I ran 50+ AI code reviews against hand-written review skills on a production codebase. The generated skills found a P1 data loss bug the humans missed. The humans caught platform-specific edge cases the AI never would. Neither won.
So I built vibecheck — it scans your repo, detects your tech stack, builds a useful servicemap, and generates Claude Code skills with platform-specific pre-flight checks and four review modes (PR, component, deep, and full-repo). Supports 30+ platforms across backend, web, mobile, and infrastructure.
The generated security review builds a threat model first, then runs OWASP-mapped checklists, then does agentic attack chain reasoning. The peer review evaluates across 8 lenses. Both have a deep mode that traces flows across service boundaries instead of just scanning files — that's the mode that found the data loss bugs.
So I built vibecheck — it scans your repo, detects your tech stack, builds a useful servicemap, and generates Claude Code skills with platform-specific pre-flight checks and four review modes (PR, component, deep, and full-repo). Supports 30+ platforms across backend, web, mobile, and infrastructure.
The generated security review builds a threat model first, then runs OWASP-mapped checklists, then does agentic attack chain reasoning. The peer review evaluates across 8 lenses. Both have a deep mode that traces flows across service boundaries instead of just scanning files — that's the mode that found the data loss bugs.