Peter Steinberger, creator of OpenClaw, recently released Clawpatch, described as an automated code review tool that "goes beyond traditional linters" by mapping a repository into semantic work units, sending those units to a provider for review, and persisting findings for later audit and follow-up.
The tool focuses on feature-level context rather than file-by-file scanning. Its feature records can include entrypoints, owned files, nearby tests, and trust boundaries, and each finding carries severity, confidence, evidence, and a recommendation. The project also groups reviews around semantic units such as routes, commands, packages, CLI scripts, and tests.
Clawpatch’s workflow is built around an explicit patch loop. The documented process includes clawpatch init, clawpatch map, clawpatch review --limit 10, clawpatch report, clawpatch fix --finding abc123, and clawpatch revalidate --finding abc123. Validation for patch attempts can include format, type, lint, and test checks, with results recorded in the patch history.
The project particularly emphasizes guardrails: according to the documentation, it uses a clean-worktree check by default for fixes, avoids implicit commits and destructive Git operations, and keeps an audit trail in .clawpatch/. It also validates provider responses against strict JSON schemas before use.
Clawpatch’s default provider shells out to the local Codex CLI. The source also notes that initialization auto-detects project types including Node.js, TypeScript, Next.js, Python, Flask, FastAPI, Go, Rust/Cargo, and SwiftPM, and that settings are written to .clawpatch/config.json. A full configuration reference is linked from the project documentation at GitHub.
Source: Clawpatch
