Files
EJClaw/src/phantom-notification-guard.test.ts
Codex 007e95fe40 Extract + unit-test the phantom auto-continue injection path
Move the storeMessage → enqueueMessageCheck side effect out of the delivery
callback into scheduleAutoContinueForPhantomFinal so the injection path is
directly unit-tested: an auto-continue outcome for a delivered final injects the
corrective nudge as an ipc_injected_human message and re-enqueues a turn, while
a non-auto-continue outcome or an undelivered final does nothing. Behavior is
identical to the inline version — this only adds test coverage and a no-op
refactor, so no redeploy is required.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-08-25 23:12:18 +09:00

231 lines
6.8 KiB
TypeScript

import { beforeEach, describe, expect, it, vi } from 'vitest';
import {
applyPhantomNotificationGuard,
isPhantomNotificationDeadEnd,
PHANTOM_CONTINUE_NUDGE,
PHANTOM_NOTIFICATION_NOTICE,
resolvePhantomFinalOutcome,
scheduleAutoContinueForPhantomFinal,
_resetPhantomAutoContinueForTests,
} from './phantom-notification-guard.js';
describe('isPhantomNotificationDeadEnd', () => {
it('detects the reported "I\'ll wait for the build … I\'ll be notified" final', () => {
expect(
isPhantomNotificationDeadEnd(
"I'll wait for the build to complete — I'll be notified.",
),
).toBe(true);
});
it('detects close variants of the notification promise', () => {
expect(
isPhantomNotificationDeadEnd('I will be notified when it finishes.'),
).toBe(true);
expect(
isPhantomNotificationDeadEnd('Waiting for the deploy to finish now.'),
).toBe(true);
expect(
isPhantomNotificationDeadEnd("I'll be notified once CI is green."),
).toBe(true);
});
it('does not flag ordinary status finals', () => {
expect(
isPhantomNotificationDeadEnd(
'TASK_DONE\n빌드를 끝냈고 테스트 1597개가 모두 통과했습니다.',
),
).toBe(false);
expect(
isPhantomNotificationDeadEnd('The build completed and I pushed the fix.'),
).toBe(false);
expect(isPhantomNotificationDeadEnd('')).toBe(false);
});
});
describe('applyPhantomNotificationGuard', () => {
const deadEnd = "I'll wait for the build to complete — I'll be notified.";
it('appends the follow-up notice for a dead-end final with no active watcher', () => {
const result = applyPhantomNotificationGuard(deadEnd, {
hasActiveWatcher: false,
});
expect(result.startsWith(deadEnd)).toBe(true);
expect(result).toContain(PHANTOM_NOTIFICATION_NOTICE.trim());
});
it('leaves the final untouched when a watcher is actually active', () => {
expect(
applyPhantomNotificationGuard(deadEnd, { hasActiveWatcher: true }),
).toBe(deadEnd);
});
it('leaves ordinary finals untouched', () => {
const ok = 'TASK_DONE\n작업을 마쳤습니다.';
expect(applyPhantomNotificationGuard(ok, { hasActiveWatcher: false })).toBe(
ok,
);
});
it('is idempotent — never appends the notice twice', () => {
const once = applyPhantomNotificationGuard(deadEnd, {
hasActiveWatcher: false,
});
const twice = applyPhantomNotificationGuard(once, {
hasActiveWatcher: false,
});
expect(twice).toBe(once);
});
});
describe('resolvePhantomFinalOutcome', () => {
const deadEnd = "I'll wait for the build to complete — I'll be notified.";
const chatJid = 'dc:test-room';
beforeEach(() => {
_resetPhantomAutoContinueForTests();
});
it('auto-continues once, then falls back to the notice on a repeated dead-end', () => {
const first = resolvePhantomFinalOutcome({
chatJid,
text: deadEnd,
hasActiveWatcher: false,
canAutoContinue: true,
});
expect(first.kind).toBe('auto-continue');
// Second consecutive dead-end for the same chat hits the cap → notice, no
// further auto-continue (this is the loop guard).
const second = resolvePhantomFinalOutcome({
chatJid,
text: deadEnd,
hasActiveWatcher: false,
canAutoContinue: true,
});
expect(second.kind).toBe('notice');
expect(second.text).toContain(PHANTOM_NOTIFICATION_NOTICE.trim());
});
it('resets the streak after a clean final so a later dead-end can auto-continue again', () => {
expect(
resolvePhantomFinalOutcome({
chatJid,
text: deadEnd,
hasActiveWatcher: false,
canAutoContinue: true,
}).kind,
).toBe('auto-continue');
// A clean final resets the streak.
expect(
resolvePhantomFinalOutcome({
chatJid,
text: 'TASK_DONE\n작업을 마쳤습니다.',
hasActiveWatcher: false,
canAutoContinue: true,
}).kind,
).toBe('clean');
// Streak reset → next dead-end auto-continues again.
expect(
resolvePhantomFinalOutcome({
chatJid,
text: deadEnd,
hasActiveWatcher: false,
canAutoContinue: true,
}).kind,
).toBe('auto-continue');
});
it('never auto-continues when a watcher is active (legitimate wait)', () => {
expect(
resolvePhantomFinalOutcome({
chatJid,
text: deadEnd,
hasActiveWatcher: true,
canAutoContinue: true,
}).kind,
).toBe('clean');
});
it('emits the notice (never auto-continue) when auto-continue is disabled', () => {
const outcome = resolvePhantomFinalOutcome({
chatJid,
text: deadEnd,
hasActiveWatcher: false,
canAutoContinue: false,
});
expect(outcome.kind).toBe('notice');
});
});
describe('scheduleAutoContinueForPhantomFinal', () => {
it('injects the corrective nudge and re-enqueues when the dead-end final delivered', () => {
const storeMessage = vi.fn();
const enqueueMessageCheck = vi.fn();
const scheduled = scheduleAutoContinueForPhantomFinal({
outcomeKind: 'auto-continue',
delivered: true,
chatJid: 'dc:test-room',
groupFolder: 'test-room',
runId: 'run-xyz',
storeMessage,
enqueueMessageCheck,
});
expect(scheduled).toBe(true);
expect(storeMessage).toHaveBeenCalledTimes(1);
expect(storeMessage).toHaveBeenCalledWith(
expect.objectContaining({
chat_jid: 'dc:test-room',
content: PHANTOM_CONTINUE_NUDGE,
is_from_me: false,
is_bot_message: false,
message_source_kind: 'ipc_injected_human',
}),
);
expect(enqueueMessageCheck).toHaveBeenCalledWith('dc:test-room', 'test-room');
});
it('does nothing when the outcome is not auto-continue', () => {
const storeMessage = vi.fn();
const enqueueMessageCheck = vi.fn();
const scheduled = scheduleAutoContinueForPhantomFinal({
outcomeKind: 'notice',
delivered: true,
chatJid: 'dc:test-room',
groupFolder: 'test-room',
runId: 'run-xyz',
storeMessage,
enqueueMessageCheck,
});
expect(scheduled).toBe(false);
expect(storeMessage).not.toHaveBeenCalled();
expect(enqueueMessageCheck).not.toHaveBeenCalled();
});
it('does nothing when the final was not delivered', () => {
const storeMessage = vi.fn();
const enqueueMessageCheck = vi.fn();
const scheduled = scheduleAutoContinueForPhantomFinal({
outcomeKind: 'auto-continue',
delivered: false,
chatJid: 'dc:test-room',
groupFolder: 'test-room',
runId: 'run-xyz',
storeMessage,
enqueueMessageCheck,
});
expect(scheduled).toBe(false);
expect(storeMessage).not.toHaveBeenCalled();
expect(enqueueMessageCheck).not.toHaveBeenCalled();
});
});