elanous 0.2.35 → 0.2.36

package/src/ux-sim/fault-tui-400-autoheal.ts

@@ -0,0 +1,297 @@
+// 결함 주입 심 «TUI 턴 400 자동 회복» ── classic TUI 턴이 모델 API 400(대화 기록 어긋남)을 받으면
+// ① 새 백엔드 세션 ⊕ 직전 요약으로 1회 재시도 → ② 또 400 이면 llm.fallbackChain 다음 공급자로 그 턴 재시도
+// → 둘 다 실패면 턴당 2회 상한에서 사람에게 이유 한 줄 — 를 프로세스 안(S1)에서 재현한다.
+//
+// 🚨 왜 이것이 있나 (TUI-400-AUTOHEAL-FALLBACK · 대표 10-22)
+// 10-22 07:1x OPT TUI 턴이 «Codex API 400(대화 기록 어긋남 의심)» 로 끝나고 멈췄다. 지문 `codex-400-bad-request` 는
+// «/session new 뒤 다시» 를 사람에게 보여줄 뿐이었고 OP 가 손으로 쳐서 회복했다. 이 심이 그 멈춤(RED)과 회복(GREEN)을 가른다.
+//
+// ⛔ 흉내 내지 않는다 — 생산 턴 경로를 그대로 부른다:
+//   `runDashboardChatMainPlainTurn`(classic TUI plain 턴) → `bootDashboardAcpSession`(in-process ACP · 대시보드와 같은 부트)
+//   → `bridgeCoreTurnToAcp` → `runCoreTurn` → `streamLLMWithTools` → 생산 `PROVIDERS['openai-codex']`(Responses API HTTP)
+//   ⊕ 폴백 대상 생산 `PROVIDERS.glm`(zai 코딩 플랜 · 자격 env 하나만 가짜). 자동 회복 배선도 대시보드와 같은
+//   `createDashboardTurnAutoheal` 이고, 요약은 생산 `compactConversation`(같은 codex 경로로 요약 요청)이다.
+// 주입은 HTTP 경계 하나뿐: codex 쪽은 스크립트된 400(실제 «No tool call found for function call output» 꼴) 또는
+// 정상 SSE · zai 코딩 엔드포인트는 정상 SSE 또는 400. 화면 쪽은 줄 배열(렌더 없음 — 렌더는 headless-tui 심의 몫).
+// 관측 = 화면 줄 · self-health.autoheal 이벤트 · 공급자별 HTTP 호출 수 · 영수증(명령 ⊕ injectionId ⊕ 결과) · 층 L1.
+
+import { randomUUID } from 'node:crypto';
+import { mkdtempSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { _resetKeyCacheForTests } from '../config.js';
+import { debug } from '../debug/log.js';
+import { PROVIDERS } from '../llm.js';
+import { ZAI_CODING_PLAN_URL } from '../llm/zai-coding-plan.js';
+import { ZAI_POD_SECRET_KEY } from '../task-orchestrator/surfaces/pod-zai-child.js';
+import { setUserConfigOverlay } from '../user-config.js';
+import { getCompactSystemPrompt } from '../compact/index.js';
+import { bootDashboardAcpSession } from '../dashboard/acp-boot.js';
+import {
+  runDashboardChatMainPlainTurn,
+  type DashboardChatMainPlainTurnRuntimeDeps,
+} from '../dashboard/input/chat-main-plain-turn-runtime.js';
+import { createDashboardTurnAutoheal } from '../dashboard/input/turn-autoheal.js';
+import type { ChatMessage } from '../chat/index.js';
+
+export const TUI_400_AUTOHEAL_LEVEL = 'L1' as const;
+export const TUI_400_AUTOHEAL_EVIDENCE_COMMAND = 'bun test src/ux-sim/fault-tui-400-autoheal.test.ts';
+
+/** 실제 codex 400 «대화 기록 어긋남» 본문 꼴(seat-autowake 가 잡는 `No tool call found for function call output with call_id`). */
+export const TUI_400_HISTORY_MISMATCH_BODY = JSON.stringify({
+  error: {
+    message: 'No tool call found for function call output with call_id call_autoheal_sim.',
+    type: 'invalid_request_error', param: 'input', code: null,
+  },
+});
+export const TUI_400_CODEX_ANSWER = '[tui-400-autoheal] codex OK';
+export const TUI_400_ZAI_ANSWER = '[tui-400-autoheal] zai OK';
+export const TUI_400_SUMMARY = '[tui-400-autoheal] 요약: 앞 질문은 src/a.ts 읽기';
+
+/** 시나리오 — codex «HTTP 호출»별 응답 스크립트(요약 요청은 따로 정상 응답) ⊕ zai 응답.
+ *  ⚠️ 생산 재시도 정책(`fetchApiWithRetry` → `decideRetry`)이 400 을 같은 공급자로 한 번 더 보낸다 —
+ *  그래서 «턴 한 번 실패» = codex 400 두 번이다(llm.retry 관측으로 센다). 스크립트는 그 실측 꼴을 따른다. */
+export type Tui400Scenario = 'heal-new-session' | 'heal-provider' | 'exhausted';
+
+const SCRIPTS: Readonly<Record<Tui400Scenario, { codex: ReadonlyArray<'400' | 'ok'>; zai: '400' | 'ok' }>> = {
+  'heal-new-session': { codex: ['400', '400', 'ok'], zai: 'ok' },
+  'heal-provider': { codex: ['400', '400', '400', '400'], zai: 'ok' },
+  exhausted: { codex: ['400', '400', '400', '400'], zai: '400' },
+};
+
+const ISOLATED_LLM_ENV_KEYS = [
+  'OPENAI_API_KEY', 'XAI_API_KEY', 'GROK_API_KEY', 'GROK_CODE_XAI_API_KEY', 'ANTHROPIC_API_KEY',
+  'GEMINI_API_KEY', 'GOOGLE_API_KEY', 'LOCAL_LLM_URL', 'OPENROUTER_API_KEY', 'OPENGATEWAY_API_KEY', 'ZAI_API_KEY',
+  'ELANOUS_TEST_FAKE_LLM', 'ELANOUS_TEST_FAKE_LLM_AS_PROVIDER',
+] as const;
+
+export interface Tui400Observation {
+  readonly level: typeof TUI_400_AUTOHEAL_LEVEL;
+  readonly injectionId: string;
+  readonly scenario: Tui400Scenario;
+  /** 턴 마감 상태(finalizeTurn). */
+  readonly status: string;
+  /** 'recovered-new-session' | 'recovered-provider' | 'stopped' */
+  readonly result: 'recovered-new-session' | 'recovered-provider' | 'stopped';
+  /** 사람이 본 줄(빈 줄 제외). */
+  readonly screen: readonly string[];
+  /** 화면에 흐른 답(스트림 텍스트). */
+  readonly answer: string;
+  readonly codexTurnCalls: number;
+  readonly codexSummaryCalls: number;
+  readonly zaiCalls: number;
+  /** llm.retry «retry» 줄 수 — 생산 재시도 정책이 같은 공급자로 다시 보낸 횟수. */
+  readonly sameProviderRetries: number;
+  /** 새 백엔드 세션으로 바뀌었나(ACP 세션 id 비교). */
+  readonly sessionRenewed: boolean;
+  /** 재시도 직전 TUI 기록의 역할 줄 — [system?, 요약 assistant, 현재 user]. */
+  readonly historyRoles: readonly string[];
+  readonly autoheal: ReadonlyArray<{ event: string; data: Record<string, unknown> }>;
+}
+
+export function tui400Receipt(o: Pick<Tui400Observation, 'injectionId' | 'result'>): string {
+  return `영수증: ${TUI_400_AUTOHEAL_EVIDENCE_COMMAND} ⊕ ${o.injectionId} ⊕ ${o.result}`;
+}
+
+export function renderTui400(o: Tui400Observation): string[] {
+  return [
+    `tui-400-autoheal · 층 ${o.level} · 주입 ${o.injectionId} · 시나리오 ${o.scenario} · 결과 ${o.result} · 턴 ${o.status}`,
+    `  HTTP: codex 턴 ${o.codexTurnCalls} · codex 요약 ${o.codexSummaryCalls} · zai ${o.zaiCalls} · 같은 공급자 재시도 ${o.sameProviderRetries} · 새 세션 ${o.sessionRenewed ? '예' : '아니오'}`,
+    ...o.screen.map((line) => `  화면│ ${line}`),
+    `  답: ${o.answer || '(없음)'}`,
+    ...o.autoheal.map((row) => `  self-health.autoheal ${row.event} ${JSON.stringify(row.data)}`),
+    tui400Receipt(o),
+  ];
+}
+
+/** 정상 SSE 한 토막 — 엔드포인트 모양을 따른다(Chat Completions `choices[].delta` · Responses `output_text.delta`). */
+function okStream(url: string, text: string): Response {
+  const events: unknown[] = url.includes('/chat/completions') || url === ZAI_CODING_PLAN_URL
+    ? [{ choices: [{ delta: { content: text } }] }, { choices: [], usage: { prompt_tokens: 3, completion_tokens: 1 } }]
+    : [{ type: 'response.output_text.delta', delta: text }, { type: 'response.completed', response: { id: 'ok' } }];
+  const body = events.map((e) => `data: ${JSON.stringify(e)}\n\n`).join('') + 'data: [DONE]\n\n';
+  return new Response(body, { status: 200, headers: { 'content-type': 'text/event-stream' } });
+}
+
+/** 결함 주입 본체 — 생산 classic TUI plain 턴을 그대로 한 번 돈다. */
+export async function simTui400Autoheal(
+  scenario: Tui400Scenario,
+  opts: {
+    injectionId?: string;
+    /** false = 자동 회복 배선 없이(수리 전 동작 · 10-22 실측 멈춤 재현). 기본 true. */
+    autoheal?: boolean;
+  } = {},
+): Promise<Tui400Observation> {
+  const injectionId = opts.injectionId ?? `inj-${randomUUID()}`;
+  const script = SCRIPTS[scenario];
+  const savedEnv = { ...process.env };
+  const savedGrok = PROVIDERS.grok;
+  const savedFetch = globalThis.fetch;
+  const savedLog = debug.log;
+  const home = mkdtempSync(join(tmpdir(), 'fault-tui-400-'));
+  const rows: Array<{ category: string; event: string; data: Record<string, unknown> }> = [];
+  let codexTurnCalls = 0;
+  let codexSummaryCalls = 0;
+  let zaiCalls = 0;
+  let session: Awaited<ReturnType<typeof bootDashboardAcpSession>>['session'] | null = null;
+  try {
+    process.env.HOME = home;
+    process.env.XDG_CONFIG_HOME = join(home, '.config');
+    process.env.CODEX_HOME = join(home, '.codex');
+    process.env.ELANOUS_STATE_DIR = join(home, 'state');
+    for (const key of ISOLATED_LLM_ENV_KEYS) delete process.env[key];
+    process.env.ELANOUS_KEEP_ENV_KEYS = '1';
+    _resetKeyCacheForTests();
+    // ② 의 «다음 공급자» = 생산 glm(zai 코딩 플랜). 자격 env 하나만 가짜 값 — 공급자 코드는 그대로.
+    process.env[ZAI_POD_SECRET_KEY] = 'fake-zai-test-only';
+    // grok 자격은 os.homedir() 의 auth.json 이라 HOME 격리로 못 지운다 — 체인 밖으로 새지 않게 이 한 칸만 막는다.
+    PROVIDERS.grok = { ...savedGrok!, available: () => false };
+    setUserConfigOverlay((config) => ({
+      ...config,
+      llm: {
+        ...config.llm, provider: 'openai-codex', apiKey: 'fake-codex-test-only',
+        fallbackChain: ['codex-rotate', 'zai'], goalLoop: { ...(config.llm.goalLoop ?? {}), enabled: false },
+      },
+    }));
+    globalThis.fetch = Object.assign(
+      async (input: string | URL | Request, init?: RequestInit): Promise<Response> => {
+        const url = typeof input === 'string' ? input : input instanceof URL ? input.href : input.url;
+        const body = typeof init?.body === 'string' ? init.body : '';
+        if (url === ZAI_CODING_PLAN_URL) {
+          zaiCalls += 1;
+          if (script.zai === '400') return new Response(TUI_400_HISTORY_MISMATCH_BODY, { status: 400 });
+          return okStream(url, TUI_400_ZAI_ANSWER);
+        }
+        // 요약(생산 compactConversation)도 같은 codex 경로로 간다 — 요약 요청은 정상 응답(스크립트 밖).
+        // 판별은 생산 요약 프롬프트(`getCompactSystemPrompt`) 그대로 — 문구가 바뀌어도 같이 따라간다.
+        if (body.includes(JSON.stringify(getCompactSystemPrompt()).slice(1, 61))) {
+          codexSummaryCalls += 1;
+          return okStream(url, TUI_400_SUMMARY);
+        }
+        const step = script.codex[codexTurnCalls] ?? '400';
+        codexTurnCalls += 1;
+        if (step === '400') return new Response(TUI_400_HISTORY_MISMATCH_BODY, { status: 400 });
+        return okStream(url, TUI_400_CODEX_ANSWER);
+      },
+      { preconnect: (savedFetch as unknown as { preconnect?: unknown }).preconnect },
+    ) as typeof fetch;
+    debug.log = ((category: string, event: string, data: Record<string, unknown>): void => {
+      rows.push({ category, event, data: data ?? {} });
+    }) as typeof debug.log;
+
+    // 대화 기록 — 짝 없는 도구 호출이 남은 «어긋난» 기록(실측 400 의 모양).
+    const history: ChatMessage[] = [
+      { role: 'system', content: 'tui-400-autoheal system' },
+      { role: 'user', content: '앞 질문: src/a.ts 읽어줘' },
+      { role: 'assistant', content: [{ type: 'tool_use', id: 'call_autoheal_sim', name: 'Read', input: { path: 'src/a.ts' } }] },
+    ] as unknown as ChatMessage[];
+    const cwd = home;
+    const boot = await bootDashboardAcpSession({
+      getCwd: () => cwd,
+      getChatHistory: () => history,
+      getPreamble: () => [],
+      // 실제 TUI 턴처럼 도구가 실린 턴(streamLLMWithTools 도구 경로) — 모델이 안 부르면 실행은 0.
+      getTools: () => [{ name: 'tui_400_probe', description: 'tui-400-autoheal probe', parameters: { type: 'object' } }],
+      getActiveModel: () => null,
+      dispatchTool: async () => ({ error: 'no tools in tui-400-autoheal' }),
+      pushTurnToolHistory: (msgs) => { for (const m of msgs) history.push(m as ChatMessage); },
+    });
+    session = boot.session;
+    session.setOriginSessionIdReader(() => null);
+    const firstSessionId = session.currentSessionId;
+    const screen: string[] = [];
+    let answer = '';
+    let status = 'unknown';
+    let historyRoles: string[] = [];
+    const userText = `tui-400-autoheal ${injectionId}: 다시 해줘`;
+    const userMsg = { role: 'user', content: userText };
+    const liveSession = session;
+    const deps = {
+      debugLog: (category: string, event: string, payload: Record<string, unknown>) => { debug.log(category, event, payload); },
+      userText, contextText: '', turnStartedAt: Date.now(),
+      thinking: { update() {}, updateAnimated() {}, updateMetrics() {}, reflow() {}, stop() {} },
+      benchmarkMode: false, attachedSessionId: null,
+      chat: { history }, chatLines: screen, contextRegistry: {}, sessionRegistry: {}, sync: {},
+      chatFooterLine: { current: null }, acpTurnRef: {},
+      blockAttach: (line: string) => line,
+      pushChatLine: (line: string) => { screen.push(line); },
+      pushDebugLine: () => {}, setChatScrollBottom: () => {}, draw: () => {}, pinChatTail: () => {},
+      termCols: () => 140, getSessionCwd: () => cwd,
+      getUserConfig: () => ({ voice: { tts: { drainCooldownMs: 0 } } }),
+      inspectActiveProvider: () => ({ model: 'gpt-6-sol' }), getActivePluginName: () => undefined,
+      buildTurnPromptRuntime: () => ({ promptBankContext: '', turnProfile: {} }),
+      loadAttachments: async () => {},
+      buildTurnMessage: () => ({ userMsg }),
+      beginCodeEditTurn: async () => {},
+      buildTurnPreamble: () => [],
+      runAutoCompact: async () => {},
+      readToolDispatchCount: () => 0,
+      attachChatStreamingKeys: () => () => {},
+      createOptionalSpecs: async () => [],
+      createTurnStreamRuntime: () => ({
+        onText(chunk: string) { answer += chunk; }, onToolCall() {}, onToolResult() {}, getAssistantStart: () => 0,
+      }),
+      runTurnUsageRuntime: () => {},
+      runTurnPrelude: () => ({ searchPlannerState: {} }),
+      armAcpTurnRef: () => {}, resetAcpTurnRef: () => {},
+      acpSession: {
+        send: (o: Parameters<typeof liveSession.send>[0]) => {
+          historyRoles = history.map((m) => `${m.role}${typeof m.content === 'string' && m.content.startsWith('Compacted conversation summary') ? ':summary' : ''}`);
+          return liveSession.send(o);
+        },
+      },
+      ...(opts.autoheal === false ? {} : {
+        autoheal: createDashboardTurnAutoheal({
+          session: liveSession, getCwd: () => cwd, remote: false, activeModel: () => undefined, summaryTimeoutMs: 10_000,
+        }),
+      }),
+      autoTts: { pushChunk() {}, commit() {}, cancel() {} },
+      voiceChat: { getPhase: () => 'idle', transitionToSpeaking() {}, notifyResponseDone() {} },
+      finalizeStreamLifecycle: (_c: () => void, _a: boolean, s: 'completed' | 'interrupted' | 'failed') => s,
+      recordTurnMetrics: async () => {},
+      commitAssistantRenderState: () => ({ lastAssistantRaw: null, lastAssistantRange: null, lastAssistantMode: 'rendered' }),
+      applyAssistantRenderState: () => {},
+      runTailAutoCopy: async () => {}, runCodeEditPostTurn: async () => {}, runHandoffMirror: async () => {},
+      parseActionBlock: () => null, isBrowseMode: () => false, enterSyncMode: () => {},
+      applyActionBlock: () => ({}), runActionEffects: async () => {},
+      warningLine: (message: string) => message,
+      pushErrorLine: (message: string) => { screen.push(message); },
+      pushFailureNotice: (notice: string) => { screen.push(notice); },
+      finalizeTurn: (s: string) => { status = s; },
+    } as unknown as DashboardChatMainPlainTurnRuntimeDeps;
+    await runDashboardChatMainPlainTurn(deps);
+    const autoheal = rows.filter((row) => row.category === 'self-health.autoheal')
+      .map(({ event, data }) => ({ event, data }));
+    const recovered = autoheal.find((row) => row.event === 'recovered');
+    const result = status !== 'completed' || !recovered ? 'stopped'
+      : recovered.data.step === 'provider-fallback' ? 'recovered-provider' : 'recovered-new-session';
+    return {
+      level: TUI_400_AUTOHEAL_LEVEL,
+      injectionId,
+      scenario,
+      status,
+      result,
+      screen: screen.filter((line) => line !== ''),
+      answer,
+      codexTurnCalls,
+      codexSummaryCalls,
+      zaiCalls,
+      sameProviderRetries: rows.filter((row) => row.category === 'llm.retry' && row.event === 'retry').length,
+      sessionRenewed: liveSession.currentSessionId !== firstSessionId,
+      historyRoles,
+      autoheal,
+    };
+  } finally {
+    try { await session?.close(); } catch { /* best-effort teardown */ }
+    debug.log = savedLog;
+    globalThis.fetch = savedFetch;
+    PROVIDERS.grok = savedGrok!;
+    setUserConfigOverlay(null);
+    for (const key of Object.keys(process.env)) {
+      if (!(key in savedEnv)) delete process.env[key];
+    }
+    Object.assign(process.env, savedEnv);
+    _resetKeyCacheForTests();
+    rmSync(home, { recursive: true, force: true });
+  }
+}