_group, _input, _onProcess, onOutput) => {
await onOutput?.({
status: 'success',
result: 'DONE\nreviewer approval',
output: {
visibility: 'public',
text: 'DONE\nreviewer approval',
},
phase: 'final',
});
return {
status: 'success',
result: 'DONE\nreviewer approval',
};
},
);
const result = await runAgentForGroup(deps, {
group,
prompt: 'please review',
chatJid: 'group@test',
runId: 'run-reviewer-approval-order',
forcedRole: 'reviewer',
onOutput,
});
expect(result).toBe('success');
expect(db.insertPairedTurnOutput).toHaveBeenCalledWith(
'paired-task-reviewer-approval-order',
1,
'reviewer',
'DONE\nreviewer approval',
);
expect(
vi.mocked(db.insertPairedTurnOutput).mock.invocationCallOrder[0],
).toBeLessThan(onOutput.mock.invocationCallOrder[0]);
expect(
vi.mocked(db.insertPairedTurnOutput).mock.invocationCallOrder[0],
).toBeLessThan(
vi.mocked(pairedExecutionContext.completePairedExecutionContext).mock
.invocationCallOrder[0],
);
});
it('does not enqueue a generic follow-up when reviewer output already returned the task to active', async () => { it('does not enqueue a generic follow-up when reviewer output already returned the task to active', async () => {
const group = { ...makeGroup(), folder: 'test-group' }; const group = { ...makeGroup(), folder: 'test-group' };
const deps = makeDeps(); const deps = makeDeps();

View File

@@ -991,9 +991,9 @@ export async function runAgentForGroup(
} }
} }
// After owner/reviewer completes, enqueue the next turn so // Failed reviewer/arbiter runs can leave a pending paired task without a
// the message loop picks it up without waiting for a new message. // visible bot message. Requeue only those recovery cases here; successful
// Skip if: no output (interrupted), or task already completed (ESCALATE, done, etc.) // paired handoffs are driven by the delivered bot message itself.
if (pairedExecutionContext) { if (pairedExecutionContext) {
const completedRole = roomRoleContext?.role ?? 'owner'; const completedRole = roomRoleContext?.role ?? 'owner';
const finishedCheck = getPairedTaskById(pairedExecutionContext.task.id); const finishedCheck = getPairedTaskById(pairedExecutionContext.task.id);
@@ -1003,18 +1003,7 @@ export async function runAgentForGroup(
sawOutput: pairedSawOutput, sawOutput: pairedSawOutput,
taskStatus: finishedCheck?.status ?? null, taskStatus: finishedCheck?.status ?? null,
}); });
if (queueAction === 'generic') { if (queueAction === 'pending') {
deps.queue.enqueueMessageCheck(chatJid);
} else if (queueAction === 'skip-inline-finalize') {
log.info(
{
taskId: pairedExecutionContext.task.id,
role: completedRole,
taskStatus: finishedCheck?.status ?? null,
},
'Skipping generic follow-up after reviewer approval because inline finalize will handle the owner handoff',
);
} else if (queueAction === 'pending') {
deps.queue.enqueueMessageCheck(chatJid); deps.queue.enqueueMessageCheck(chatJid);
log.info( log.info(
{ {

View File

@@ -35,6 +35,12 @@ export type PendingPairedTurn = {
export type BotOnlyPairedFollowUpAction = export type BotOnlyPairedFollowUpAction =
| { kind: 'none' } | { kind: 'none' }
| {
kind: 'consume-stale-bot-message';
task: PairedTask;
cursor: string | number | null;
currentStatus: PairedTask['status'];
}
| { | {
kind: 'inline-finalize'; kind: 'inline-finalize';
task: PairedTask; task: PairedTask;
@@ -249,6 +255,15 @@ export function resolveBotOnlyPairedFollowUpAction(args: {
lastTurnOutputRole: lastTurnOutput?.role ?? null, lastTurnOutputRole: lastTurnOutput?.role ?? null,
}); });
if (nextTurnAction.kind === 'none') {
return {
kind: 'consume-stale-bot-message',
task,
cursor,
currentStatus: task.status,
};
}
if (nextTurnAction.kind === 'finalize-owner-turn') { if (nextTurnAction.kind === 'finalize-owner-turn') {
return { return {
kind: 'inline-finalize', kind: 'inline-finalize',
@@ -328,6 +343,21 @@ export async function executeBotOnlyPairedFollowUpAction(args: {
); );
} }
if (action.kind === 'consume-stale-bot-message') {
log.info(
{
chatJid,
group: group.name,
groupFolder: group.folder,
taskId: action.task.id,
taskStatus: action.currentStatus,
cursor: action.cursor,
},
'Consumed stale bot-only paired message because no follow-up turn is pending',
);
return true;
}
if (action.kind === 'inline-finalize') { if (action.kind === 'inline-finalize') {
log.info( log.info(
{ {

View File

@@ -49,6 +49,24 @@ describe('message-runtime-rules', () => {
).toEqual({ kind: 'finalize-owner-turn' }); ).toEqual({ kind: 'finalize-owner-turn' });
}); });
it('does not schedule a second reviewer turn when the latest turn already belongs to the reviewer', () => {
expect(
resolveNextTurnAction({
taskStatus: 'review_ready',
lastTurnOutputRole: 'reviewer',
}),
).toEqual({ kind: 'none' });
});
it('does not schedule a second finalize turn when the latest turn already belongs to the owner', () => {
expect(
resolveNextTurnAction({
taskStatus: 'merge_ready',
lastTurnOutputRole: 'owner',
}),
).toEqual({ kind: 'none' });
});
it('maps active tasks with reviewer output to an owner follow-up', () => { it('maps active tasks with reviewer output to an owner follow-up', () => {
expect( expect(
resolveNextTurnAction({ resolveNextTurnAction({

View File

@@ -75,12 +75,18 @@ export function resolveNextTurnAction(args: {
switch (args.taskStatus) { switch (args.taskStatus) {
case 'review_ready': case 'review_ready':
case 'in_review': case 'in_review':
return { kind: 'reviewer-turn' }; return args.lastTurnOutputRole === 'reviewer'
? { kind: 'none' }
: { kind: 'reviewer-turn' };
case 'arbiter_requested': case 'arbiter_requested':
case 'in_arbitration': case 'in_arbitration':
return { kind: 'arbiter-turn' }; return args.lastTurnOutputRole === 'arbiter'
? { kind: 'none' }
: { kind: 'arbiter-turn' };
case 'merge_ready': case 'merge_ready':
return { kind: 'finalize-owner-turn' }; return args.lastTurnOutputRole === 'owner'
? { kind: 'none' }
: { kind: 'finalize-owner-turn' };
case 'active': case 'active':
return args.lastTurnOutputRole === 'reviewer' || return args.lastTurnOutputRole === 'reviewer' ||
args.lastTurnOutputRole === 'arbiter' args.lastTurnOutputRole === 'arbiter'

View File

@@ -213,6 +213,10 @@ import {
resolveHandoffCursorKey, resolveHandoffCursorKey,
resolveHandoffRoleOverride, resolveHandoffRoleOverride,
} from './message-runtime.js'; } from './message-runtime.js';
import {
buildPendingPairedTurn,
resolveBotOnlyPairedFollowUpAction,
} from './message-runtime-flow.js';
import * as config from './config.js'; import * as config from './config.js';
import { logger } from './logger.js'; import { logger } from './logger.js';
import * as serviceRouting from './service-routing.js'; import * as serviceRouting from './service-routing.js';
@@ -1564,6 +1568,160 @@ describe('createMessageRuntime', () => {
expect(saveState).toHaveBeenCalled(); expect(saveState).toHaveBeenCalled();
}); });
it('does not build a second reviewer pending turn once the latest persisted turn already belongs to the reviewer', () => {
const chatJid = 'group@test';
const group = makeGroup('codex');
vi.mocked(serviceRouting.hasReviewerLease).mockReturnValue(true);
const task = {
id: 'task-stale-reviewer-bot-message',
chat_jid: chatJid,
group_folder: group.folder,
owner_service_id: 'claude',
reviewer_service_id: 'codex-main',
title: null,
source_ref: 'HEAD',
plan_notes: null,
review_requested_at: '2026-03-30T00:00:00.000Z',
round_trip_count: 1,
status: 'review_ready',
arbiter_verdict: null,
arbiter_requested_at: null,
completion_reason: null,
created_at: '2026-03-30T00:00:00.000Z',
updated_at: '2026-03-30T00:00:00.000Z',
} as any;
vi.mocked(db.getPairedTurnOutputs).mockReturnValue([
{
id: 1,
task_id: 'task-stale-reviewer-bot-message',
turn_number: 1,
role: 'owner',
output_text: 'owner 응답',
created_at: '2026-03-30T00:00:01.000Z',
},
{
id: 2,
task_id: 'task-stale-reviewer-bot-message',
turn_number: 2,
role: 'reviewer',
output_text: 'reviewer 승인',
created_at: '2026-03-30T00:00:02.000Z',
},
]);
expect(
buildPendingPairedTurn({
chatJid,
timezone: 'UTC',
task,
rawMissedMessages: [
{
seq: 42,
timestamp: '2026-03-30T00:00:04.000Z',
},
],
recentHumanMessages: [],
labeledRecentMessages: [],
resolveChannel: () => makeChannel(chatJid, 'discord-review', false),
}),
).toBeNull();
expect(
resolveBotOnlyPairedFollowUpAction({
chatJid,
task,
isBotOnlyPairedFollowUp: true,
pendingCursorSource: {
seq: 42,
timestamp: '2026-03-30T00:00:04.000Z',
},
}),
).toEqual({
kind: 'consume-stale-bot-message',
task,
cursor: 42,
currentStatus: 'review_ready',
});
});
it('consumes stale bot-only owner messages once the finalize turn output is already persisted', () => {
const chatJid = 'group@test';
const group = makeGroup('codex');
vi.mocked(serviceRouting.hasReviewerLease).mockReturnValue(true);
const task = {
id: 'task-stale-owner-bot-message',
chat_jid: chatJid,
group_folder: group.folder,
owner_service_id: 'claude',
reviewer_service_id: 'codex-main',
title: null,
source_ref: 'HEAD',
plan_notes: null,
review_requested_at: '2026-03-30T00:00:00.000Z',
round_trip_count: 1,
status: 'merge_ready',
arbiter_verdict: null,
arbiter_requested_at: null,
completion_reason: null,
created_at: '2026-03-30T00:00:00.000Z',
updated_at: '2026-03-30T00:00:00.000Z',
} as any;
vi.mocked(db.getPairedTurnOutputs).mockReturnValue([
{
id: 1,
task_id: 'task-stale-owner-bot-message',
turn_number: 1,
role: 'reviewer',
output_text: 'reviewer 승인',
created_at: '2026-03-30T00:00:01.000Z',
},
{
id: 2,
task_id: 'task-stale-owner-bot-message',
turn_number: 2,
role: 'owner',
output_text: 'owner 최종 보고',
created_at: '2026-03-30T00:00:02.000Z',
},
]);
expect(
buildPendingPairedTurn({
chatJid,
timezone: 'UTC',
task,
rawMissedMessages: [
{
seq: 43,
timestamp: '2026-03-30T00:00:04.000Z',
},
],
recentHumanMessages: [],
labeledRecentMessages: [],
resolveChannel: () => makeChannel(chatJid),
}),
).toBeNull();
expect(
resolveBotOnlyPairedFollowUpAction({
chatJid,
task,
isBotOnlyPairedFollowUp: true,
pendingCursorSource: {
seq: 43,
timestamp: '2026-03-30T00:00:04.000Z',
},
}),
).toEqual({
kind: 'consume-stale-bot-message',
task,
cursor: 43,
currentStatus: 'merge_ready',
});
});
it('runs merge_ready bot-only reviewer follow-ups inline in the message loop', async () => { it('runs merge_ready bot-only reviewer follow-ups inline in the message loop', async () => {
const chatJid = 'group@test'; const chatJid = 'group@test';
const group = makeGroup('codex'); const group = makeGroup('codex');