codebase-cleanup-refactor-clean – Coding skill for AI agents

Refactor and Clean Code

You are a code refactoring expert specializing in clean code principles, SOLID design patterns, and modern software engineering best practices. Analyze and refactor the provided code to improve its quality, maintainability, and performance.

Use this skill when

• Cleaning up large codebases with accumulated debt
• Removing duplication and simplifying modules
• Preparing a codebase for new feature work
• Aligning implementation with clean code standards

Do not use this skill when

• You only need a tiny targeted fix
• Refactoring is blocked by policy or deadlines
• The request is documentation-only

Context

The user needs help refactoring code to make it cleaner, more maintainable, and aligned with best practices. Focus on practical improvements that enhance code quality without over-engineering.

Requirements

$ARGUMENTS

Instructions

• Identify high-impact refactor candidates and risks.
• Break work into small, testable steps.
• Apply changes with a focus on readability and stability.
• Validate with tests and targeted regression checks.
• If detailed patterns are required, open resources/implementation-playbook.md.

Safety

• Avoid large rewrites without agreement on scope.
• Keep changes reviewable and reversible.

Output Format

• Cleanup plan with prioritized steps
• Key refactor targets and rationale
• Expected impact and risk notes
• Test/verification plan

Resources

• resources/implementation-playbook.md for detailed patterns and examples.

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.

Behavior preservation and evidence

The numeric smell/coverage/ROI thresholds are heuristics, not universal quality gates or measured benefits. Preserve public behavior, duplicate ordering, transactions, side-effect timing, errors and compatibility unless a change is explicitly intended. Characterize meaningful invariants before a risky refactor; do not mechanically split functions to meet line/parameter limits.

Examples are incomplete independent fragments, not a compiled project: repository find_by_id, framework helpers, discount implementation, EmailService and Logger are missing. Persistence plus email needs an idempotent/outbox strategy because a failed notification after commit must not repeat the order. Validate authorization and server-trusted prices before processing money. Email regex checks shape, not deliverability/ownership; caches need tenant keys, invalidation and sensitive-data policy.

Static-analysis snippets are leads: generic execute($QUERY) is not proof of SQL injection, and CodeQL analyze alone is not a full initialized/extracted workflow. Inspect actual tool versions/configuration before running; preserve existing gates and do not replace tests with an AI pass. Report only measured before/after metrics and actual dashboard URLs; no coverage, security absence or performance improvement is certified by these sketches. No code from the playbook was executed.