code-quality

Installation
SKILL.md

Code Quality

Improve code comprehension and change safety using evidence from the project. Do not infer authorship from style, polish, or the absence of workaround comments.

UTILITY SKILL. INVOKES: scoped code inspection and, only in simplify mode, requested code edits and checks. FOR SINGLE OPERATIONS: assess one supplied code target and, when requested, apply its behavior-preserving findings.

DO NOT USE FOR:

  • Dedicated audits whose primary goal is identifying avoidable complexity. Assess complexity here only as a dimension of a broader production-code quality assessment.

  • AI-authorship detection or formatting-only reviews.

  • Comprehensive security audits, performance investigations, or test-suite audits. Assess performance costs and behavioral tests only as dimensions of a code-quality review; do not use this skill when a performance investigation or test-suite audit is the primary task.

  • Isolated test-code reviews: the primary target is test code and its assertion quality, determinism, or isolation. Assess behavioral tests here only as evidence for the production-code target.

  • Standalone refactoring or simplification requests: the requested deliverable is restructured or simplified code, whether the goal is open-ended or the transformation is preselected. Apply edits here only as follow-through on a requested code-quality assessment. For a mixed request, use this skill only when the user requests that assessment and limits edits to its supported findings.

Mode and scope

Use review by default. Use simplify only when the user requests a code-quality assessment and application of its behavior-preserving findings. A standalone request to simplify code does not select this mode. Explicit review-only or dry-run instructions take precedence; unresolved conflicting edit instructions require clarification before edits.

Use supplied code snippets, diffs, or named paths. For snippets, cite supplied snippet lines; report unavailable required caller or test context as not-assessed review coverage. Do not select workspace changes instead of a supplied snippet. If no code target is supplied, inspect staged and unstaged tracked changes and related, nonignored untracked source/test files. Establish relevance from the requested task, affected module, imports, or tests; exclude unrelated, generated, and vendor files from automatic selection. Explicitly supplied targets override these file-selection exclusions, not the task exclusions in DO NOT USE FOR. If no target can be selected unambiguously, request one and return Blocked. Do not default to a repository-wide audit. Inspect callers, tests, and neighboring code as context without expanding the edit scope.

Installs
5
GitHub Stars
2
First Seen
Oct 10, 2026