Python coding standards for the uv + ruff + pytest + pyright toolchain:
uv owns environments, dependencies, and packaging; ruff is linter and
formatter; pytest runs the tests; pyright checks types (with Astral's
ty as the watched successor). Standards follow community consensus —
PEPs, the PyPA packaging guide, official tool docs — with the author's
preferences on contested points.
| Skill | Scope |
|---|---|
python-code-style |
naming, idioms, ruff as lint + formatter, pyproject config |
python-project-layout |
uv, pyproject.toml, src layout, entry points, module boundaries, the mise↔uv ownership boundary |
python-typing |
annotations, strictness policy, protocols vs ABCs, pyright configuration |
python-testing |
pytest structure, fixtures, parametrization, coverage |
python-cli |
typer CLIs, config handling, logging, distribution via uv tool |
python-web-api |
FastAPI layering (router → service → repository), pydantic models, async boundaries, HTTP error handling |
Each skill ships reference/ examples alongside its SKILL.md.
python-code-reviewskill — audits Python code against the standards skills; writes one review report per run todocs/code-review/.python-code-revieweragent — reviews a diff or named files, writes the report todocs/code-review/, returns findings by severity./python-reviewcommand — review the current diff or named files. Python files only; out-of-domain files are noted as out of scope.
When the working-process plugin's review-reports rule is installed (detected by its contract probe), that rule's report contract is authoritative; without it, the skill's minimal inline fallback applies — a strict subset of the same contract, so the report shape never depends on the install profile.
Naming is one python- family throughout; skills are named for the
activity (python-code-review), the agent for the actor
(python-code-reviewer) — the same activity/actor distinction
working-process draws between its *-session skills and its review
agents.
python-plan-reviewskill — plan-review checklist discovered by the working-process plan-adversary agent when a reviewed plan touches Python: interpreter versions, lockfile behavior, async boundaries, typing limits, pytest coverage.rules/python-toolchain.md— a Rules payload distributed by working-process's sync-rules engine; states the toolchain and points at the skills conditionally.
Integration assumes working-process ≥ 0.5.0 when it is installed; the plugin has no dependency on it — a standalone install is fully functional, and every working-process mention in content is conditional.