NB As of this writing, everything in this repo but this single line has been 100% AI generated, and 0% reviewed or even used by a human.
Iteratively raises a Python repo's mutation score (mutmut) to a threshold by writing tests that kill the highest-value surviving mutants. Tests only — source code is never modified.
| Layer | What | Where |
|---|---|---|
| Brain | driver-mutation report CLI: mutmut, scoring, ranking, all loop state and verdicts |
src/driver_mutation/ |
| Conductor | Stateless workflow loop: report → branch → parallel fixers | skill/workflow.js |
| Hands | Fixer agents writing killing tests | spawned by the workflow |
| Launcher | /driver-mutation Claude Code skill |
skill/SKILL.md |
In a Claude Code session, in the target repo (needs [tool.mutmut] in its
pyproject.toml and a clean git tree):
/driver-mutation --threshold 85 --top 5
Standalone report (inside the target repo's environment):
driver-mutation report --repo . --threshold 85
Ranking heuristic: 0.5·cluster + 0.3·fan-in + 0.2·churn, min-max normalized
per run. Design details: docs/superpowers/specs/2026-07-03-driver-mutation-design.md.
UV_INDEX_URL=https://pypi.org/simple uv sync
uv run pytest