diff --git a/cli/README.md b/cli/README.md index 44aaac4e..ca520cee 100644 --- a/cli/README.md +++ b/cli/README.md @@ -276,25 +276,3 @@ bun run build:single-exe - `../hub/README.md` - `../web/README.md` - -### Codex Luna Reserve - -Remote sessions read Codex's account usage through app-server. Luna Reserve is -shown only while the task is using the backend-authorized Reserve route; an -unused Reserve allowance never adds a row or a model option. Ordinary and -Reserve windows remain separate, and unavailable percentages stay unknown. - -HAPI applies the official fallback with `thread/settings/update` and restores -the task's saved model and reasoning effort after a fresh account read permits -ordinary usage. It does not replay the blocked turn. Resume first reconciles -Codex's task settings; the task-local return record uses Codex's -`CODEX_HOME/tui-luna-reserve/.json` format for TUI handoff. - -This requires the newer app-server protocol with `ordinaryUsageAllowed`, -`supportsLunaReserve`, the hidden Reserve catalog entry, and thread settings -updates. Verified against official source -[`ac192cd7937`](https://github.com/openai/codex/tree/ac192cd7937b0d73edc6dffe009940ae53782dd4). -Codex 0.153.4 does not expose the required usage capability; no minimum released -version is claimed. Older servers keep their ordinary usage display without -advertising Reserve activation. Real eligible-account exhaustion, Reserve -exhaustion, and recovery still require account-level validation. diff --git a/cli/src/codex/appServerTypes.ts b/cli/src/codex/appServerTypes.ts index dc5e2799..8d6dc4dc 100644 --- a/cli/src/codex/appServerTypes.ts +++ b/cli/src/codex/appServerTypes.ts @@ -387,11 +387,3 @@ export interface ExperimentalFeatureEnablementSetResponse { enablement: Record; [key: string]: unknown; } - -export interface ThreadSettingsUpdateParams { - threadId: string; - model: string; - effort?: string | null; - serviceTier?: string | null; - collaborationMode?: CollaborationMode; -} diff --git a/cli/src/codex/codexAppServerClient.test.ts b/cli/src/codex/codexAppServerClient.test.ts index c2cd395c..b7df81ba 100644 --- a/cli/src/codex/codexAppServerClient.test.ts +++ b/cli/src/codex/codexAppServerClient.test.ts @@ -62,26 +62,6 @@ describe('CodexAppServerClient process cwd', () => { spawnMock.mockReset(); }); - it('uses null for legacy usage reads and opts in only when Reserve is supported', async () => { - const child = fakeChild(); - child.stdin.write = vi.fn((_data: unknown, callback?: (error?: Error | null) => void) => { - callback?.(); - return true; - }); - spawnMock.mockReturnValue(child); - const client = new CodexAppServerClient({ cwd: '/neutral-home' }); - await client.connect(); - for (const supported of [false, true]) { - const pending = client.readAccountRateLimits(supported); - const request = JSON.parse(child.stdin.write.mock.lastCall![0] as string); - expect(request.method).toBe('account/rateLimits/read'); - expect(request.params).toEqual(supported ? { supportsLunaReserve: true } : null); - child.stdout.emit('data', Buffer.from(JSON.stringify({ id: request.id, result: { rateLimits: {} } }) + '\n')); - await expect(pending).resolves.toEqual({ rateLimits: {} }); - } - await client.disconnect(); - }); - it('passes an explicit neutral cwd to the app-server process', async () => { spawnMock.mockReturnValue(fakeChild()); const client = new CodexAppServerClient({ cwd: '/neutral-home' }); diff --git a/cli/src/codex/codexAppServerClient.ts b/cli/src/codex/codexAppServerClient.ts index 32a72d9c..3460ffa6 100644 --- a/cli/src/codex/codexAppServerClient.ts +++ b/cli/src/codex/codexAppServerClient.ts @@ -5,7 +5,6 @@ import { logger } from '@/ui/logger'; import { JsonLineParser } from '@/utils/jsonLineParser'; import { killProcessByChildProcess } from '@/utils/process'; import type { - ThreadSettingsUpdateParams, CollaborationModeListResponse, InitializeParams, InitializeResponse, @@ -356,14 +355,6 @@ export class CodexAppServerClient extends JsonLineParser { return response as ConfigReadResponse; } - async readAccountRateLimits(supportsLunaReserve = false): Promise { - return this.sendRequest('account/rateLimits/read', supportsLunaReserve ? { supportsLunaReserve: true } : null, { timeoutMs: 30_000 }); - } - - async updateThreadSettings(params: ThreadSettingsUpdateParams): Promise { - await this.sendRequest('thread/settings/update', params, { timeoutMs: 30_000 }); - } - async listModels(params?: ModelListParams): Promise { const response = await this.sendRequest('model/list', params ?? {}, { timeoutMs: 30_000 diff --git a/cli/src/codex/codexLocalLauncher.ts b/cli/src/codex/codexLocalLauncher.ts index a5368ee8..e50f43d6 100644 --- a/cli/src/codex/codexLocalLauncher.ts +++ b/cli/src/codex/codexLocalLauncher.ts @@ -48,7 +48,7 @@ function extractTurnContextModel(event: CodexSessionEvent): string | null | unde const model = (event.payload as Record).model; if (model === null) return null; if (typeof model !== 'string' || !model.trim()) return undefined; - return model.trim() === 'gpt-reserve' ? 'gpt-5.6-luna' : model.trim(); + return model.trim(); } export async function codexLocalLauncher(session: CodexSession): Promise<'switch' | 'exit'> { @@ -445,7 +445,7 @@ export async function codexLocalLauncher(session: CodexSession): Promise<'switch await codexLocal({ path: session.path, sessionId: resumeSessionId, - modelReasoningEffort: resumeSessionId ? undefined : (session.getModelReasoningEffort() ?? undefined) as ReasoningEffort | undefined, + modelReasoningEffort: (session.getModelReasoningEffort() ?? undefined) as ReasoningEffort | undefined, onSessionFound: handleSessionFound, abort: abortSignal, codexArgs, diff --git a/cli/src/codex/codexRemoteLauncher.test.ts b/cli/src/codex/codexRemoteLauncher.test.ts index 3f3fbb6b..83a71a3a 100644 --- a/cli/src/codex/codexRemoteLauncher.test.ts +++ b/cli/src/codex/codexRemoteLauncher.test.ts @@ -1,16 +1,8 @@ import { afterEach, describe, expect, it, vi } from 'vitest'; -import { mkdtemp, rm } from 'node:fs/promises'; -import { tmpdir } from 'node:os'; -import { join } from 'node:path'; import { MessageQueue2 } from '@/utils/MessageQueue2'; import type { EnhancedMode } from './loop'; const harness = vi.hoisted(() => ({ - appServerConnected: true, - abandonTransport: null as (() => void) | null, - resumeModel: 'gpt-5.4', - reserveUsage: null as Record | null, - reserveSettingsCalls: [] as Array>, notifications: [] as Array<{ method: string; params: unknown }>, dispatchNotification: null as ((method: string, params: unknown) => void) | null, registerRequestCalls: [] as string[], @@ -112,14 +104,14 @@ vi.mock('./codexAppServerClient', () => { private notificationHandler: ((method: string, params: unknown) => void) | null = null; private stderrHandler: ((text: string) => void) | null = null; - async connect(): Promise { harness.appServerConnected = true; } + async connect(): Promise {} isConnected(): boolean { - return harness.appServerConnected; + return true; } isInitialized(): boolean { - return harness.appServerConnected; + return true; } async initialize(params: unknown): Promise<{ protocolVersion: number }> { @@ -140,7 +132,7 @@ vi.mock('./codexAppServerClient', () => { harness.dispatchNotification = handler; } - setTransportAbandonedHandler(handler: (() => void) | null): void { harness.abandonTransport = handler; } + setTransportAbandonedHandler(_handler: (() => void) | null): void {} setStderrHandler(handler: ((text: string) => void) | null): void { this.stderrHandler = handler; @@ -154,25 +146,6 @@ vi.mock('./codexAppServerClient', () => { return harness.collaborationModeResponse; } - async readAccountRateLimits(): Promise { - if (!harness.reserveUsage) throw new Error('Unsupported'); - return structuredClone(harness.reserveUsage); - } - - async supportsMethod(): Promise { return true; } - - async listModels(): Promise { - return { data: [ - { id: 'gpt-reserve', hidden: true, defaultReasoningEffort: 'medium' }, - { id: 'gpt-5.4', hidden: false, defaultReasoningEffort: 'high' } - ] }; - } - - async updateThreadSettings(params: Record): Promise { - harness.reserveSettingsCalls.push(params); - this.notificationHandler?.('thread/settings/updated', { threadId: params.threadId, threadSettings: params }); - } - async listSkills(params: unknown): Promise { harness.listSkillsCalls.push(params); return harness.skillsListResponse; @@ -205,7 +178,7 @@ vi.mock('./codexAppServerClient', () => { if (harness.failResumeThreadIds.includes(id)) { throw new Error('resume failed'); } - return { thread: { id }, model: harness.resumeModel }; + return { thread: { id }, model: 'gpt-5.4' }; } async compactThread(params?: { threadId?: string }): Promise> { @@ -1136,7 +1109,6 @@ import { INDETERMINATE_SYMBOL } from './codexAppServerClient'; import { RPC_METHODS } from '@hapi/protocol/rpcMethods'; type FakeAgentState = { - codexUsage?: import('@hapi/protocol').AgentState['codexUsage']; requests: Record; completedRequests: Record; }; @@ -1232,10 +1204,6 @@ function createSessionStub( setModelReasoningEffort(nextEffort: EnhancedMode['modelReasoningEffort']) { currentModelReasoningEffort = nextEffort; }, - getModelReasoningEffort() { return currentModelReasoningEffort; }, - pushKeepAlive() {}, - getServiceTier() { return undefined; }, - setServiceTier() {}, getCollaborationMode() { return currentCollaborationMode; }, @@ -1431,11 +1399,6 @@ describe('codexRemoteLauncher', () => { session.queue.close(); }); afterEach(() => { - harness.appServerConnected = true; - harness.abandonTransport = null; - harness.resumeModel = 'gpt-5.4'; - harness.reserveUsage = null; - harness.reserveSettingsCalls = []; harness.notifications = []; harness.dispatchNotification = null; harness.registerRequestCalls = []; @@ -2695,118 +2658,6 @@ describe('codexRemoteLauncher', () => { expect(session.thinking).toBe(false); }); - it('preserves queued model settings when resuming an ordinary thread', async () => { - const { session } = createSessionStub(['first'], { - ...createMode(), model: 'gpt-5.6', modelReasoningEffort: 'high', serviceTier: 'fast' - }, true); - session.sessionId = 'thread-old'; - await codexRemoteLauncher(session as never); - expect(harness.startTurnParams[0]).toMatchObject({ - collaborationMode: { settings: { model: 'gpt-5.6', reasoning_effort: 'high' } }, - serviceTier: 'priority' - }); - }); - - it.each([ - ['transient', 'Codex thread entered systemError', 2, false], - ['context', "Codex ran out of room in the model's context window.", 2, true], - ['usage', 'Usage limit exceeded', 1, false] - ] as const)('preserves %s failure handling during Reserve', async (_kind, error, attempts, compact) => { - harness.resumeModel = 'gpt-reserve'; - harness.remainingThreadSystemErrors = 1; - harness.nextThreadSystemErrorMessage = error; - harness.reserveUsage = { - accountId: 'account-a', ordinaryUsageAllowed: false, - rateLimits: { limitId: 'codex', primary: { usedPercent: 100 } } - }; - const { session } = createSessionStub(['first'], createMode(), true); - session.sessionId = 'thread-reserve'; - await codexRemoteLauncher(session as never); - expect(harness.startTurnMessages).toEqual(Array(attempts).fill('first')); - expect(harness.compactThreadIds).toEqual(compact ? ['thread-reserve'] : []); - expect(harness.startThreadIds).toEqual([]); - expect(harness.startTurnParams[0]).toMatchObject({ - collaborationMode: { settings: { model: 'gpt-reserve' } } - }); - }); - - it('reconciles a resumed Reserve task before waiting for the first user message', async () => { - harness.resumeModel = 'gpt-reserve'; - harness.reserveUsage = { - accountId: 'account-a', ordinaryUsageAllowed: false, - rateLimits: { limitId: 'codex', primary: { usedPercent: 100 } }, - rateLimitsByLimitId: { base_model_inference: { limitName: 'gpt-reserve', primary: { usedPercent: 20 } } } - }; - const { session, getModel, getAgentState } = createSessionStub([], createMode(), false, false); - session.sessionId = 'thread-reserve'; - const run = codexRemoteLauncher(session as never); - try { - await vi.waitFor(() => expect(getAgentState().codexUsage?.reserve?.primary?.remainingPercent).toBe(80)); - expect(getModel()).toBe('gpt-5.6-luna'); - expect(harness.resumeThreadIds).toEqual(['thread-reserve']); - expect(harness.startTurnMessages).toEqual([]); - } finally { - session.queue.close(); - await run; - } - }); - - it('reads native settings after transport loss without replaying the interrupted message', async () => { - harness.resumeModel = 'gpt-reserve'; - harness.suppressTurnCompletion = true; - harness.reserveUsage = { - accountId: 'account-a', ordinaryUsageAllowed: false, - rateLimits: { limitId: 'codex', primary: { usedPercent: 100 } } - }; - const { session, getModel, getAgentState } = createSessionStub(['first'], createMode(), false, false); - session.sessionId = 'thread-reserve'; - const run = codexRemoteLauncher(session as never); - try { - await vi.waitFor(() => expect(harness.startTurnMessages).toEqual(['first'])); - harness.resumeModel = 'gpt-5.4'; - harness.reserveUsage.ordinaryUsageAllowed = true; - harness.appServerConnected = false; - harness.abandonTransport?.(); - await vi.waitFor(() => expect(harness.resumeThreadIds).toEqual(['thread-reserve', 'thread-reserve'])); - await vi.waitFor(() => expect(getAgentState().codexUsage?.reserve).toBeNull()); - expect(getModel()).toBe('gpt-5.4'); - expect(harness.startTurnMessages).toEqual(['first']); - } finally { - session.queue.close(); - await run; - } - }); - - it('uses accepted Reserve settings for queued turns in the same conversation without replay', async () => { - const home = await mkdtemp(join(tmpdir(), 'hapi-reserve-launcher-')); - vi.stubEnv('CODEX_HOME', home); - try { - harness.reserveUsage = { - accountId: 'account-a', ordinaryUsageAllowed: false, - rateLimits: { limitId: 'codex', primary: { usedPercent: 100 } }, - rateLimitsByLimitId: { base_model_inference: { limitName: 'gpt-reserve', primary: { usedPercent: 20 } } }, - rateLimitUpsell: { banner_type: 'luna_reserve', blocked_model_slug: 'gpt-5.4' } - }; - const { session, getModel, sessionEvents } = createSessionStub(['first', 'second'], createMode(), true); - session.sessionId = '01900000-0000-7000-8000-000000000001'; - await codexRemoteLauncher(session as never); - expect(harness.resumeThreadIds).toEqual([session.sessionId]); - expect(harness.resumeThreadParams[0]).not.toHaveProperty('model'); - expect(harness.reserveSettingsCalls).toHaveLength(1); - expect(harness.startTurnMessages).toEqual(['first', 'second']); - expect(harness.startTurnParams).toHaveLength(2); - for (const params of harness.startTurnParams) { - expect(params.collaborationMode).toMatchObject({ settings: { model: 'gpt-reserve', reasoning_effort: 'medium' } }); - expect(params.threadId).toBe(session.sessionId); - } - expect(getModel()).toBe('gpt-5.6-luna'); - expect(sessionEvents.filter(event => String(event.message).includes('Luna Reserve'))).toHaveLength(1); - } finally { - vi.unstubAllEnvs(); - await rm(home, { recursive: true, force: true }); - } - }); - it('does not create a new thread when an existing conversation cannot be resumed', async () => { harness.failResumeThreadIds = ['thread-old']; const { session, sessionEvents } = createSessionStub(['first message']); diff --git a/cli/src/codex/codexRemoteLauncher.ts b/cli/src/codex/codexRemoteLauncher.ts index c686d65c..f2290f38 100644 --- a/cli/src/codex/codexRemoteLauncher.ts +++ b/cli/src/codex/codexRemoteLauncher.ts @@ -18,7 +18,6 @@ import { registerGeneratedImageFromPath } from '@/modules/common/generatedImages import { registerAppServerPermissionHandlers } from './utils/appServerPermissionAdapter'; import { buildThreadStartParams, - withoutCodexModelOverrides, buildTurnStartParams, type CodexContextManagementConfig } from './utils/appServerConfig'; @@ -39,7 +38,6 @@ import { type RemoteLauncherExitReason } from '@/modules/common/remote/RemoteLauncherBase'; import { CodexConversationHistory } from './conversationHistory'; -import { LunaReserve } from './utils/lunaReserve'; @@ -361,9 +359,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { }); } - private lunaReserve: LunaReserve | null = null; - private usageTimer: ReturnType | null = null; - protected async runMainLoop(): Promise { const session = this.session; const messageBuffer = this.messageBuffer; @@ -1867,7 +1862,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { } }; - let hasThread = false; let activeMessage: QueuedMessage | null = null; let sameThreadRetryAttempt = 0; let sameThreadCompactAttempt = 0; @@ -1922,9 +1916,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { if (session.thinking) session.onThinkingChange(false); recoveryInFlight = false; activeMessage = null; - hasThread = false; - this.lunaReserve?.detach(); - this.currentThreadId = null; this.currentTurnId = null; wakeLoop(); }); @@ -2056,7 +2047,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { experimentalApi: true } }); - await this.lunaReserve?.initialize(); }; // Returns 'accepted' when the thread contains the steered user message @@ -2920,9 +2910,7 @@ class CodexRemoteLauncher extends RemoteLauncherBase { const isThreadStatusFailure = msgType === 'task_failed' && msg.terminal_source === 'thread_status'; const error = msgType === 'task_failed' ? asString(msg.error) : null; const explicitlyNonRetryable = msgType === 'task_failed' - && (msg.retryable === false || isPolicyBlockedCodexFailure(msg, error) - || msg.codex_error_info === 'usageLimitExceeded' - || /usage limit|rate limit|quota/i.test(error ?? '')); + && (msg.retryable === false || isPolicyBlockedCodexFailure(msg, error)); if (deferredThreadStatusFailure && isTerminalEvent && !isThreadStatusFailure) { const sameThread = !eventThreadId || eventThreadId === deferredThreadStatusFailure.threadId; @@ -3504,37 +3492,7 @@ class CodexRemoteLauncher extends RemoteLauncherBase { } }; - const reserve = new LunaReserve(appServerClient, (model, effort, serviceTier) => { - session.setModel(model); - session.setModelReasoningEffort(effort); - if (serviceTier !== undefined) session.setServiceTier(serviceTier); - session.pushKeepAlive(); - }, (codexUsage) => { - session.client.updateAgentState(state => ({ ...state, codexUsage })); - }, message => session.sendSessionEvent({ type: 'message', message })); - this.lunaReserve = reserve; - const currentMode = (): EnhancedMode => ({ - permissionMode: 'default', - model: session.getModel() ?? undefined, - modelReasoningEffort: session.getModelReasoningEffort() ?? undefined, - collaborationMode: session.getCollaborationMode() ?? 'default', - serviceTier: session.getServiceTier() - }); - const refreshUsage = () => reserve.refresh(currentMode(), () => !turnInFlight && !this.shouldExit); - appServerClient.setNotificationHandler((method, params) => { - if (method === 'account/rateLimits/updated' || method === 'account/updated') { - reserve.invalidate(); - void refreshUsage(); - } - if (method === 'thread/settings/updated') { - const record = asRecord(params); - const threadId = asString(record?.threadId); - if (threadId) reserve.onSettings(threadId, record?.threadSettings); - } - if (method === 'error' || method === 'turn/completed') { - reserve.invalidate(); - } if (method === 'skills/changed') { void refreshNativeSkills(true).catch((error) => { logger.debug(`[Codex] failed to refresh skills: ${errorMessage(error)}`); @@ -3564,9 +3522,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { }); } } - if (method === 'error' || method === 'turn/completed') { - void (codexEventQueue ?? Promise.resolve()).then(refreshUsage); - } }); appServerClient.setStderrHandler((text) => { @@ -3624,10 +3579,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { } }); - await reserve.initialize(); - this.usageTimer = setInterval(() => { void refreshUsage(); }, 60_000); - this.usageTimer.unref(); - let contextManagementConfig: CodexContextManagementConfig | undefined; try { const effectiveConfig = (await appServerClient.readConfig({ @@ -3732,6 +3683,7 @@ class CodexRemoteLauncher extends RemoteLauncherBase { logger.debug(`[Codex] collaborationMode/list failed: ${errorMessage(error)}`); } + let hasThread = false; let pending: QueuedMessage | null = null; let suppressReadyForAdminCommand = false; @@ -3825,7 +3777,7 @@ class CodexRemoteLauncher extends RemoteLauncherBase { try { const resumeResponse = await appServerClient.resumeThread({ threadId: resumeCandidate, - ...withoutCodexModelOverrides(threadParams) + ...threadParams }, { signal: this.abortController.signal }); @@ -3833,7 +3785,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { const resumeThread = resumeRecord ? asRecord(resumeRecord.thread) : null; const threadId = asString(resumeThread?.id) ?? resumeCandidate; applyResolvedModel(resumeRecord?.model); - reserve.attach(threadId, resumeResponse); this.currentThreadId = threadId; this.conversationHistory.setThreadId(threadId); void this.conversationHistory.probeCapabilities().catch(() => {}); @@ -3871,7 +3822,7 @@ class CodexRemoteLauncher extends RemoteLauncherBase { } }; - const ensureThread = async (mode: EnhancedMode): Promise => { + const ensureThreadForGoal = async (mode: EnhancedMode): Promise => { if (this.currentThreadId && this.currentThreadId !== invalidThreadId) { hasThread = true; return this.currentThreadId; @@ -3891,7 +3842,7 @@ class CodexRemoteLauncher extends RemoteLauncherBase { try { const resumeResponse = await appServerClient.resumeThread({ threadId: resumeCandidate, - ...withoutCodexModelOverrides(threadParams) + ...threadParams }, { signal: this.abortController.signal }); @@ -3899,7 +3850,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { const resumeThread = resumeRecord ? asRecord(resumeRecord.thread) : null; const threadId = asString(resumeThread?.id) ?? resumeCandidate; applyResolvedModel(resumeRecord?.model); - reserve.attach(threadId, resumeResponse); this.currentThreadId = threadId; this.conversationHistory.setThreadId(threadId); void this.conversationHistory.probeCapabilities().catch(() => {}); @@ -3907,8 +3857,8 @@ class CodexRemoteLauncher extends RemoteLauncherBase { hasThread = true; return threadId; } catch (error) { - logger.warn(`[Codex] Failed to resume app-server thread ${resumeCandidate}`, error); - sendVisibleStatus(`Codex conversation ${resumeCandidate} could not be resumed`); + logger.warn(`[Codex] Failed to resume app-server thread ${resumeCandidate} for /goal`, error); + sendVisibleStatus(`Goal failed: Codex conversation ${resumeCandidate} could not be resumed`); return null; } } @@ -3931,7 +3881,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { const thread = threadRecord ? asRecord(threadRecord.thread) : null; const threadId = asString(thread?.id); applyResolvedModel(threadRecord?.model); - if (threadId) reserve.attach(threadId, threadResponse); if (!threadId) { throw new Error('app-server thread/start did not return thread.id'); } @@ -3985,7 +3934,7 @@ class CodexRemoteLauncher extends RemoteLauncherBase { return true; } - const threadId = await ensureThread(message.mode); + const threadId = await ensureThreadForGoal(message.mode); if (!threadId) { return true; } @@ -4062,7 +4011,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { invalidThreadId = null; hasThread = false; session.resetCodexThread(); - reserve.detach(); sendVisibleStatus('Context was reset'); return true; } @@ -4097,19 +4045,10 @@ class CodexRemoteLauncher extends RemoteLauncherBase { return true; }; - const reconcileIdleThread = async () => { - if (!hasThread && session.queue.size() === 0 && session.sessionId && session.sessionId !== session.sourceSessionId) { - await ensureThread(currentMode()); - await refreshUsage(); - } - }; - await reconcileIdleThread(); - while (!this.shouldExit) { logActiveHandles('loop-top'); if (!appServerClient.isConnected() || !appServerClient.isInitialized()) { await ensureAppServerInitialized(); - await reconcileIdleThread(); } if (pendingSteerReconciliations.size > 0) { await runSteerReconciliation(); @@ -4197,13 +4136,13 @@ class CodexRemoteLauncher extends RemoteLauncherBase { const response = shouldForkImportedSource ? await appServerClient.forkThread({ threadId: resumeCandidate, - ...withoutCodexModelOverrides(threadParams) + ...threadParams }, { signal: this.abortController.signal }) : await appServerClient.resumeThread({ threadId: resumeCandidate, - ...withoutCodexModelOverrides(threadParams) + ...threadParams }, { signal: this.abortController.signal }); @@ -4211,7 +4150,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { const responseThread = responseRecord ? asRecord(responseRecord.thread) : null; threadId = asString(responseThread?.id) ?? resumeCandidate; applyResolvedModel(responseRecord?.model); - if (threadId) reserve.attach(threadId, response, shouldForkImportedSource ? resumeCandidate : undefined); logger.debug(shouldForkImportedSource ? `[Codex] Forked imported app-server thread ${resumeCandidate} -> ${threadId}` : `[Codex] Resumed app-server thread ${threadId}`); @@ -4237,7 +4175,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { const thread = threadRecord ? asRecord(threadRecord.thread) : null; threadId = asString(thread?.id); applyResolvedModel(threadRecord?.model); - if (threadId) reserve.attach(threadId, threadResponse); if (!threadId) { throw new Error('app-server thread/start did not return thread.id'); } @@ -4261,22 +4198,13 @@ class CodexRemoteLauncher extends RemoteLauncherBase { } } - const beforeRefresh = currentMode(); - await reserve.refresh(message.mode, () => !turnInFlight && !this.shouldExit); - const afterRefresh = currentMode(); - const transitioned = beforeRefresh.model !== afterRefresh.model - || beforeRefresh.modelReasoningEffort !== afterRefresh.modelReasoningEffort - || beforeRefresh.serviceTier !== afterRefresh.serviceTier; setTurnInFlight(true); this.conversationHistory.setBusy(true); allowAnonymousTerminalEvent = false; - const mode = reserve.turnMode({ + const mode = { ...message.mode, - model: transitioned ? afterRefresh.model : message.mode.model ?? afterRefresh.model, - modelReasoningEffort: transitioned ? afterRefresh.modelReasoningEffort - : message.mode.modelReasoningEffort ?? afterRefresh.modelReasoningEffort, - serviceTier: transitioned ? afterRefresh.serviceTier : message.mode.serviceTier ?? afterRefresh.serviceTier - }); + model: session.getModel() ?? message.mode.model + }; usageModel = typeof mode.model === 'string' && mode.model.trim() ? mode.model.trim() : null; @@ -4414,10 +4342,6 @@ class CodexRemoteLauncher extends RemoteLauncherBase { protected async cleanup(): Promise { logger.debug('[codex-remote]: cleanup start'); - if (this.usageTimer) clearInterval(this.usageTimer); - this.usageTimer = null; - this.lunaReserve?.dispose(); - this.lunaReserve = null; this.appServerClient.setTransportAbandonedHandler(null); this.appServerClient.setStderrHandler(null); try { diff --git a/cli/src/codex/shared/root.test.ts b/cli/src/codex/shared/root.test.ts index 9bfc6725..c69ea530 100644 --- a/cli/src/codex/shared/root.test.ts +++ b/cli/src/codex/shared/root.test.ts @@ -31,7 +31,10 @@ vi.mock('../utils/buildHapiMcpBridge', () => ({ buildHapiMcpBridge: async () => }) })); const cleanups: Array<() => Promise> = []; -afterEach(async () => { for (const cleanup of cleanups.splice(0)) await cleanup(); }); +afterEach(async () => { + try { for (const cleanup of cleanups.splice(0)) await cleanup(); } + finally { vi.useRealTimers(); } +}); async function fixture() { const directory = await mkdtemp('/tmp/hapi-shared-root-'); @@ -45,7 +48,7 @@ async function fixture() { updateAgentState: updateState, keepAlive() {}, onUserMessage() {}, onCancelQueuedMessage() {}, onRetryQueuedMessage() {}, onReconnect: (fn: (() => void) | null) => { reconnect = fn; }, - rpcHandlerManager: { registerHandler() {} }, sendSessionEvent() {}, sendAgentMessage() {}, + rpcHandlerManager: { registerHandler() {} }, sendSessionEvent() {}, sendAgentMessage() {}, emitSessionReady() {}, sendSessionDeath() {}, async flush() {}, close() {} } as unknown as ApiSessionClient; const root = new SharedCodexRoot({ session, workingDirectory: directory } as SessionBootstrapResult, { @@ -66,6 +69,21 @@ async function fixture() { } describe('shared steering availability', () => { + it('keeps idle sessions online without polling usage or publishing agent-state updates', async () => { + const f = await fixture(); + const requests = vi.spyOn(f.root.client, 'request'); + const heartbeat = vi.spyOn(f.root.session, 'keepAlive'); + const updates = f.updateState.mock.calls.length; + vi.useFakeTimers(); + + await f.root.activate(); + await vi.advanceTimersByTimeAsync(5 * 60_000); + + expect(heartbeat).toHaveBeenCalled(); + expect(requests).not.toHaveBeenCalled(); + expect(f.updateState).toHaveBeenCalledTimes(updates); + }); + it('publishes root turn transitions, ignores child turns, and clears on shutdown', async () => { const f = await fixture(); expect(f.state().steeringActive).toBe(false); diff --git a/cli/src/codex/shared/root.ts b/cli/src/codex/shared/root.ts index 9bd5cca0..9e4bbde7 100644 --- a/cli/src/codex/shared/root.ts +++ b/cli/src/codex/shared/root.ts @@ -12,7 +12,6 @@ import { CodexAppServerClient } from '../codexAppServerClient'; import { buildHapiMcpBridge, type HapiMcpBridge } from '../utils/buildHapiMcpBridge'; import { buildUserInputFromMessage } from '../utils/appServerConfig'; import { resolveCodexPermissionModeConfig } from '../utils/permissionModeConfig'; -import { LunaReserve } from '../utils/lunaReserve'; import { resolveCodexSlashCommand } from '../utils/slashCommands'; import { parseReasoningEffortValue } from '../utils/reasoningEffort'; import { SharedCodexPermissions } from './permissions'; @@ -49,9 +48,7 @@ export class SharedCodexRoot { private projection!: SharedCodexProjection; private readonly children = new Map(); private readonly ancestry = new Map(); - private readonly reserve: LunaReserve; private heartbeat?: ReturnType; - private usageTimer?: ReturnType; private work: Promise = Promise.resolve(); private notifications = Promise.resolve(); private currentTurn: string | undefined; @@ -75,9 +72,6 @@ export class SharedCodexRoot { constructor(readonly bootstrap: SessionBootstrapResult, private readonly host: RootHost) { this.session = bootstrap.session; this.client = new CodexAppServerClient({ endpoint: host.endpoint, token: host.token, cwd: bootstrap.workingDirectory }); - this.reserve = new LunaReserve(this.client, - (model, effort, serviceTier) => { this.settings = { ...this.settings, model, modelReasoningEffort: effort, serviceTier }; }, - codexUsage => this.session.updateAgentState(state => ({ ...state, codexUsage })), message => this.notice(message)); this.client.setNotificationHandler((method, params) => { if (method === 'serverRequest/resolved') { this.permissions?.resolved(string(record(params).threadId) ?? '', record(params).requestId); return; @@ -105,7 +99,6 @@ export class SharedCodexRoot { const text = formatMessageWithAttachments(message.content.text, message.content.attachments); const resolved = text.trim().startsWith('/') ? await this.queue.command(id, () => this.command(text)) : text; if (resolved === null) { this.session.emitMessagesConsumed([id], { clearQueuedThinkingGrace: true }); return; } - await this.refreshUsage(); await this.queue.enqueue(id, buildUserInputFromMessage(resolved), this.interrupted); }).catch(error => this.notice(`Message not confirmed: ${error instanceof Error ? error.message : error}. Inspect the queue before retrying.`)); }); @@ -186,7 +179,7 @@ export class SharedCodexRoot { [id, { ...request, completedAt: Date.now(), status: 'canceled' as const }])) } })); if (subscribe) response = record(await this.client.request('thread/resume', { threadId })); - this.acceptSettings(response); this.acceptSettings(this.host.settingsFor(threadId) ?? {}); this.reserve.attach(threadId, response); + this.acceptSettings(response); this.acceptSettings(this.host.settingsFor(threadId) ?? {}); await this.projection.history(response.thread); await this.refresh(); await this.refreshChildren(true); } async activate(options: SharedLaunchOptions = {}): Promise { @@ -195,8 +188,6 @@ export class SharedCodexRoot { if (this.stopping) throw new Error('Codex execution is stopping'); this.registerControls(); this.ready(); this.heartbeat = setInterval(() => this.alive(), 2_000); this.alive(); this.session.emitSessionReady(); - void this.reserve.initialize().catch(error => logger.debug('[Codex shared] usage initialization', error)); - this.usageTimer = setInterval(() => { void this.refreshUsage(); }, 60_000); } private publishSteering(): void { const active = Boolean(this.currentTurn) && !this.stopping && !this.closed && !this.reconnecting && this.client.isInitialized(); @@ -220,7 +211,6 @@ export class SharedCodexRoot { ...('serviceTier' in value ? { serviceTier: value.serviceTier === 'priority' ? 'fast' : 'standard' } : {}), ...('collaborationMode' in value ? { collaborationMode: record(value.collaborationMode).mode === 'plan' ? 'plan' as const : 'default' as const } : {}) }; - this.reserve.onSettings(this.threadId, value); for (const listener of this.settingsListeners) listener(); } private async notification(method: string, params: unknown, modelAtReceipt?: string): Promise { @@ -303,11 +293,6 @@ export class SharedCodexRoot { } catch (error) { logger.debug('[Codex shared] child history unavailable', { id, error }); } } } - private refreshUsage(): Promise { - return this.reserve.refresh({ permissionMode: this.settings.permissionMode === 'yolo' ? 'yolo' : 'default', - collaborationMode: this.settings.collaborationMode ?? 'default', model: this.settings.model ?? undefined }, - () => !this.closed && !this.currentTurn); - } nativeQueueDeleted(nativeId: string): Promise { return this.queue.deleted(nativeId); } replaySettings(): Array<{ method: string; params: unknown }> { return this.settingsNotification ? [{ method: 'thread/settings/updated', params: { @@ -482,9 +467,9 @@ export class SharedCodexRoot { return slash.kind === 'replace' ? slash.text : null; } stopAccepting(): void { - this.stopping = true; this.ready(); clearInterval(this.usageTimer); + this.stopping = true; this.ready(); this.publishSteering(); - this.session.onReconnect(null); this.client.setTransportAbandonedHandler(null); this.reserve.dispose(); + this.session.onReconnect(null); this.client.setTransportAbandonedHandler(null); } async suspend(): Promise { this.stopAccepting(); diff --git a/cli/src/codex/utils/appServerConfig.ts b/cli/src/codex/utils/appServerConfig.ts index 16c75f5b..fcd70510 100644 --- a/cli/src/codex/utils/appServerConfig.ts +++ b/cli/src/codex/utils/appServerConfig.ts @@ -332,13 +332,3 @@ export function buildTurnStartParams(args: { return params; } - -/** Resume the actual task settings before reconciling HAPI's queued configuration. */ -export function withoutCodexModelOverrides(params: ThreadStartParams): ThreadStartParams { - const { model, serviceTier, ...rest } = params; - const config = { ...rest.config }; - delete config.model; - delete config.model_reasoning_effort; - delete config.service_tier; - return { ...rest, config }; -} diff --git a/cli/src/codex/utils/lunaReserve.test.ts b/cli/src/codex/utils/lunaReserve.test.ts deleted file mode 100644 index 114d0629..00000000 --- a/cli/src/codex/utils/lunaReserve.test.ts +++ /dev/null @@ -1,175 +0,0 @@ -import { afterEach, describe, expect, it, vi } from 'vitest'; -import { mkdtemp, readFile, rm } from 'node:fs/promises'; -import { tmpdir } from 'node:os'; -import { join } from 'node:path'; -import { LunaReserve } from './lunaReserve'; -import { withoutCodexModelOverrides } from './appServerConfig'; -import type { EnhancedMode } from '../loop'; - -const threadId = '01900000-0000-7000-8000-000000000001'; -const mode: EnhancedMode = { permissionMode: 'yolo', collaborationMode: 'plan', model: 'gpt-5.6', modelReasoningEffort: 'high' }; -const dirs: string[] = []; -afterEach(async () => { await Promise.all(dirs.splice(0).map(dir => rm(dir, { recursive: true, force: true }))); }); - -async function fixture() { - const dir = await mkdtemp(join(tmpdir(), 'hapi-reserve-')); - dirs.push(dir); - const response = { - accountId: 'account-a', ordinaryUsageAllowed: true as boolean | null, - rateLimits: { limitId: 'codex', primary: { usedPercent: 40 }, secondary: { usedPercent: 60 }, spendControlReached: false, rateLimitReachedType: null as string | null }, - rateLimitsByLimitId: { - base_model_inference: { limitId: 'base_model_inference', limitName: 'gpt-reserve', primary: { usedPercent: 10 as unknown, resetsAt: 2000000000, windowDurationMins: 123 }, secondary: { usedPercent: 25 } } - }, - rateLimitUpsell: null as unknown - }; - const client = { - readAccountRateLimits: vi.fn(async () => structuredClone(response) as unknown), - supportsMethod: vi.fn(async () => true), - listModels: vi.fn(async () => ({ data: [ - { id: 'gpt-reserve', hidden: true, defaultReasoningEffort: 'medium', supportedReasoningEfforts: [{ reasoningEffort: 'medium' }] }, - { id: 'gpt-5.6', hidden: false, defaultReasoningEffort: 'medium', supportedReasoningEfforts: [{ reasoningEffort: 'high' }] } - ] })), - updateThreadSettings: vi.fn(async () => {}) - }; - const sync = vi.fn(); const publish = vi.fn(); const notice = vi.fn(); - const reserve = new LunaReserve(client, sync, publish, notice, dir); - await reserve.initialize(); - reserve.attach(threadId, { model: 'gpt-5.6', reasoningEffort: 'high', serviceTier: 'priority' }); - const enter = async () => { - response.ordinaryUsageAllowed = false; - response.rateLimitUpsell = { banner_type: 'luna_reserve', blocked_model_slug: 'gpt-5.6' }; - await reserve.refresh(mode, () => true); - }; - const usage = () => publish.mock.lastCall?.[0]; - return { dir, response, client, reserve, sync, publish, notice, enter, usage }; -} - -describe('Luna Reserve authoritative transitions', () => { - it('hides an available bucket and does not infer eligibility from exhausted percentages', async () => { - const f = await fixture(); - f.response.rateLimits.primary.usedPercent = 100; - f.response.rateLimits.secondary.usedPercent = 100; - await f.reserve.refresh(mode, () => true); - expect(f.usage().reserve).toBeNull(); - expect(f.client.updateThreadSettings).not.toHaveBeenCalled(); - f.response.ordinaryUsageAllowed = false; - await f.reserve.refresh(mode, () => true); - expect(f.client.updateThreadSettings).not.toHaveBeenCalled(); - }); - - it('activates only after acceptance, separates buckets, preserves return settings, and hides on recovery', async () => { - const f = await fixture(); - await f.enter(); - expect(f.client.readAccountRateLimits).toHaveBeenLastCalledWith(true); - expect(f.client.updateThreadSettings).toHaveBeenCalledWith({ threadId, model: 'gpt-reserve', effort: 'medium', serviceTier: null, - collaborationMode: { mode: 'plan', settings: { model: 'gpt-reserve', reasoning_effort: 'medium', developer_instructions: null } } }); - expect(f.usage().ordinary.secondary.remainingPercent).toBe(40); - expect(f.usage().reserve.secondary.remainingPercent).toBe(75); - expect(f.usage().reserve.primary).toEqual({ remainingPercent: 90, resetsAt: 2000000000, windowDurationMins: 123 }); - expect(f.sync).toHaveBeenLastCalledWith('gpt-5.6-luna', 'medium', 'standard'); - expect(JSON.parse(await readFile(join(f.dir, 'tui-luna-reserve', `${threadId}.json`), 'utf8'))).toEqual({ account_id: 'account-a', model: 'gpt-5.6', effort: 'high' }); - expect(f.reserve.turnMode(mode)).toMatchObject({ model: 'gpt-reserve', modelReasoningEffort: 'medium', serviceTier: 'standard' }); - f.response.ordinaryUsageAllowed = true; - f.response.rateLimitUpsell = null; - await f.reserve.refresh(mode, () => true); - expect(f.sync).toHaveBeenLastCalledWith('gpt-5.6', 'high', 'standard'); - expect(f.usage().reserve).toBeNull(); - expect(f.notice).toHaveBeenCalledTimes(2); - }); - - it('retains exhausted Reserve without another switch or retry and treats missing/invalid usage as unknown', async () => { - const f = await fixture(); await f.enter(); - for (const used of [undefined, null, '50', NaN, Infinity, -1, 101, 100]) { - f.response.rateLimitsByLimitId.base_model_inference.primary.usedPercent = used; - await f.reserve.refresh(mode, () => true); - expect(f.usage().reserve.primary.remainingPercent).toBe(used === 100 ? 0 : null); - } - expect(f.client.updateThreadSettings).toHaveBeenCalledTimes(1); - }); - - it('requires a fresh account-bound recovery read; sparse notifications and unknown banners cannot recover', async () => { - const f = await fixture(); await f.enter(); - f.response.ordinaryUsageAllowed = true; - f.response.rateLimitUpsell = { banner_type: 'unknown_future_banner' }; - await f.reserve.refresh(mode, () => true); - f.response.rateLimitUpsell = null; - f.response.accountId = 'account-b'; - await f.reserve.refresh(mode, () => true); - expect(f.client.updateThreadSettings).toHaveBeenCalledTimes(1); - f.response.accountId = 'account-a'; - let finish!: (value: unknown) => void; - f.client.readAccountRateLimits.mockImplementationOnce(() => new Promise(resolve => { finish = resolve; })); - const pending = f.reserve.refresh(mode, () => true); - await vi.waitFor(() => expect(finish).toBeTypeOf('function')); - f.reserve.invalidate(); - finish(f.response); await pending; - expect(f.client.updateThreadSettings).toHaveBeenCalledTimes(1); - }); - - it('reconciles a resumed Reserve task with the TUI return target and ignores another thread settings', async () => { - const f = await fixture(); await f.enter(); f.reserve.dispose(); - const resumed = new LunaReserve(f.client, f.sync, f.publish, f.notice, f.dir); - await resumed.initialize(); - resumed.attach(threadId, { model: 'gpt-reserve', reasoningEffort: 'medium' }); - resumed.onSettings('other-thread', { model: 'gpt-5.6', effort: 'high' }); - expect(resumed.turnMode(mode).model).toBe('gpt-reserve'); - f.response.ordinaryUsageAllowed = true; f.response.rateLimitUpsell = null; - await resumed.refresh(mode, () => true); - expect(f.sync).toHaveBeenLastCalledWith('gpt-5.6', 'high', 'standard'); - expect(f.usage().reserve).toBeNull(); - }); - - it('copies the native fork return target without deleting the parent recovery record', async () => { - const f = await fixture(); await f.enter(); - const forkId = '01900000-0000-7000-8000-000000000002'; - f.reserve.attach(forkId, { model: 'gpt-reserve', reasoningEffort: 'medium', thread: { forkedFromId: threadId } }); - await f.reserve.refresh(mode, () => true); - const parent = join(f.dir, 'tui-luna-reserve', `${threadId}.json`); - const child = join(f.dir, 'tui-luna-reserve', `${forkId}.json`); - expect(await readFile(child, 'utf8')).toBe(await readFile(parent, 'utf8')); - f.response.ordinaryUsageAllowed = true; f.response.rateLimitUpsell = null; - await f.reserve.refresh(mode, () => true); - expect(f.client.updateThreadSettings).toHaveBeenLastCalledWith(expect.objectContaining({ threadId: forkId, model: 'gpt-5.6' })); - expect(JSON.parse(await readFile(parent, 'utf8')).model).toBe('gpt-5.6'); - await expect(readFile(child, 'utf8')).rejects.toThrow(); - f.reserve.detach(); - expect(f.usage()).toBeNull(); - }); - - it('keeps selection and visibility unchanged on a rejected settings update without looping', async () => { - const f = await fixture(); - f.client.updateThreadSettings.mockRejectedValue(new Error('Method not found')); - await f.enter(); await f.reserve.refresh(mode, () => true); - expect(f.client.updateThreadSettings).toHaveBeenCalledTimes(1); - expect(f.usage().reserve).toBeNull(); - expect(f.reserve.turnMode(mode).model).toBe('gpt-5.6'); - expect(f.notice).toHaveBeenCalledTimes(1); - }); - - it('does not advertise or activate Reserve on an older protocol', async () => { - const f = await fixture(); - f.client.readAccountRateLimits.mockImplementation(async () => ({ rateLimits: f.response.rateLimits, accountId: 'account-a', rateLimitUpsell: { banner_type: 'luna_reserve' } })); - const old = new LunaReserve(f.client, f.sync, f.publish, f.notice, f.dir); - await old.initialize(); old.attach(threadId, { model: 'gpt-5.6' }); - await old.refresh(mode, () => true); - expect(f.client.readAccountRateLimits).toHaveBeenLastCalledWith(false); - expect(f.client.updateThreadSettings).not.toHaveBeenCalled(); - }); - - it('waits until idle and hides state after exit/disposal', async () => { - const f = await fixture(); - f.response.ordinaryUsageAllowed = false; f.response.rateLimitUpsell = { banner_type: 'luna_reserve' }; - await f.reserve.refresh(mode, () => false); - expect(f.client.updateThreadSettings).not.toHaveBeenCalled(); - await f.enter(); - f.reserve.onSettings(threadId, { model: 'gpt-5.6', effort: 'high' }); - expect(f.usage().reserve).toBeNull(); - f.reserve.dispose(); - expect(f.usage()).toBeNull(); - }); - - it('omits model overrides on resume while retaining permissions, tools and instructions', () => { - expect(withoutCodexModelOverrides({ model: 'stale', serviceTier: 'priority', sandbox: 'workspace-write', config: { model_reasoning_effort: 'high', 'mcp_servers.hapi': {}, developer_instructions: 'hapi' } })) - .toEqual({ sandbox: 'workspace-write', config: { 'mcp_servers.hapi': {}, developer_instructions: 'hapi' } }); - }); -}); diff --git a/cli/src/codex/utils/lunaReserve.ts b/cli/src/codex/utils/lunaReserve.ts deleted file mode 100644 index 629dcd02..00000000 --- a/cli/src/codex/utils/lunaReserve.ts +++ /dev/null @@ -1,301 +0,0 @@ -import { mkdir, open, rename, rm } from 'node:fs/promises'; -import { homedir } from 'node:os'; -import { join } from 'node:path'; -import { randomUUID } from 'node:crypto'; -import { z } from 'zod'; -import { logger } from '@/ui/logger'; -import type { AgentState } from '@hapi/protocol'; -import type { CodexAppServerClient } from '../codexAppServerClient'; -import type { EnhancedMode } from '../loop'; -import type { ModelListItem, ThreadSettingsUpdateParams } from '../appServerTypes'; - -// This quota alias belongs only at the Codex boundary, never in HAPI's model picker. -const RESERVE_MODEL = 'gpt-reserve'; -const LUNA_MODEL = 'gpt-5.6-luna'; -const WindowSchema = z.object({ - usedPercent: z.unknown().optional(), - windowDurationMins: z.unknown().optional(), - resetsAt: z.unknown().optional() -}); -const BucketSchema = z.object({ - limitId: z.string().nullish(), - limitName: z.string().nullish(), - normalModelSlug: z.string().nullish(), - primary: WindowSchema.nullish(), - secondary: WindowSchema.nullish(), - credits: z.object({ hasCredits: z.boolean(), unlimited: z.boolean() }).nullish(), - spendControlReached: z.boolean().nullish(), - rateLimitReachedType: z.string().nullish() -}); -const UsageSchema = z.object({ - accountId: z.string().nullish(), - ordinaryUsageAllowed: z.boolean().nullish(), - rateLimits: BucketSchema, - rateLimitsByLimitId: z.record(z.string(), BucketSchema).nullish(), - rateLimitUpsell: z.unknown().optional() -}); -const BannerSchema = z.object({ - banner_type: z.literal('luna_reserve'), - blocked_model_slug: z.string().min(1).max(256).nullish() -}); -const ReturnSchema = z.object({ - account_id: z.string().min(1), - model: z.string().min(1).max(256), - effort: z.string().nullable() -}); -const SettingsSchema = z.object({ - model: z.string().min(1), - effort: z.string().nullish(), - serviceTier: z.string().nullish() -}); -type Usage = z.infer; -type Bucket = z.infer; -type Settings = z.infer; -type ReserveClient = Pick; - -function usageBucket(bucket?: Bucket | null): NonNullable['ordinary'] { - const window = (value: Bucket['primary']) => { - if (!value) return null; - const finite = (n: unknown): n is number => typeof n === 'number' && Number.isFinite(n); - return { - remainingPercent: finite(value.usedPercent) && value.usedPercent >= 0 && value.usedPercent <= 100 - ? 100 - value.usedPercent : null, - windowDurationMins: finite(value.windowDurationMins) && value.windowDurationMins > 0 ? value.windowDurationMins : null, - resetsAt: finite(value.resetsAt) && value.resetsAt >= 0 ? value.resetsAt : null - }; - }; - return { primary: window(bucket?.primary), secondary: window(bucket?.secondary) }; -} - -/** Account reads authorize transitions; settings accepted by the thread establish activation. */ -export class LunaReserve { - private threadId: string | null = null; - private settings: Settings | null = null; - private settingsRevision = 0; - private forkedFrom: string | null = null; - private snapshot: Usage | null = null; - private models: ModelListItem[] = []; - private supported = false; - private revision = 0; - private disposed = false; - private queue: Promise = Promise.resolve(); - - constructor( - private readonly client: ReserveClient, - private readonly sync: (model: string, effort: string | null, serviceTier: string | null | undefined) => void, - private readonly publish: (usage: AgentState['codexUsage']) => void, - private readonly notice: (message: string) => void, - private readonly codexHome = process.env.CODEX_HOME || join(homedir(), '.codex') - ) {} - - async initialize(): Promise { - this.supported = false; - this.models = []; - try { - const response = UsageSchema.parse(await this.client.readAccountRateLimits()); - this.snapshot = response; - // Field presence is a protocol capability, not account eligibility. - if ('ordinaryUsageAllowed' in response && await this.client.supportsMethod('thread/settings/update')) { - let cursor: string | undefined; - do { - const page = await this.client.listModels({ includeHidden: true, ...(cursor ? { cursor } : {}) }); - this.models.push(...(page.data ?? [])); - cursor = page.nextCursor ?? undefined; - } while (cursor); - this.supported = this.models.some(model => (model.model ?? model.id) === RESERVE_MODEL); - } - this.publishUsage(); - } catch { - // Older app-servers and API-key accounts can lack the usage endpoint. - this.publish(null); - } - } - - attach(threadId: string, response: unknown, forkedFrom?: string): void { - const parsed = z.object({ model: z.string(), reasoningEffort: z.string().nullish(), serviceTier: z.string().nullish(), thread: z.object({ forkedFromId: z.string().nullish() }).optional() }).safeParse(response); - this.threadId = threadId; - this.forkedFrom = forkedFrom ?? (parsed.success ? parsed.data.thread?.forkedFromId : null) ?? null; - this.invalidate(); - this.snapshot = null; - if (parsed.success) { - this.acceptSettings({ ...parsed.data, effort: parsed.data.reasoningEffort }); - } else { - this.settings = null; - } - this.publishUsage(); - } - - onSettings(threadId: string, value: unknown): void { - if (threadId !== this.threadId || this.disposed) return; - const parsed = SettingsSchema.safeParse(value); - if (!parsed.success) return; - this.invalidate(); - this.settingsRevision++; - this.acceptSettings(parsed.data); - this.publishUsage(); - } - - private acceptSettings(settings: Settings): void { - this.settings = settings; - this.sync(settings.model === RESERVE_MODEL ? LUNA_MODEL : settings.model, settings.effort ?? null, - settings.serviceTier === 'priority' ? 'fast' : settings.serviceTier === null ? 'standard' : undefined); - } - - invalidate(): void { - this.revision++; - } - - detach(): void { - this.invalidate(); - this.threadId = null; - this.settings = null; - this.snapshot = null; - this.publishUsage(); - } - - dispose(): void { - this.disposed = true; - this.invalidate(); - this.publish(null); - } - - private publishUsage(): void { - const response = this.snapshot; - if (!response || this.disposed) { - this.publish(null); - return; - } - // Never substitute a model-specific bucket for ordinary account windows. - const ordinary = response.rateLimitsByLimitId?.codex - ?? (!response.rateLimits.limitId || response.rateLimits.limitId === 'codex' ? response.rateLimits : null); - const reserve = Object.values(response.rateLimitsByLimitId ?? {}).find(bucket => bucket.limitName === RESERVE_MODEL) - ?? (response.rateLimits.limitName === RESERVE_MODEL ? response.rateLimits : null); - this.publish({ - ordinary: usageBucket(ordinary), - reserve: this.settings?.model === RESERVE_MODEL && response.accountId - ? usageBucket(reserve) : null - }); - } - - /** Serialize polling with pre-submission reconciliation; never submit or replay a message here. */ - refresh(mode: EnhancedMode, canSwitch: () => boolean): Promise { - this.queue = this.queue.then(() => this.readAndReconcile(mode, canSwitch)); - return this.queue; - } - - private returnPath(threadId = this.threadId): string | null { - if (!z.string().uuid().safeParse(threadId).success) return null; - // Shared with the official TUI so local/remote handoff preserves the task's return target. - return join(this.codexHome, 'tui-luna-reserve', `${threadId}.json`); - } - - private async loadReturn(threadId = this.threadId): Promise | null> { - const path = this.returnPath(threadId); - if (!path) return null; - try { - const handle = await open(path, 'r'); - try { - if ((await handle.stat()).size > 4096) return null; - return ReturnSchema.parse(JSON.parse(await handle.readFile('utf8'))); - } finally { - await handle.close(); - } - } catch { return null; } - } - - private async saveReturn(value: z.infer): Promise { - const path = this.returnPath(); - if (!path) throw new Error('Invalid Codex thread ID'); - await mkdir(join(this.codexHome, 'tui-luna-reserve'), { recursive: true }); - const temporary = `${path}.${randomUUID()}.tmp`; - try { - const handle = await open(temporary, 'wx', 0o600); - try { - await handle.writeFile(JSON.stringify(value)); - await handle.sync(); - } finally { await handle.close(); } - await rename(temporary, path); - } finally { await rm(temporary, { force: true }); } - } - - private async readAndReconcile(mode: EnhancedMode, canSwitch: () => boolean): Promise { - if (this.disposed) return; - const revision = this.revision; - let applying = false; - try { - const response = UsageSchema.parse(await this.client.readAccountRateLimits(this.supported)); - if (this.disposed || revision !== this.revision) return; - this.snapshot = response; - this.publishUsage(); - if (!this.supported || !this.threadId || !this.settings || !response.accountId || !canSwitch()) return; - const current = this.settings; - const inReserve = current.model === RESERVE_MODEL; - let previous = inReserve ? await this.loadReturn() : null; - if (inReserve && !previous && this.forkedFrom) { - const inherited = await this.loadReturn(this.forkedFrom); - if (inherited?.account_id === response.accountId) { - await this.saveReturn(inherited); - previous = inherited; - } - } - const ordinary = response.rateLimits; - const recovered = response.ordinaryUsageAllowed != null - && (response.ordinaryUsageAllowed || ordinary.credits?.hasCredits || ordinary.credits?.unlimited) - && response.rateLimitUpsell == null - && ordinary.spendControlReached !== true && ordinary.rateLimitReachedType == null; - const banner = BannerSchema.safeParse(response.rateLimitUpsell); - const entering = !inReserve && (!mode.model || mode.model === current.model) - && response.ordinaryUsageAllowed === false && banner.success - && (!banner.data.blocked_model_slug || banner.data.blocked_model_slug === current.model); - const targetId = entering ? RESERVE_MODEL - : inReserve && recovered && previous?.account_id === response.accountId ? previous.model : null; - const target = this.models.find(model => (model.model ?? model.id) === targetId && (entering || model.hidden !== true)); - if (!target || !targetId || revision !== this.revision || !canSwitch()) return; - const desiredEffort = entering ? current.effort : previous?.effort; - const effort = target.supportedReasoningEfforts?.some(item => item.reasoningEffort === desiredEffort) - ? desiredEffort : target.defaultReasoningEffort; - applying = true; - if (entering) await this.saveReturn({ account_id: response.accountId, model: current.model, effort: current.effort ?? null }); - if (this.disposed || revision !== this.revision || !canSwitch()) return; - const params: ThreadSettingsUpdateParams = { - threadId: this.threadId, - model: targetId, - effort: effort ?? null, - serviceTier: null, - ...(mode.collaborationMode ? { collaborationMode: { - mode: mode.collaborationMode, - settings: { model: targetId, reasoning_effort: effort ?? null, developer_instructions: null } - } } : {}) - }; - const settingsRevision = this.settingsRevision; - await this.client.updateThreadSettings(params); - if (this.disposed || params.threadId !== this.threadId) return; - if (this.settingsRevision !== settingsRevision - && (this.settings?.model !== targetId || (this.settings.effort ?? null) !== (effort ?? null))) return; - this.acceptSettings({ model: targetId, effort, serviceTier: null }); - if (!entering) { - const path = this.returnPath(); - if (path) await rm(path, { force: true }).catch(error => logger.debug('[Codex] Failed to remove Reserve return record', error)); - } - this.publishUsage(); - this.notice(entering - ? 'Ordinary Codex usage is exhausted. Switched to GPT-5.6 Luna using Luna Reserve.' - : `Ordinary Codex usage is available again. Switched back to ${targetId}.`); - } catch (error) { - logger.debug('[Codex] Reserve reconciliation unavailable', error); - if (applying) { - this.supported = false; - this.notice('Codex could not apply the Reserve transition. Task settings are unchanged; no message was retried.'); - } - this.publishUsage(); - } - } - - turnMode(mode: EnhancedMode): EnhancedMode { - if (this.settings?.model !== RESERVE_MODEL) return mode; - const target = this.models.find(model => (model.model ?? model.id) === RESERVE_MODEL); - const effort = target?.supportedReasoningEfforts?.some(item => item.reasoningEffort === mode.modelReasoningEffort) - ? mode.modelReasoningEffort : this.settings.effort ?? undefined; - return { ...mode, model: RESERVE_MODEL, modelReasoningEffort: effort, serviceTier: 'standard' }; - } -} diff --git a/shared/src/schemas.ts b/shared/src/schemas.ts index fb5b2ba8..0fa480ea 100644 --- a/shared/src/schemas.ts +++ b/shared/src/schemas.ts @@ -196,22 +196,7 @@ export const AgentStateCompletedRequestSchema = z.object({ export type AgentStateCompletedRequest = z.infer -const CodexUsageWindowSchema = z.object({ - remainingPercent: z.number().min(0).max(100).nullable(), - windowDurationMins: z.number().positive().nullable(), - resetsAt: z.number().nonnegative().nullable() -}) - -const CodexUsageBucketSchema = z.object({ - primary: CodexUsageWindowSchema.nullable(), - secondary: CodexUsageWindowSchema.nullable() -}) - export const AgentStateSchema = z.object({ - codexUsage: z.object({ - ordinary: CodexUsageBucketSchema, - reserve: CodexUsageBucketSchema.nullable() - }).nullish(), controlledByUser: z.boolean().nullish(), // True while the CLI is delivering a queued message into the active turn // (Steer). Surfaced so the web can reflect the inject in progress. diff --git a/web/src/components/AssistantChat/CodexUsage.test.tsx b/web/src/components/AssistantChat/CodexUsage.test.tsx deleted file mode 100644 index 94bf7e9d..00000000 --- a/web/src/components/AssistantChat/CodexUsage.test.tsx +++ /dev/null @@ -1,28 +0,0 @@ -import { fireEvent, render, screen } from '@testing-library/react' -import { expect, it } from 'vitest' -import { I18nProvider } from '@/lib/i18n-context' -import { CodexUsage } from './CodexUsage' -import type { AgentState } from '@/types/api' - -it('shows Reserve only while active, with separate ordinary limits and unknown data', () => { - const usage: NonNullable = { - ordinary: { primary: null, secondary: { remainingPercent: 40, windowDurationMins: null, resetsAt: null } }, - reserve: null - } - const view = () => - const { rerender } = render(view()) - fireEvent.click(screen.getByRole('button', { name: 'Codex Usage' })) - expect(screen.queryByText(/Luna Reserve/)).toBeNull() - expect(screen.getByText('40% remaining')).toBeTruthy() - usage.reserve = { primary: { remainingPercent: 0, windowDurationMins: 123, resetsAt: 2000000000 }, secondary: { remainingPercent: null, windowDurationMins: null, resetsAt: null } } - rerender(view()) - expect(screen.getByText('Luna Reserve · GPT-5.6 Luna')).toBeTruthy() - expect(screen.getByText('40% remaining')).toBeTruthy() - expect(screen.getByText('0% remaining')).toBeTruthy() - expect(screen.getByText('Unknown')).toBeTruthy() - expect(screen.getByText(/123 min window/)).toBeTruthy() - expect(screen.getByText(/Resets/)).toBeTruthy() - usage.reserve = null - rerender(view()) - expect(screen.queryByText(/Luna Reserve/)).toBeNull() -}) diff --git a/web/src/components/AssistantChat/CodexUsage.tsx b/web/src/components/AssistantChat/CodexUsage.tsx deleted file mode 100644 index ef317113..00000000 --- a/web/src/components/AssistantChat/CodexUsage.tsx +++ /dev/null @@ -1,50 +0,0 @@ -import * as Popover from '@radix-ui/react-popover' -import type { AgentState } from '@/types/api' -import { useTranslation } from '@/lib/use-translation' - -type Usage = NonNullable - -export function CodexUsage({ usage }: { usage: Usage }) { - const { t } = useTranslation() - const bucket = (value: Usage['ordinary']) => { - const windows = [value.primary, value.secondary].filter(window => window !== null) - if (windows.length === 0) return
{t('codexUsage.unknown')}
- return windows.map((window, index) => ( -
- {window.remainingPercent === null - ? t('codexUsage.unknown') - : t('codexUsage.remaining', { percent: window.remainingPercent })} - {window.windowDurationMins !== null ? ( - · {t('codexUsage.window', { minutes: window.windowDurationMins })} - ) : null} - {window.resetsAt !== null ? ( -
- {t('codexUsage.resets', { time: new Date(window.resetsAt * 1000).toLocaleString() })} -
- ) : null} -
- )) - } - return ( - - - - - - -

{t('codexUsage.title')}

-
{t('codexUsage.ordinary')}
- {bucket(usage.ordinary)} - {usage.reserve ? ( -
-

Luna Reserve · GPT-5.6 Luna

- {bucket(usage.reserve)} -
- ) : null} - - - - ) -} diff --git a/web/src/components/AssistantChat/StatusBar.tsx b/web/src/components/AssistantChat/StatusBar.tsx index b2f89ee8..5b1f71cb 100644 --- a/web/src/components/AssistantChat/StatusBar.tsx +++ b/web/src/components/AssistantChat/StatusBar.tsx @@ -18,7 +18,6 @@ import { shouldShowReasoningStatusLabel } from '@/lib/codexStatusLabels' import { isFastServiceTier } from './codexFastMode' -import { CodexUsage } from './CodexUsage' import { useTranslation } from '@/lib/use-translation' import { useSessionHeaderMetadata } from '@/hooks/useSessionHeaderMetadata' @@ -303,9 +302,6 @@ export function StatusBar(props: { {connectionStatus.text} - {props.agentFlavor === 'codex' && props.agentState?.codexUsage ? ( - - ) : null} {contextUsageLabel ? ( diff --git a/web/src/lib/locales/en.ts b/web/src/lib/locales/en.ts index f4c6f072..ac237550 100644 --- a/web/src/lib/locales/en.ts +++ b/web/src/lib/locales/en.ts @@ -1,11 +1,4 @@ export default { - 'codexUsage.title': 'Codex Usage', - 'codexUsage.ordinary': 'Ordinary usage', - 'codexUsage.unknown': 'Unknown', - 'codexUsage.remaining': '{percent}% remaining', - 'codexUsage.window': '{minutes} min window', - 'codexUsage.resets': 'Resets {time}', - // Loading states 'loading': 'Loading…', 'authorizing': 'Authorizing…', diff --git a/web/src/lib/locales/zh-CN.ts b/web/src/lib/locales/zh-CN.ts index 9f63d9cc..63994476 100644 --- a/web/src/lib/locales/zh-CN.ts +++ b/web/src/lib/locales/zh-CN.ts @@ -1,11 +1,4 @@ export default { - 'codexUsage.title': 'Codex 用量', - 'codexUsage.ordinary': '常规用量', - 'codexUsage.unknown': '未知', - 'codexUsage.remaining': '剩余 {percent}%', - 'codexUsage.window': '{minutes} 分钟窗口', - 'codexUsage.resets': '重置时间 {time}', - // Loading states 'loading': '加载中…', 'authorizing': '认证中…',