repovibe

A one-shot, zero-agent static scanner that grades how well a repository is structured for AI-agent development, then roasts it.

npx repovibe .

No API key, no LLM call, runs offline and in CI. It measures the gap between a "vibecoded" repo and one an AI agent can actually consume: giant files, dumping-ground modules, circular imports, missing CLAUDE.md, committed secrets, and no way to run the tests. That gap becomes a single letter grade.

Usage

npx repovibe [path] [flags]

| Flag | Behaviour | |---|---| | (default) | Terminal grade card (the roast). | | --json | Stable JSON report to stdout. Exits non-zero below --min-score or on a hard cap. CI-friendly. | | --verbose | Per-axis remediation with fix links. | | --min-score N | Failing threshold for the --json exit code (default 60). | | --no-color | Plain output, no ANSI colour. |

What it grades

Seven weighted axes: AI-context files (CLAUDE.md/README), hygiene & secrets, code-quality tells, file-size health, structure & modularity, import coupling & cycles, and runnability. A committed secret or tracked build output caps the grade at F regardless of the rest.

CI example

npx repovibe . --json --min-score 70   # exit 1 if the repo grades below 70 or hard-caps

Fix a bad grade

repovibe tells you how bad your repo is; agent-starter's adopt fixes it.

npx agent-starter adopt

MIT © sneg55