How Code Mower works

Install Code Mower from the public repository at github.com/codemower-ai/code-mower, then run code-mower init --easy in the repository you want to pilot.

code-mower doctor --supervised-pilot --repo OWNER/REPO checks Python, GitHub auth, private-repo Actions cost traps, provider CLI setup, token posture, and optional cloud readiness. Warnings are setup guidance before any reviewer lane surprises the team. --preflight remains the compatibility preset.

Codex and Claude are the first structured local audit lanes. Hosted reviewers and extra provider integrations start informational until calibration data proves they are useful on the actual codebase.

Build a calibration corpus from known-clean and known-blocked pull requests. Code Mower turns that into a value report with useful-rate, false positives, cost, latency, and lane recommendations.

Once uploads include PR-linked builder metadata and adjudicated reviewer runs for the same work type, CodeMower.com's Lane Intelligence view pairs the best-evidenced builder with the best-evidenced reviewer lane, one pairing per work type, scoped to your repository only. Builder and reviewer evidence stay separate: each side carries its own confidence and sample size, and pairings with too little evidence read "collect more evidence" instead of guessing a winner.

Lane Intelligence is descriptive, not automatic. There is no cross-customer or cross-repo benchmark, and enabling a lane, adding it to branch protection, or gating merges on it stays an operator decision made outside the dashboard.

Teams that want private dashboards can create a team token on CodeMower.com and upload sanitized metadata. Cross-team cohort benchmarks are roadmap work, not live product value yet. Local reports remain useful without a hosted account.

Start with the early adopter setup or review the privacy and data boundary.