# Final Review Report: Job f3699a96 (PR #48 / Issue #45)

## 1. Review Identity & Metadata
- **Loop ID**: `loop-20261005T064728Z-a4e9c8`
- **Job ID**: `f3699a96`
- **Reviewer**: agy (session: `herdr:reviewer-agy-01`)
- **Base Commit**: `c5c7eff6f758c76ac37a138dbc3c301f3f7e10cf` (main, v4.8.4)
- **Head Commit**: `35935f9e8a052ca9e9d9b806a62d17eb72afea78` (10 commits)
- **Tree Hash**: `dff9db13f3e018528d04a024acda942f501513ff`
- **Verification Path**: `/var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1`
- **Git Status**: Clean (working tree clean, no uncommitted changes)
- **Whitespace Check (`git diff base...HEAD --check`)**: Clean (0 errors)

---

## 2. Review Findings & Reasons for NOT PASS

During verification with the specified test command (`AGENT_PYTHON_BIN=... python -m pytest ...`), multiple critical test failures occurred on macOS / Python < 3.11:

### 1) Unconditional `import tomllib` at Module Level in Test Files
- **Failure 1**: [`tests/test_issue45_herdr_config.py:16`](file:///var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1/tests/test_issue45_herdr_config.py#L16)
  ```python
  import tomllib
  ```
  - **Error**: `ModuleNotFoundError: No module named 'tomllib'`
  - **Impact**: Pytest fails collection immediately with exit code 2. The entire test module cannot be run on Python < 3.11 (such as Python 3.9 in the project virtualenv).
  - **Context**: While [`lib_py/herdr_config.py:152-155`](file:///var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1/.agents/skills/lib_py/herdr_config.py#L152-L155) guards the import with `try ... except ImportError: tomllib = None`, the test module itself imports `tomllib` unconditionally at line 16 and in test functions.

- **Failure 2**: [`tests/test_issue9_server_env.py:1592`](file:///var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1/tests/test_issue9_server_env.py#L1592)
  ```python
  def test_bootstrap_hands_the_server_a_no_restore_config(mam_sandbox):
      import tomllib
  ```
  - **Error**: `ModuleNotFoundError: No module named 'tomllib'`
  - **Impact**: Test fails immediately on Python < 3.11.

---

### 2) Non-Portable Linux Assumptions in Real Herdr Server Test (`test_issue45_restore.py`)
Four out of 5 tests in [`tests/test_issue45_restore.py`](file:///var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1/tests/test_issue45_restore.py) fail on macOS:

- **Failure 3 (Linux `/proc` filesystem hardcoding)**: [`tests/test_issue45_restore.py:227`](file:///var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1/tests/test_issue45_restore.py#L227)
  ```python
  environ = Path(f"/proc/{rig.server_pids()[0]}/environ").read_bytes().split(b"\0")
  ```
  - **Error**: `FileNotFoundError: [Errno 2] No such file or directory: '/proc/26854/environ'`
  - **Impact**: Tests `test_a_mam_server_does_not_relaunch_saved_agents[plain-user-config]` and `test_a_mam_server_does_not_relaunch_saved_agents[user-config-with-session-table]` both crash. macOS (Darwin) does not have a `/proc` filesystem. Reading `/proc/<pid>/environ` is non-portable.

- **Failure 4 (macOS `/tmp` vs `/private/tmp` symlink mismatch)**: [`tests/test_issue45_restore.py:263`](file:///var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1/tests/test_issue45_restore.py#L263)
  ```python
  assert _pane_with(rig, ids["Y"])["cwd"] == str(rig.root)
  ```
  - **Error**: `AssertionError: assert '/private/tmp/m45phdbspj1' == '/tmp/m45phdbspj1'`
  - **Impact**: `test_b_only_the_resumed_sessions_restored_shell_is_closed` fails. On macOS, `/tmp` is a symlink to `/private/tmp`. Herdr canonicalizes working directories to `/private/tmp/...`. Exact string comparison against uncanonicalized `str(rig.root)` fails. Comparing with `Path(cwd).resolve()` or `os.path.samefile` is required.

- **Failure 5 (Pane lookup failure due to uncanonicalized `/tmp` path)**: [`tests/test_issue45_restore.py:325`](file:///var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1/tests/test_issue45_restore.py#L325)
  ```python
  pane = next(p for p in rig.panes() if p.get("cwd") == str(ws))
  ```
  - **Error**: `StopIteration`
  - **Impact**: `test_b_resume_leaves_no_restored_shell_behind` fails. `ws` is constructed under `/tmp/...`, while `p.get("cwd")` is reported as `/private/tmp/...`. The generator finds no match and raises `StopIteration`.

---

## 3. Evaluation of Brief Review Questions

1. **A's config merging**:
   - On Python 3.11+, parsing and rewriting logic in `lib_py/herdr_config.py` preserves tables, dotted keys, and verifies with `tomllib.loads(text) == expected` followed by `herdr config check`.
   - On Python < 3.11 (e.g. system Python 3.9), `tomllib` is absent, causing `server_config_text()` to emit a warning and fall back to `MINIMAL` (`[session]\nresume_agents_on_restore = false`), thereby losing user-defined herdr settings on Python 3.9 systems.
2. **Coverage across server boot paths**:
   - `lib.sh:230-236` runs `python3 -m lib_py.herdr_config server-config` during headless bootstrap.
   - Accurately documented that manually attached/started servers outside MAM do not pick up `mam-server-config.toml`.
3. **B's pane closing conditions**:
   - Clean fail-safe in `lib_py/herdr_restore.py`: checks `HERDR_SESSION_NAME` is not default, ensures no live agent is running in the pane, checks matching `agent_session`, and verifies foreground process group is the shell. If herdr cannot be queried, it fails-safe by leaving panes untouched and emitting a warning in `resume_session.sh:614`.
4. **Test isolation & portability**:
   - Test suites properly use temporary HOME and isolated session names, preventing mutation of live user servers or configs.
   - However, portability on macOS is broken due to hardcoded `/proc` access and uncanonicalized `/tmp` symlink path comparisons.
5. **Version and Inventory**:
   - Framework and 8 skills were updated to `4.8.5`.
   - `tests/data/skill_code_blocks.json` updated line ranges for `create` and `resume` while preserving classification.
6. **PR purge failure note**:
   - Baseline purge test passes 23/23 on macOS and is unrelated.
7. **Boundary (#17)**:
   - Complex logic is implemented in Python (`herdr_config.py`, `herdr_restore.py`), while shell scripts delegate to them.

---

## 4. Test Suite Execution Summary

Executed in verification worktree (`/var/folders/q_/68sxgw3j073dgjg8wl9s08vw0000gn/T/mam-review/loop-20261005T064728Z-a4e9c8/1`):

| Test Suite | Result | Details |
|---|---|---|
| `tests/test_issue45_herdr_config.py` | **FAILED** | Collection error: `ModuleNotFoundError: No module named 'tomllib'` at line 16 |
| `tests/test_issue9_server_env.py` | **FAILED** | `test_bootstrap_hands_the_server_a_no_restore_config` failed at line 1592 (`ModuleNotFoundError`) |
| `tests/test_issue45_restore.py` | **FAILED** (4/5 failed) | Lines 227 (`/proc` FileNotFoundError), 263 (`/tmp` vs `/private/tmp` AssertionError), 325 (`StopIteration`) |
| `tests/test_issue23_24_startup_path.py` | 27 passed | Clean |
| `tests/test_issue42_skill_blocks.py` | 37 passed, 1 skipped | Clean |
| `tests/test_version_consistency.py` | 3 passed | Clean |

---

## 5. Conclusion

Due to the critical test failures on macOS and Python < 3.11 (`ModuleNotFoundError: tomllib` in test suites, Linux `/proc` hardcoding, and macOS `/private/tmp` path matching failures in `test_issue45_restore.py`), the review verdict is NOT PASS.

[VERDICT: NOT PASS]