# Plan: fix_test_failures_20260624

3 phases, 3 tasks, 3 atomic commits. TDD: write the failing test invocation first (the existing 14 tests are the red), fix the code (green), commit.

## Phase 1: NormalizedResponse dual-signature (1 task)

Focus: Add custom `__init__` to `NormalizedResponse` that accepts both new nested `usage: UsageStats` and legacy flat kwargs. This fixes 12 of the 14 failing tests in one place.

- [x] Task 1.1: Add custom `__init__` to `NormalizedResponse` in `src/openai_schemas.py`. [1b39aae7]
  - WHERE: `src/openai_schemas.py:75-80` (the `@dataclass(frozen=True) class NormalizedResponse:` block)
  - WHAT:
    - Change `@dataclass(frozen=True)` to `@dataclass(frozen=True, init=False)` on line 75
    - Add a custom `__init__` method (after the class fields) that:
      - Signature: `(self, text: str, tool_calls: tuple[ToolCall, ...] = (), usage: Optional[UsageStats] = None, raw_response: Any = None, usage_input_tokens: Optional[int] = None, usage_output_tokens: Optional[int] = None, usage_cache_read_tokens: Optional[int] = None, usage_cache_creation_tokens: Optional[int] = None) -> None`
      - If `usage is None` and any legacy flat kwarg is non-None: build `UsageStats(input_tokens=usage_input_tokens or 0, output_tokens=usage_output_tokens or 0, cache_read_tokens=usage_cache_read_tokens or 0, cache_creation_tokens=usage_cache_creation_tokens or 0)`
      - If `usage is None` and all legacy flat kwargs are None: build `UsageStats(0, 0)`
      - Otherwise: use the provided `usage`
      - Use `object.__setattr__(self, "text", text)` for all 4 field assignments (required because `frozen=True` locks `__setattr__`)
  - HOW: Use `manual-slop_py_update_definition` to replace the entire class block. The new content includes the `init=False` decorator and the custom `__init__` method.
  - SAFETY:
    - Verify with `ast.parse(open("src/openai_schemas.py").read())` (no syntax errors)
    - Run `uv run python -c "from src.openai_schemas import NormalizedResponse, UsageStats; r = NormalizedResponse(text='hi', tool_calls=(), usage_input_tokens=10, usage_output_tokens=5, raw_response=None); print(r.usage.input_tokens, r.usage.output_tokens)"`
    - Run the 12 affected tests
  - COMMIT: `fix(schemas): add legacy-kwarg backward compat to NormalizedResponse.__init__`
  - GIT NOTE: 12 tests fixed by single backward-compat __init__; supports both new nested usage: UsageStats and legacy flat usage_input_tokens=... kwargs
  - VERIFY: `uv run pytest tests/test_ai_client_cli.py tests/test_ai_client_tool_loop.py tests/test_ai_client_tool_loop_builder.py tests/test_ai_client_tool_loop_send_func.py tests/test_gemini_cli_integration.py tests/test_gemini_cli_parity_regression.py tests/test_gemini_cli_edge_cases.py -v` shows 12/12 pass

## Phase 2: Session frozen fix (1 task)

Focus: Update the test to use `dataclasses.replace()` instead of mutating the frozen `Session`.

- [ ] Task 2.1: Update `test_auto_whitelist_keywords` to use `dataclasses.replace`.
  - WHERE: `tests/test_auto_whitelist.py:20`
  - WHAT: Change `reg.data[session_id]["whitelisted"] = True` to `reg.data[session_id] = dataclasses.replace(reg.data[session_id], whitelisted=True)`
  - HOW: Use `manual-slop_edit_file` with the exact old/new text. Add `import dataclasses` to the top of the file if not already imported.
  - SAFETY: Verify with `uv run pytest tests/test_auto_whitelist.py -v`
  - COMMIT: `test(auto-whitelist): use dataclasses.replace for frozen Session mutation`
  - GIT NOTE: 1 test fixed; the old `["whitelisted"] = True` was invalid on a frozen dataclass
  - VERIFY: `uv run pytest tests/test_auto_whitelist.py -v` shows 0 failures

## Phase 3: Toggle race fix (1 task)

Focus: Replace the non-deterministic toggle with a deterministic close path.

- [ ] Task 3.1: Update `test_palette_starts_hidden` to use a deterministic close callback.
  - WHERE: `tests/test_command_palette_sim.py:38` (the `client.push_event("custom_callback", {"callback": "_toggle_command_palette", "args": []})` call)
  - WHAT:
    - Check `_predefined_callbacks` in `src/api_hooks.py` for a close callback (e.g., `_close_command_palette` or similar). If found, use it.
    - If no close callback exists, push `_toggle_command_palette` once, poll for state, then push again to guarantee the palette is closed.
  - HOW: Use `manual-slop_edit_file` with the exact old/new text.
  - SAFETY: The test must end with `show_command_palette is False` regardless of starting state. Verify with `uv run pytest tests/test_command_palette_sim.py -v` in isolation AND in the full batch (per the `Isolated-Pass Verification Fallacy` rule in `conductor/workflow.md`).
  - COMMIT: `test(palette): use deterministic close instead of toggle for palette_starts_hidden test`
  - GIT NOTE: 1 test fixed; toggle is non-deterministic (close-if-open, open-if-closed); new code forces close regardless of starting state
  - VERIFY: `uv run pytest tests/test_command_palette_sim.py -v` AND `uv run python scripts/run_tests_batched.py` (tier-3-live_gui)

## Phase 4: Verification (1 task)

Focus: Confirm all 10 verification criteria pass.

- [ ] Task 4.1: Run all 6 VCs; write the track's end-of-track report.
  - WHERE: Full test suite + audit gates
  - WHAT:
    - Run VC1-VC6 (the 6 verification criteria from the spec)
    - Write `docs/reports/TRACK_COMPLETION_fix_test_failures_20260624.md` with the verification results
    - Update this track's `state.toml` to `status = "completed"`, `current_phase = "complete"`, all 4 phases `completed`
    - Update `conductor/tracks.md` to add a row for this track
  - HOW: Run each VC command, capture output, write the report.
  - SAFETY: The 2 pre-existing-violation audit gates (NG1, NG2 from the polish track) are still out of scope. Do not regress them.
  - COMMIT: 3 commits: `conductor(state): fix_test_failures_20260624 SHIPPED`, `docs(reports): TRACK_COMPLETION for fix_test_failures_20260624`, `conductor(tracks): add fix_test_failures_20260624 row`
  - GIT NOTE: 1 per commit per workflow.md
  - VERIFY: All 6 VCs pass; the 14 previously-failing tests are now green; no new failures introduced

## Commit Log (Expected)

1. `fix(schemas): add legacy-kwarg backward compat to NormalizedResponse.__init__` (Task 1.1)
2. `test(auto-whitelist): use dataclasses.replace for frozen Session mutation` (Task 2.1)
3. `test(palette): use deterministic close instead of toggle for palette_starts_hidden test` (Task 3.1)
4. `conductor(state): fix_test_failures_20260624 SHIPPED` (Task 4.1)
5. `docs(reports): TRACK_COMPLETION for fix_test_failures_20260624` (Task 4.1)
6. `conductor(tracks): add fix_test_failures_20260624 row` (Task 4.1)

## Verification Commands (run at end of Phase 4)

```bash
# VC1: the 12 NormalizedResponse tests pass
uv run pytest tests/test_ai_client_cli.py tests/test_ai_client_tool_loop.py tests/test_ai_client_tool_loop_builder.py tests/test_ai_client_tool_loop_send_func.py tests/test_gemini_cli_integration.py tests/test_gemini_cli_parity_regression.py tests/test_gemini_cli_edge_cases.py -v

# VC2: test_auto_whitelist_keywords passes
uv run pytest tests/test_auto_whitelist.py -v

# VC3: test_palette_starts_hidden passes
uv run pytest tests/test_command_palette_sim.py -v

# VC4: full batched suite is green
uv run python scripts/run_tests_batched.py

# VC5: 4 audit gates
uv run python scripts/audit_exception_handling.py
uv run python scripts/audit_weak_types.py
uv run python scripts/audit_main_thread_imports.py
uv run python scripts/audit_no_models_config_io.py

# VC6: confirm no new failures (diff against pre-fix summary)
```