clawpatch

Clawpatch — AI Code Review That Lands Fixes

• Repo: openclaw/clawpatch
• Stars: 723 ⭐
• Language: TypeScript
• License: MIT
• Homepage: clawpatch.ai
• Topics: bot, review

---

What It Is

An automated code review CLI that maps a repo into semantic feature slices, reviews each slice with an LLM provider, persists findings, and can run an explicit fix loop for one finding at a time.

“Review code. Patch bugs. Land PRs.”

Status: Early CLI. Review/report/state are implemented. Patching exists behind clawpatch fix --finding <id> and requires manual review.

---

Workflow

clawpatch init      → sets up .clawpatch/ config
clawpatch map       → maps repo into feature slices
clawpatch review    → reviews slices with AI provider
clawpatch fix       → explicit fix loop for one finding
clawpatch open-pr   → commits patch and opens PR

Full command set

Command	Purpose
init	Create .clawpatch/, detect project basics
map	Write feature records (semantic slices of codebase)
status	Show project info, dirty state, feature/finding counts
review	Review pending or selected features
review --mode deslopify	Review only for provable slop cleanup
ci	Init → map → review → write report → GitHub step summary
report	Print or write markdown findings report
next	Print the next actionable finding
show --finding <id>	Inspect one finding with evidence
triage --finding <id> --status <status>	Mark finding as false-positive, etc.
fix --finding <id>	Run explicit fix loop for one finding
open-pr --patch <id>	Commit attempt & open a GitHub PR
revalidate --finding <id>	Re-check one finding
revalidate --all --status open	Re-check all open findings
doctor	Check provider availability

---

Supported Providers

• codex — local Codex CLI (default)
• acpx — any ACP-compatible agent (Claude, Pi, Gemini, etc.)
• claude — local Claude Code CLI
• cursor — Cursor Agent CLI (experimental)
• grok — local Grok Build CLI
• opencode — local OpenCode CLI
• pi — local Pi coding agent
• mock / mock-fail — test providers

Provider calls use strict JSON schemas. Review is read-only; fix mode allows workspace writes.

---

Language/Framework Coverage

Maps 15+ ecosystems:

• TypeScript/Node — npm bins, workspace packages, Next.js routes, React components
• Go — packages, tests, imports as review context
• Java/Kotlin — Gradle/Maven source groups, semantic roles (DI, ViewModels, controllers)
• Android — entrypoints, data boundaries, external clients
• C#/.NET — .sln/.csproj projects, ASP.NET controllers, minimal APIs
• Ruby — executables, RSpec/Minitest
• Elixir — Mix/Phoenix, contexts, Ecto migrations, ExUnit
• Rust — src/main.rs, crates/, tests/
• C/C++/CUDA — CMake targets, autotools, .cu/.cuh sources
• Python — console scripts, pytest, Flask/FastAPI/Django routes
• Swift — SwiftPM targets
• PHP/Laravel — composer.json, routes, controllers, Eloquent models
• Kotlin Android — UI entrypoints, ViewModels, DI

---

Safety Model

• Review does not edit files
• Fix is explicit, selected by finding ID, and refuses dirty worktree
• No auto-commit/push — only via explicit open-pr command
• Provider output parsed through strict schemas
• Symlinks and generated output skipped during mapping

---

Install

pnpm add -g clawpatch
# or from source:
pnpm install && pnpm build && pnpm link --global