@1agh/maude 1.0.11 → 1.3.0

package/apps/studio/test/acp-bridge.test.ts

@@ -1,545 +0,0 @@
-// End-to-end bridge test against a mock ACP agent (test/fixtures/mock-acp-agent.mjs)
-// so no real `claude` install is needed. Proves: spawn + handshake + streamed
-// update + turn completion, AND that ANTHROPIC_API_KEY never reaches the child
-// (DDR-123 guardrail #1, verified end-to-end — the mock echoes its own env).
-
-import { afterEach, describe, expect, test } from 'bun:test';
-import { mkdtemp, readFile, rm } from 'node:fs/promises';
-import { tmpdir } from 'node:os';
-import { join } from 'node:path';
-
-import { AcpBridge } from '../acp/bridge.ts';
-import {
-  getClaudeAuthStatus,
-  probeAcpAvailability,
-  probeAcpAvailabilityAuthed,
-} from '../acp/probe.ts';
-
-const FIXTURE = join(import.meta.dir, 'fixtures', 'mock-acp-agent.mjs');
-const TEST_ENV_KEYS = [
-  'ANTHROPIC_API_KEY',
-  'MAUDE_ACP_ADAPTER_ENTRY',
-  'MAUDE_ACP_RUNTIME',
-  'MAUDE_CLAUDE_BIN',
-];
-
-afterEach(() => {
-  for (const key of TEST_ENV_KEYS) delete process.env[key];
-});
-
-describe('AcpBridge — round-trip + subscription guardrail', () => {
-  test('connects, streams an update, completes a turn, and scrubs ANTHROPIC_API_KEY', async () => {
-    process.env.ANTHROPIC_API_KEY = 'sk-must-be-scrubbed';
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath; // run the mock under this bun
-    process.env.MAUDE_CLAUDE_BIN = process.execPath; // satisfy the claude-present check
-
-    const updates: unknown[] = [];
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: (u) => updates.push(u) });
-    try {
-      const result = await bridge.prompt('hello', 'c1');
-      expect(bridge.connected).toBe(true);
-      expect(bridge.sessionId).toBe('mock-session-1');
-      expect(result.stopReason).toBe('end_turn');
-
-      // The mock streamed its own ANTHROPIC_API_KEY — it must read `<unset>`,
-      // proving scrubAgentEnv stripped it from the child env before spawn.
-      const streamed = JSON.stringify(updates);
-      expect(streamed).toContain('apiKey=<unset>');
-      expect(streamed).not.toContain('sk-must-be-scrubbed');
-      // DDR-123 guardrail #2 — the bridge pins the adapter to the user's own
-      // `claude` (here MAUDE_CLAUDE_BIN = this bun) via CLAUDE_CODE_EXECUTABLE, so
-      // it never falls back to the unshipped ~210 MB native binary.
-      expect(streamed).toContain(`claudeExe=${process.execPath}`);
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('model/effort are no longer env-at-spawn — a config change never respawns the running adapter', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-
-    const updates: unknown[] = [];
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: (u) => updates.push(u) });
-    try {
-      // A persisted pick this mock doesn't advertise (it has no configOptions at
-      // all) is skipped best-effort, never forwarded to ANTHROPIC_MODEL/env.
-      bridge.setConfig('opus', 'thorough');
-      await bridge.prompt('hi', 'c1');
-      const sid = bridge.sessionId;
-      const first = JSON.stringify(updates);
-      expect(first).toContain('model=<unset>');
-      expect(first).toContain('thinking=<unset>');
-
-      // Changing the config again must NOT tear down / respawn the live process
-      // (Task A3 — the old configChanged()-triggered stop() is gone).
-      updates.length = 0;
-      bridge.setConfig(null, 'fast');
-      await bridge.prompt('again', 'c1');
-      expect(bridge.sessionId).toBe(sid); // same session — no respawn happened
-      const second = JSON.stringify(updates);
-      expect(second).toContain('model=<unset>');
-      expect(second).toContain('thinking=<unset>');
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('a second prompt reuses the same live session', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: () => {} });
-    try {
-      await bridge.prompt('first', 'c1');
-      const sid = bridge.sessionId;
-      await bridge.prompt('second', 'c1');
-      expect(bridge.sessionId).toBe(sid);
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('different chat ids get separate claude sessions; same id reuses its own', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: () => {} });
-    try {
-      await bridge.prompt('a', 'chatA');
-      const sa = bridge.sessionId;
-      await bridge.prompt('b', 'chatB');
-      const sb = bridge.sessionId;
-      expect(sa).not.toBe(sb); // separate per-chat contexts
-      await bridge.prompt('a again', 'chatA');
-      expect(bridge.sessionId).toBe(sa); // chatA reuses its own session
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-});
-
-describe('AcpBridge — cross-restart session resume (DDR-125 gap)', () => {
-  let dir: string;
-
-  afterEach(async () => {
-    if (dir) await rm(dir, { recursive: true, force: true });
-  });
-
-  test('a persisted sessionId resumes via loadSession instead of spawning a new session', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-    dir = await mkdtemp(join(tmpdir(), 'acp-bridge-test-'));
-    const storePath = join(dir, 'c1.session.json');
-    await Bun.write(storePath, JSON.stringify({ sessionId: 'persisted-session-abc' }));
-
-    const updates: unknown[] = [];
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: (u) => updates.push(u) });
-    try {
-      bridge.setSessionStorePath(storePath);
-      await bridge.prompt('pokracuj', 'c1');
-      // Resumed the persisted id verbatim — never fell through to session/new
-      // (which would have minted a fresh `mock-session-N`).
-      expect(bridge.sessionId).toBe('persisted-session-abc');
-      // The mock's replay notification must NOT reach the UI sink.
-      expect(JSON.stringify(updates)).not.toContain('REPLAYED-HISTORY-SHOULD-NOT-SURFACE');
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('a resume that fails (pruned session) falls back to newSession and re-persists the new id', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-    dir = await mkdtemp(join(tmpdir(), 'acp-bridge-test-'));
-    const storePath = join(dir, 'c1.session.json');
-    await Bun.write(storePath, JSON.stringify({ sessionId: 'not-found' }));
-
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: () => {} });
-    try {
-      bridge.setSessionStorePath(storePath);
-      await bridge.prompt('pokracuj', 'c1');
-      expect(bridge.sessionId).toBe('mock-session-1'); // fresh session, not the stale id
-      const stored = JSON.parse(await readFile(storePath, 'utf8'));
-      expect(stored.sessionId).toBe('mock-session-1'); // re-persisted so the NEXT restart resumes it
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('no prior chat for this sidecar behaves exactly as before — a brand new session is created and persisted', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-    dir = await mkdtemp(join(tmpdir(), 'acp-bridge-test-'));
-    const storePath = join(dir, 'brand-new.session.json');
-
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: () => {} });
-    try {
-      bridge.setSessionStorePath(storePath);
-      await bridge.prompt('hi', 'c1');
-      expect(bridge.sessionId).toBe('mock-session-1');
-      const stored = JSON.parse(await readFile(storePath, 'utf8'));
-      expect(stored.sessionId).toBe('mock-session-1');
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('no sidecar wired at all (e.g. a warm-up before any prompt) never touches disk and behaves exactly as before', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: () => {} });
-    try {
-      // setSessionStorePath is never called — sessionStorePath stays null.
-      await bridge.prompt('hi', 'c1');
-      expect(bridge.sessionId).toBe('mock-session-1');
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('a malformed persisted sessionId (not a plausible shape) is rejected — falls back to newSession', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-    dir = await mkdtemp(join(tmpdir(), 'acp-bridge-test-'));
-    const storePath = join(dir, 'c1.session.json');
-    // Not a plausible sessionId shape (embedded newline + non-UUID charset) —
-    // must never reach the wire as a `loadSession` sessionId.
-    await Bun.write(storePath, JSON.stringify({ sessionId: 'evil\nsessionId; rm -rf /' }));
-
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: () => {} });
-    try {
-      bridge.setSessionStorePath(storePath);
-      await bridge.prompt('hi', 'c1');
-      expect(bridge.sessionId).toBe('mock-session-1'); // rejected — fresh session instead
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-
-  test('warm and prompt racing for the same chat share one resume attempt (no replaying race)', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_ACP_RUNTIME = process.execPath;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-    dir = await mkdtemp(join(tmpdir(), 'acp-bridge-test-'));
-    const storePath = join(dir, 'c1.session.json');
-    await Bun.write(storePath, JSON.stringify({ sessionId: 'persisted-race-session' }));
-
-    const updates: unknown[] = [];
-    const bridge = new AcpBridge({ repoRoot: process.cwd(), onUpdate: (u) => updates.push(u) });
-    try {
-      bridge.setSessionStorePath(storePath);
-      // Fire warmUp and prompt concurrently for the same chatId — both call
-      // sessionFor('c1') before either has resolved.
-      const [, result] = await Promise.all([bridge.warmUp('c1'), bridge.prompt('pokracuj', 'c1')]);
-      expect(bridge.sessionId).toBe('persisted-race-session');
-      expect(result.stopReason).toBe('end_turn');
-      // Only ONE resume attempt happened — the mock's replay text never leaked,
-      // and the prompt's own turn still streamed normally.
-      expect(JSON.stringify(updates)).not.toContain('REPLAYED-HISTORY-SHOULD-NOT-SURFACE');
-    } finally {
-      await bridge.stop();
-    }
-  }, 15000);
-});
-
-describe('probeAcpAvailability — not-connected detection', () => {
-  test('reports not-available with a Claude-Code reason when the CLI is absent', () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = join(import.meta.dir, 'no-such-claude-bin');
-    const probe = probeAcpAvailability();
-    expect(probe.available).toBe(false);
-    expect(probe.reason ?? '').toMatch(/Claude Code/i);
-  });
-
-  test('reports available when both the adapter and a claude binary resolve', () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = process.execPath;
-    const probe = probeAcpAvailability();
-    expect(probe.available).toBe(true);
-    expect(probe.adapterEntry).toBe(FIXTURE);
-  });
-});
-
-describe('probeAcpAvailabilityAuthed — DDR-166 T0d, logged-out must not read as available', () => {
-  const AUTH_FIXTURE = join(import.meta.dir, 'fixtures', 'fake-claude-auth.mjs');
-
-  afterEach(() => {
-    delete process.env.FAKE_CLAUDE_LOGGED_IN;
-    delete process.env.FAKE_CLAUDE_STDOUT_NOISE;
-  });
-
-  test('installed but NOT signed in reports available:false (the bug this closes)', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = AUTH_FIXTURE;
-    process.env.FAKE_CLAUDE_LOGGED_IN = '0';
-    const probe = await probeAcpAvailabilityAuthed();
-    expect(probe.available).toBe(false);
-    expect(probe.reason ?? '').toMatch(/signed in/i);
-  });
-
-  test('installed AND signed in reports available:true', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = AUTH_FIXTURE;
-    process.env.FAKE_CLAUDE_LOGGED_IN = '1';
-    const probe = await probeAcpAvailabilityAuthed();
-    expect(probe.available).toBe(true);
-  });
-
-  test('not installed still reports the plain not-installed reason (no auth spawn attempted)', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = join(import.meta.dir, 'no-such-claude-bin');
-    const probe = await probeAcpAvailabilityAuthed();
-    expect(probe.available).toBe(false);
-    expect(probe.reason ?? '').toMatch(/Claude Code/i);
-  });
-
-  // Issue #107 — the reported `claude` was a mise shim printing a banner to
-  // STDOUT before the JSON. Parsing the whole blob threw, the catch returned
-  // null, and a signed-in user was told they were signed out.
-  test('signed in through a stdout-noisy shim still reports available:true (#107)', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = AUTH_FIXTURE;
-    process.env.FAKE_CLAUDE_LOGGED_IN = '1';
-    process.env.FAKE_CLAUDE_STDOUT_NOISE = '1';
-    const probe = await probeAcpAvailabilityAuthed();
-    expect(probe.available).toBe(true);
-  });
-
-  test('an UNREADABLE status says so — it does not accuse the user of being signed out (#107)', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = join(import.meta.dir, 'fixtures', 'noisy-claude-auth.mjs');
-    process.env.FAKE_CLAUDE_OUT = 'mise: something went sideways\nno json here\n';
-    try {
-      const probe = await probeAcpAvailabilityAuthed();
-      expect(probe.available).toBe(false);
-      expect(probe.reason ?? '').toContain("couldn't read");
-      expect(probe.reason ?? '').not.toMatch(/not signed in/i);
-    } finally {
-      delete process.env.FAKE_CLAUDE_OUT;
-    }
-  });
-
-  test('signed OUT through a stdout-noisy shim is still reported as signed out (no false positive)', async () => {
-    process.env.MAUDE_ACP_ADAPTER_ENTRY = FIXTURE;
-    process.env.MAUDE_CLAUDE_BIN = AUTH_FIXTURE;
-    process.env.FAKE_CLAUDE_LOGGED_IN = '0';
-    process.env.FAKE_CLAUDE_STDOUT_NOISE = '1';
-    const probe = await probeAcpAvailabilityAuthed();
-    expect(probe.available).toBe(false);
-    expect(probe.reason ?? '').toMatch(/signed in/i);
-  });
-});
-
-// Issue #107 — direct coverage of the stdout shape tolerated by the auth probe.
-// A wrapper on PATH (mise/asdf/volta, corporate `bin` shims) can prepend noise
-// to stdout; the fast path (clean JSON) and the honest failure (nothing
-// parseable → null) both have to keep working alongside it.
-describe('getClaudeAuthStatus — wrapper-stdout tolerance', () => {
-  const NOISY = join(import.meta.dir, 'fixtures', 'noisy-claude-auth.mjs');
-
-  afterEach(() => {
-    delete process.env.FAKE_CLAUDE_OUT;
-    delete process.env.FAKE_CLAUDE_HEAD;
-    delete process.env.FAKE_CLAUDE_PAD_BYTES;
-    delete process.env.FAKE_CLAUDE_TAIL;
-  });
-
-  const statusFor = async (stdout: string) => {
-    process.env.MAUDE_CLAUDE_BIN = NOISY;
-    process.env.FAKE_CLAUDE_OUT = stdout;
-    return getClaudeAuthStatus();
-  };
-
-  const PAYLOAD = '{"loggedIn":true,"apiProvider":"firstParty","subscriptionType":"max"}';
-
-  test('clean JSON — the unchanged fast path', async () => {
-    expect(await statusFor(PAYLOAD)).toEqual({
-      loggedIn: true,
-      apiProvider: 'firstParty',
-      subscriptionType: 'max',
-    });
-  });
-
-  test('a banner line ahead of the JSON (the reported mise shim)', async () => {
-    const out = `mise ~/.config/mise/config.toml tools: claude@2.1.246\n${PAYLOAD}`;
-    expect((await statusFor(out))?.loggedIn).toBe(true);
-  });
-
-  test('noise carrying braces of its own does not outrank the real payload', async () => {
-    const out = `warning: {"loggedIn":false} is not a real status\nresolving {claude}\n${PAYLOAD}`;
-    expect((await statusFor(out))?.subscriptionType).toBe('max');
-  });
-
-  test('trailing noise after the JSON', async () => {
-    expect((await statusFor(`${PAYLOAD}\nmise: done in 12ms\n`))?.loggedIn).toBe(true);
-  });
-
-  test('a brace inside a string value never miscounts the object bounds', async () => {
-    const out = `noise\n{"orgName":"a } brace \\" in a string","loggedIn":true}\n`;
-    expect((await statusFor(out))?.loggedIn).toBe(true);
-  });
-
-  test('nothing parseable still returns null — an unreadable status is not a signed-in one', async () => {
-    expect(await statusFor('mise: command failed\nno json here at all\n')).toBeNull();
-  });
-
-  test('a signed-out payload behind noise stays signed out', async () => {
-    expect((await statusFor(`mise banner\n{"loggedIn":false}`))?.loggedIn).toBe(false);
-  });
-
-  // ── Security-review regressions (defender M1, attacker F1/F2/F5) ──
-  // Every one of these is a way the FIRST cut of this parser failed SOFT and
-  // re-created #107 — a confident false claim — through a new door.
-
-  test('an object with no `loggedIn` is not a status document → null, never a false "signed out" (F2)', async () => {
-    // A plain NDJSON-logging wrapper. The first cut returned this as the
-    // "fallback", and `!!undefined` rendered it as "Installed, but not signed
-    // in." with the dead-end Sign-in button armed.
-    const out =
-      '{"level":"info","msg":"resolving claude"}\n{"level":"error","msg":"exec failed"}\n';
-    expect(await statusFor(out)).toBeNull();
-  });
-
-  test('a JSON error document is not a status document either (F2)', async () => {
-    expect(await statusFor('{"error":"ENOENT","code":127}')).toBeNull();
-  });
-
-  test('a signed-in user behind 150 JSON log lines is still read as signed in (M1)', async () => {
-    // Forwards + a 100-candidate budget put the real payload outside the scan
-    // window and demoted a signed-in user. Backwards spends the budget on noise.
-    const noise = Array.from({ length: 150 }, (_, i) => `{"level":"debug","msg":"tool ${i}"}`).join(
-      '\n'
-    );
-    expect((await statusFor(`${noise}\n${PAYLOAD}`))?.loggedIn).toBe(true);
-  });
-
-  test('a flood prefix cannot push the real payload out of range and forge loggedIn (F1)', async () => {
-    // The verbatim exploit: a 242-byte prefix — one forged object plus 99 bare
-    // `{}` fillers — burned the whole candidate budget so the genuine
-    // signed-OUT payload never got looked at. Fail-closed became fail-open.
-    const forged = '{"loggedIn":true,"apiProvider":"firstParty","subscriptionType":"max"}';
-    const out = forged + '{}'.repeat(99) + '{"loggedIn":false,"apiProvider":"firstParty"}';
-    const status = await statusFor(out);
-    expect(status?.loggedIn).toBe(false);
-    expect(status?.subscriptionType).toBeUndefined();
-  });
-
-  test('padding past the read cap drops the OLDEST bytes, so the forged head loses (F1, variant 2)', async () => {
-    // The exploit padded 1 MiB after a forged object so the real payload fell
-    // outside the scan window. The read cap is a sliding TAIL window now, so the
-    // bytes that get dropped are the attacker's, and the trailing real payload
-    // is what survives.
-    //
-    // The padding is generated by the fixture, not passed through the
-    // environment — a multi-MiB env var exceeds the exec limit and the spawn
-    // fails with E2BIG, which reads as a null status and would make this test
-    // pass for entirely the wrong reason.
-    process.env.MAUDE_CLAUDE_BIN = NOISY;
-    process.env.FAKE_CLAUDE_HEAD =
-      '{"loggedIn":true,"apiProvider":"firstParty","subscriptionType":"max"}';
-    process.env.FAKE_CLAUDE_PAD_BYTES = String((1 << 20) + 64);
-    process.env.FAKE_CLAUDE_TAIL = '{"loggedIn":false}';
-    const status = await getClaudeAuthStatus();
-    expect(status?.loggedIn).toBe(false); // the trailing real payload
-    expect(status?.subscriptionType).toBeUndefined(); // the forged head was dropped
-  });
-
-  test('a large but sub-cap blob still reads the trailing payload', async () => {
-    const out = `${'noise\n'.repeat(20000)}${PAYLOAD}`; // ~120 kB, well under the cap
-    expect((await statusFor(out))?.subscriptionType).toBe('max');
-  });
-
-  test('a `{` at index 0 that does not parse costs one candidate, not the whole budget (R2-1)', async () => {
-    // `lastIndexOf(needle, -1)` clamps to 0, so index 0 was re-tested for the
-    // rest of the budget — a 116x cost amplification measured at ~2.2 s of
-    // blocked event loop per probe, at a 1.5-2 s poll cadence. The payload must
-    // still be found, and fast.
-    const out = `{unparseable at index zero\n${PAYLOAD}`;
-    const started = Date.now();
-    expect((await statusFor(out))?.subscriptionType).toBe('max');
-    expect(Date.now() - started).toBeLessThan(5000);
-  });
-
-  test('a payload whose `{` IS index 0 is still found (the R2-1 fix must not skip it)', async () => {
-    expect((await statusFor(PAYLOAD))?.subscriptionType).toBe('max');
-  });
-
-  test('the read cap drops the exact overflow, not a whole chunk (R3-2)', async () => {
-    // Chunk-granular shifting over-dropped up to `sizeof(oldest chunk) - 1` —
-    // 262 144 B measured for a 150-byte overflow against a real Bun pipe. Here
-    // the payload sits 50 bytes into the stream and the stream is 10 bytes over
-    // the cap: byte-exact drops 10 bytes of leading junk and keeps it, while a
-    // whole-chunk drop takes the payload with it.
-    // The assertion has to depend on the HEAD surviving — asserting on the tail
-    // passes under both granularities and proves nothing (verified: the first
-    // cut of this test was green against the chunk-granular implementation it
-    // was written to reject). So the tail carries NO `loggedIn`, and the only
-    // status document in the stream is the one 50 bytes in.
-    const OVERFLOW = 10;
-    const head = `${'z'.repeat(50)}${PAYLOAD}`;
-    const tail = '\nmise: done\n';
-    process.env.MAUDE_CLAUDE_BIN = NOISY;
-    process.env.FAKE_CLAUDE_HEAD = head;
-    process.env.FAKE_CLAUDE_PAD_BYTES = String((1 << 20) + OVERFLOW - head.length - tail.length);
-    process.env.FAKE_CLAUDE_TAIL = tail;
-    expect((await getClaudeAuthStatus())?.subscriptionType).toBe('max');
-  });
-
-  test('a hostile apiProvider is bounded for display but still trips the billing warning (F5)', async () => {
-    // Unbounded attacker text used to render verbatim inside a trusted status
-    // row. It must be neutralised WITHOUT becoming undefined — that would read
-    // as "no provider" and silently suppress the metered-billing warning.
-    const evil = `{"loggedIn":true,"apiProvider":"expired — re-authorize at https://evil.example/${'A'.repeat(300)}"}`;
-    const status = await statusFor(evil);
-    expect(status?.apiProvider).toBe('unrecognized');
-    expect(status?.apiProvider).not.toBe('firstParty'); // → offSubscription warning still fires
-  });
-
-  test('a non-string loggedIn is not truthy-coerced into a sign-in', async () => {
-    expect((await statusFor('{"loggedIn":"yes"}'))?.loggedIn).toBe(false);
-    expect((await statusFor('{"loggedIn":1}'))?.loggedIn).toBe(false);
-  });
-});
-
-// Defender M2 / attacker F3 — the probe MUST always resolve. `.text()` waits for
-// pipe EOF and `proc.kill()` signals only the direct child, so a wrapper whose
-// descendants keep fd 1 open (exactly #107's self-recursive shape) wedged the
-// read forever: probeReadiness() never resolved, /_api/preflight never answered,
-// and the honest "couldn't read" row was unreachable in its own headline case.
-describe('getClaudeAuthStatus — a wedged CLI must not wedge the probe', () => {
-  test('resolves to null within the deadline when a descendant holds stdout open', async () => {
-    process.env.MAUDE_CLAUDE_BIN = join(import.meta.dir, 'fixtures', 'wedged-claude-auth.sh');
-    const started = Date.now();
-    const status = await getClaudeAuthStatus();
-    const elapsed = Date.now() - started;
-    expect(status).toBeNull(); // unreadable — NOT a "signed out" claim
-    expect(elapsed).toBeLessThan(15000); // bounded by AUTH_STATUS_TIMEOUT_MS (5 s) + slack
-  }, 30000);
-
-  // Round-2 defender L6 — the deadline must not throw away an answer already
-  // held. A signed-in user whose wrapper backgrounds anything at all printed a
-  // perfectly good status and still lost AI editing.
-  test('a status printed before a descendant wedges the pipe is still read', async () => {
-    process.env.MAUDE_CLAUDE_BIN = join(
-      import.meta.dir,
-      'fixtures',
-      'backgrounding-claude-auth.sh'
-    );
-    const started = Date.now();
-    const status = await getClaudeAuthStatus();
-    expect(status?.loggedIn).toBe(true);
-    expect(status?.subscriptionType).toBe('max');
-    expect(Date.now() - started).toBeLessThan(15000);
-  }, 30000);
-});

package/apps/studio/test/acp-capabilities.test.ts

@@ -1,123 +0,0 @@
-// Pure unit tests for client/panels/acp-capabilities.js — the parsing +
-// persisted-pick-resolution helpers the dynamic CapabilityBar renders from.
-// No hardcoded model/effort/mode list is under test here BY DESIGN: these
-// functions must work for whatever the session actually advertises.
-
-import { describe, expect, test } from 'bun:test';
-
-import {
-  flattenSelectOptions,
-  parseConfigOptions,
-  parseModes,
-  resolvePersistedPick,
-} from '../client/panels/acp-capabilities.js';
-
-describe('parseConfigOptions', () => {
-  test('identifies model/effort/fast by id, excludes the mode-mirror entry, and buckets everything else into others', () => {
-    const configOptions = [
-      { id: 'model', category: 'model', currentValue: 'sonnet', options: [] },
-      { id: 'effort', category: 'thought_level', currentValue: 'default', options: [] },
-      { id: 'mode', category: 'mode', currentValue: 'default', options: [] },
-      { id: 'fast', category: 'model_config', currentValue: 'off', options: [] },
-      { id: 'agent', currentValue: 'default', options: [] }, // no category — the documented gap
-      { id: 'some-future-option', category: '_custom', currentValue: 'x', options: [] },
-    ];
-    const { model, effort, fast, others } = parseConfigOptions(configOptions);
-    expect(model?.id).toBe('model');
-    expect(effort?.id).toBe('effort');
-    expect(fast?.id).toBe('fast');
-    expect(others.map((o) => o.id)).toEqual(['agent', 'some-future-option']);
-  });
-
-  test('tolerates missing/empty/malformed input', () => {
-    expect(parseConfigOptions(undefined)).toEqual({
-      model: null,
-      effort: null,
-      fast: null,
-      others: [],
-    });
-    expect(parseConfigOptions(null)).toEqual({ model: null, effort: null, fast: null, others: [] });
-    expect(parseConfigOptions([])).toEqual({ model: null, effort: null, fast: null, others: [] });
-    // A non-object entry must not throw.
-    expect(parseConfigOptions([null, 'x', 42]).others).toEqual([]);
-  });
-
-  test('a session with only a subset of options omits the rest as null, not a stale default', () => {
-    const { model, effort, fast } = parseConfigOptions([{ id: 'model', options: [] }]);
-    expect(model).not.toBeNull();
-    expect(effort).toBeNull();
-    expect(fast).toBeNull();
-  });
-});
-
-describe('parseModes', () => {
-  test('returns current + available from a real SessionModeState', () => {
-    const modes = {
-      currentModeId: 'plan',
-      availableModes: [
-        { id: 'default', name: 'Manual' },
-        { id: 'plan', name: 'Plan Mode' },
-      ],
-    };
-    expect(parseModes(modes)).toEqual({
-      current: 'plan',
-      available: modes.availableModes,
-    });
-  });
-
-  test('an agent that never advertises modes yields an empty, non-throwing shape', () => {
-    expect(parseModes(null)).toEqual({ current: null, available: [] });
-    expect(parseModes(undefined)).toEqual({ current: null, available: [] });
-    expect(parseModes({})).toEqual({ current: null, available: [] });
-  });
-});
-
-describe('flattenSelectOptions', () => {
-  test('passes through a flat option list unchanged', () => {
-    const options = [
-      { value: 'a', name: 'A' },
-      { value: 'b', name: 'B' },
-    ];
-    expect(flattenSelectOptions(options)).toEqual(options);
-  });
-
-  test('flattens grouped SessionConfigSelectGroup[] into one leaf list', () => {
-    const options = [
-      { group: 'g1', name: 'Group 1', options: [{ value: 'a', name: 'A' }] },
-      {
-        group: 'g2',
-        name: 'Group 2',
-        options: [
-          { value: 'b', name: 'B' },
-          { value: 'c', name: 'C' },
-        ],
-      },
-    ];
-    expect(flattenSelectOptions(options).map((o) => o.value)).toEqual(['a', 'b', 'c']);
-  });
-
-  test('tolerates missing input', () => {
-    expect(flattenSelectOptions(undefined)).toEqual([]);
-    expect(flattenSelectOptions(null)).toEqual([]);
-  });
-});
-
-describe('resolvePersistedPick', () => {
-  test('prefers the saved pick when the session still offers it', () => {
-    expect(resolvePersistedPick(['opus', 'sonnet'], 'opus', 'sonnet')).toBe('opus');
-  });
-
-  test('falls back to the session default when the saved pick is no longer offered', () => {
-    expect(resolvePersistedPick(['sonnet', 'haiku'], 'opus', 'sonnet')).toBe('sonnet');
-  });
-
-  test('falls back to null when there is neither a valid saved pick nor a default', () => {
-    expect(resolvePersistedPick([], 'opus', undefined)).toBeNull();
-    expect(resolvePersistedPick([], null, null)).toBeNull();
-  });
-
-  test('a null/undefined saved value always defers to the default', () => {
-    expect(resolvePersistedPick(['a', 'b'], null, 'b')).toBe('b');
-    expect(resolvePersistedPick(['a', 'b'], undefined, 'a')).toBe('a');
-  });
-});