菜单
Deterministic, advisory lint/format/type audit for Python. Uses ruff lint (with dead-code and complexity codes excluded), config-gated ruff format --check, and a type checker to emit LINT / FORMAT / TYPE findings. Never mutates source.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | quality-audit |
| version | 0.8.0 |
| description | Deterministic, advisory lint/format/type audit for Python. Uses ruff lint (with dead-code and complexity codes excluded), config-gated ruff format --check, and a type checker to emit LINT / FORMAT / TYPE findings. Never mutates source. |
A code-health leaf skill reporting lint violations, declared-standard formatting
drift, and type errors as advisory findings. It never runs ruff --fix or
reformats.
python3 scripts/quality_audit.py \
--root /path/to/repo \
--source-prefix src/pkg/ \
--out-dir /tmp/quality
quality_findings.json — sorted findings (shared schema).quality_report.md — summary grouped by signal plus format-check status.format_check and suppressed_format_files.0 clean, 1 advisory findings present, 2 tool/config error.--config JSON keys: type_checker ("mypy"|"ty"), ruff_select, ruff_ignore. The
default ruff ignore excludes F401/F811/F841/C901 (owned by other leaves). Findings are
deterministic.
ruff format --check runs only when the audited root declares a formatting
standard via .ruff.toml, ruff.toml, [tool.ruff...], or [tool.black...]
in pyproject.toml. Without one, FORMAT findings are skipped and counted as
suppressed_format_files; lint and type checks still run.
Deterministic, advisory mutation-testing audit for Python. Wraps mutmut==3.6.0 to measure per-module mutation kill rates, emitting TEST findings (shared code-health schema) for modules whose kill rate falls below the configured threshold. Never mutates source in-place; runs in a sandbox copy. The machine-readable answer to "how good are my tests at catching bugs?".
Umbrella that runs the code-health leaf skills (complexity, duplication, dead-code, structure, quality) in parallel, merges and ranks their findings into one report, and emits a supervisor decision with exit codes 0/1/2. Reads a leaf registry so new language leaves plug in without changing the orchestrator. Advisory only.
Deterministic, advisory complexity and maintainability audit for Python. Uses lizard (per-function cyclomatic complexity, length, parameters) and radon mi (per-module maintainability index) to emit SIMPLIFY / DECOMPOSE findings to the shared code-health finding schema. Never mutates source.
Deterministic, advisory testedness audit for Python. Consumes coverage.py JSON report(s) and emits TEST findings (shared code-health schema) for production files with no or insufficient test coverage. Never runs tests, never mutates source. The machine-readable answer to "is this file safe to refactor?".
Deterministic, advisory dead-code audit for Python. Uses vulture (unused functions/classes/methods/properties) and ruff F401/F811/F841 (unused imports, redefinitions, unused locals) to emit DELETE findings to the shared code-health finding schema. Never mutates source.
Deterministic, advisory dependency audit for Python. Compiles declared dependencies (pyproject.toml [project] or requirements*.txt) against AST-collected imports and reports unused declarations, undeclared imports, test-only runtime usage, and optional C-8 advisory enrichment. Uses stdlib only, requires Python >=3.11, never mutates source, and never makes network calls.