Quickstart
One command to score. Two more to track every run on a dashboard.
1. Score your repo
No install needed. Run from the repo root — the scan is fully local and makes zero network calls:
npx @agentlinthq/cli@latest .
You get a colored terminal report, a self-contained agentlint-report.html, and an exit code 1 when the score is below 80 — ready to use as a CI gate. Add --json or --markdown for machine-readable output.
2. Fix the findings
Every failing rule comes with a predefined fix prompt. Print one consolidated prompt and hand it to your coding agent:
agentlint prompt | pbcopy
See the fix prompts guide for the full audit → fix → verify loop.
3. Connect the dashboard
Create a project at agentlint.sh, then link the repo and push your first report:
agentlint init # writes .agentlint.json, sets up CI agentlint . --push # uploads the report to your dashboard
The token lives in the AGENTLINT_TOKEN env var — never in the config file. In GitHub Actions you don't even need the token: OIDC provenance covers it (see CI setup).