| name | testing-strategy-review |
| description | Review and improve SoulMap AI test strategy so pytest coverage, eval coverage, and smoke checks match the real repo risks. |
Testing strategy review
Use this skill when deciding what tests to add, tighten, or remove after changes to
code, docs contracts, local workflow layers, or eval surfaces.
Do not use this skill for
Mission
Keep test coverage targeted, honest, and proportional to the repo's real contracts.
For this repo, prefer Python-focused tests over doc or workflow policing tests.
Sources to check first
../rules/evals-and-testing.md
docs/engineering/TESTER.md
tests/
evals/
- the changed implementation or docs files
What to look for
- behavior changes with no matching regression test
- docs or workflow contracts that should be protected by evals, smoke checks, or
review instead of broad pytest policing
- duplicated tests that do not protect meaningful behavior
- eval-only claims that are missing source markers or coverage links
- stale tests asserting an old contract
- pytest files that only police Markdown, docs, README wording, local AI workflow
structure, or other non-executable surfaces
Workflow
- Read the changed files and identify the real contract that moved.
- Decide whether the right protection is a unit test, eval case, smoke check, or
explicit review step.
- Default to Python runtime tests. Avoid new pytest files for non-Python surfaces
unless executable behavior would otherwise be unprotected.
- Add the smallest high-signal coverage that would catch the regression again.
- Remove or update stale assertions that no longer match the real contract.
- Run the narrowest checks first, then the broader suite when needed.
Expected output
Findings
List the coverage gaps or stale tests first.
Fixes
Summarize what test strategy was added, tightened, or cleaned up.
Validation
State which test commands or eval commands were run.
Definition of done
The changed surface should be:
- protected by the right kind of test
- free of stale contract assertions
- consistent with docs and eval claims
- validated with the commands that matter for that layer