Your AI coding agent scores 10/100. Here's what it's missing.

After testing 2,431 checks across 8 AI coding platforms on real repos, I found a clear pattern: most projects use barely 10% of what's available. I built Nerviq — a zero-dependency CLI that audits your AI coding agent setup and scores it 0-100. npx @nerviq/cli audit Most projects score 10-20 out of 100 . After running setup, they jump to 60-80+ . The Top 10 Things You're Probably Missing 1. Instructions file (Critical) Every AI coding platform has one: CLAUDE.md , AGENTS.md , .cursorrules , GEMINI.md . Without it, the agent doesn't know your build commands, code style, or project rules. 2. Architecture diagrams (73% token savings) A Mermaid diagram gives your agent the project structure in a fraction of the tokens that prose requires. 3. Hooks > instructions (100% vs 80% compliance) Written instructions are advisory (~80% compliance). Hooks are deterministic (100%). Auto-lint after every edit. Every time. 4. Verification commands This is the single highest-leverage thing you can do. —