rafaie/codex_home/test-runner

1) Read codex.toml (if present) for canonical commands.

• When a work item is provided, read its test-plan.md for smoke applicability.

2) Resolve commands (prefer codex.toml, otherwise defaults):

• format: uv run ruff format . --check
• lint: uv run ruff check .
• typecheck: uv run mypy src
• docstrings: optional [commands].docstrings
• quick tests: test_quick, else test, else uv run pytest -q
• full tests: test_full, else test, else uv run pytest -q
• smoke: smoke, else uv run python scripts/smoke.py if present

3) Support two modes:

• Quick mode:
• format/lint
• docstrings when [commands].docstrings is configured and the work item changes Python APIs
• work-item-scoped or quick tests (test_quick fallback chain)
• smoke-test when applicability is required
• Full mode:
• format/lint
• typecheck
• docstrings when [commands].docstrings is configured
• full test suite (test_full fallback chain)
• smoke-test when applicability is required

4) Run smoke as its own stage after tests/checks and report it as a separate result bucket.

• Run smoke only after all preceding stages pass; otherwise report smoke: not_run (blocked by earlier gate) to avoid unnecessary live calls.
• Skip smoke only when the work item explicitly records not_applicable with a non-runnable rationale; report smoke: not_applicable and the rationale.
• If no work item is provided, run smoke when a smoke command/script exists; otherwise report not_configured rather than inventing evidence.
• In full mode, this smoke stage is the canonical smoke execution consumed by ship-feature.
• Emit or capture the smoke artifact path so downstream skills do not rerun live smoke unnecessarily.

5) Summarize results:

• Mode used and commands run
• What passed
• What failed (top 3 failure signatures)
• Smoke result (pass/fail + artifact path when available)
• Smoke applicability and any skip/block reason
• Docstring result when configured
• Likely category: env/setup vs flaky vs deterministic bug vs expectation mismatch

6) Recommendation logic:

• If unit/tests pass but smoke fails, prioritize packaging/runtime and invoke debug-loop.
• If there are many failures, invoke failure-triage.
• If failures appear intermittent, invoke flaky-test-hunter.