Systematic code review skill for AI agents — finds real bugs, not style issues
Build a systematic code review skill system for AI agents to detect real bugs and automate GitHub/GitLab issue workflows
I'm building a code review skill system for AI coding agents like Claude Code. I want to create two main skills that help developers systematically find bugs and manage GitHub/GitLab issues.
The first skill, called "find-mismatch", should be a comprehensive bug detection tool that scans entire codebases looking for real bugs, not style issues. It should detect cross-boundary contract mismatches like function name/parameter/return type mismatches, serialization gaps like casing issues and optional vs required field mismatches, logic bugs like off-by-one errors and dead code, property access errors and null dereferences, async and concurrency bugs like missing awaits and race conditions, placeholder code like TODOs and unused imports, and language-specific type system gaps. The skill should be smart enough to understand the specific language being reviewed.
The second skill, called "work-on-issues", should automate the entire issue workflow. It connects to GitHub or GitLab, lists open issues, lets the user pick which ones to work on, then automatically creates branches, makes commits, opens pull requests/merge requests, and closes the issues when done. It should auto-detect whether the repo uses GitHub or GitLab based on the git remote.
Both skills should be installed via an npm command and work as slash commands within Claude Code or similar AI agents. The system should be structured so users can call /find-mismatch to start bug detection or /work-on-issues to start the issue workflow. Make this installable as a skill package, organized in a skills directory with proper documentation for each skill. The whole thing should feel like a natural extension of AI coding agents, making it easy for developers to integrate systematic code review and issue management into their workflows without complex setup.