mikiships/agentmd

Analyzes codebases across 6+ languages and automatically detects subsystem boundaries to generate tiered context files (Tier 1 manifest + per-subsystem Tier 2 files), validated against Next.js standards. Core commands include `scan` for project inspection, `generate` for multi-agent output (Claude/Cursor/Copilot), and `eval` for measuring performance impact via optional coderace integration. Research-backed `--minimal` mode strips verbose guidance to focus on commands agents can't infer, addressing findings that verbose context files reduce task success by ~20%.