Omen: Code Analysis CLI for AI-Assisted Development
Context
Tech debt backlogs that would take months to clear with a team of 30 devs, but no way to prioritize what actually mattered. PRs slipped through fast reviews in complex areas. Developers new to parts of the codebase didn't understand the nuances. Every change looked the same on the surface, but some touched high-churn, high-complexity code where bugs hide. AI tools could see the code, but not the weight behind it - that knot in your stomach when you open a god object knowing one wrong move could break everything.
Action
Built an open-source code analysis CLI in Go that surfaces the 'omens' - complexity hotspots, defect-prone files, architectural coupling, and technical debt. Started from PMAT/PAIML but rebuilt it for fast installs via Homebrew, cross-platform support, and seamless LLM integration. Added MCP server integration so Claude, Cursor, and Codex can see where the landmines are before writing code. Built quality gates via githooks that force AI to be more careful when touching risky areas. Supports 12+ languages through tree-sitter parsing.
Result
Refactored a large Rails API monolith - complexity dropped from 30 to under 12 across the codebase in 2 hours, zero regressions. Upgraded an Angular app from version 3 to 13 in 2 hours with significant code improvements and no regressions. Historical PR analysis proved the defect prediction works - flagging commits that would have caused production issues. Teams can now track codebase health trends over time and enforce quality standards beyond just test coverage.