diff to merge ready code

The problem

Standard AI code reviews often give generic advice that ignores your project's specific idioms or dumps too many low-value suggestions at once. You end up manually filtering noise or fixing style nits that don't match your team's real patterns.

What it does

• Identifies the review scope by automatically detecting staged changes or specific commit ranges.
• Learns your local codebase conventions, including error handling patterns and naming styles, before reviewing.
• Categorizes findings by severity (Critical, Warning, Suggestion) and provides concrete failure scenarios for each.
• Applies approved fixes directly to your local files while maintaining your project's architectural style.
• Provides a clear audit trail of what was fixed and what was skipped, leaving you in full control of the final commit.

Frameworks & tools

Works with any language or framework via git. Integrates with CLAUDE.md, CONTRIBUTING.md, and local linter configurations.

Why this beats prompting it yourself

Generic prompts just summarize code or apply boilerplate patterns. This skill enforces a multi-step workflow that understands repo-specific context, mandates a user-approval gate, and focuses only on findings with plausible failure scenarios to minimize noise.

Use cases

• Review and fix staged changes before running git commit.
• Audit the delta between your feature branch and the main branch.
• Scan the last commit for security vulnerabilities or performance bottlenecks.
• Ensure new code strictly follows existing project naming and testing conventions.