← Back to Skills Marketplace
PR Review
by
Misha Kolesnik
· GitHub ↗
· v0.2.1
· MIT-0
121
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install review-github-pr
Description
GitHub PR code review - fetches the diff, runs automated checks, launches 3 parallel review agents (correctness, convention compliance, efficiency) to analyz...
Usage Guidance
This skill's behavior (cloning/checking out PRs, running local validation, launching three parallel model-based review agents, and preparing GitHub reviews) is reasonable for a PR-reviewer, but the packaging is inconsistent. Before installing or running: (1) verify you have the GitHub 'gh' CLI and git on PATH and that 'gh' is authenticated (it will need permission to read the repo and, if you post reviews, to write), (2) confirm whether your platform allows the skill to invoke models — the SKILL.md expects sub-agents but the skill metadata sets disable-model-invocation: true, so ask the publisher to clarify/fix this, (3) treat evaluation fixture scripts as test-only — they create a mock 'gh' but are not a runtime install; don't run untrusted setup scripts unless you inspect them, (4) test the skill in a sandboxed environment (non-production account) to confirm actual behavior and that it will not auto-post reviews, and (5) ask the publisher to update metadata to declare required binaries and any needed environment variables (e.g., GITHUB_TOKEN or GH auth) and to remove the contradictory disable-model-invocation flag. If the publisher cannot explain or fix these mismatches, avoid granting the skill access to real repositories or credentials.
Capability Analysis
Type: OpenClaw Skill
Name: review-github-pr
Version: 0.2.1
The skill automates GitHub PR reviews by cloning repositories and executing commands defined in a 'CLAUDE.md' file. A significant security vulnerability exists in the execution flow described in SKILL.md: the agent checks out the PR branch before Phase 1 (Automated Checks), which allows a malicious PR to modify the CLAUDE.md file and achieve Remote Code Execution (RCE) by specifying arbitrary shell commands. While the instructions include explicit security warnings and use delimiters to handle untrusted PR content, the logic of executing project-defined scripts from an unverified branch poses a high risk of exploitation.
Capability Tags
Capability Assessment
Purpose & Capability
The SKILL.md implements a GitHub PR review flow that expects a local git repo, the GitHub 'gh' CLI, and the ability to post reviews. However, the declared requirements list no binaries, no config paths, and no primary credential. At minimum 'gh' (and git) and user authentication for GitHub are required for the described behavior. The presence of evaluation fixtures that create a mock gh CLI suggests tests assume the operator will add a mock to PATH — but that is not the same as the skill declaring its real runtime dependencies.
Instruction Scope
The instructions instruct the agent to clone/check out PRs, run local validation commands (from CLAUDE.md), read every changed file, and launch three parallel review sub-agents with model: 'opus' while wrapping PR content as untrusted. Those operations are in-scope for a PR reviewer, but the SKILL.md also requires model-based sub-agent invocation while the skill metadata sets disable-model-invocation: true (contradiction). The skill reads repository files and may clone into /tmp; it explicitly forbids executing arbitrary commands found in PR content, which is good, but the mismatch between declared capabilities and the runtime instructions is a practical problem.
Install Mechanism
There is no install specification (instruction-only), which minimizes automatic disk writes. The package does include evaluation fixture scripts and mock 'gh' creation scripts used for local testing; these are test artifacts that a user would need to run manually to reproduce evals. No external downloads or installers are embedded in the skill.
Credentials
The skill will interact with GitHub via the 'gh' CLI and may need to post reviews, but it declares no required environment variables (no GITHUB_TOKEN/GH_TOKEN). That omission is disproportionate: either the operator must already have GH auth configured for 'gh', or the skill should declare and request credentials. The SKILL.md's requirement that reviews be posted only after explicit user confirmation mitigates auto-post risk, but the credential omission remains an inconsistency.
Persistence & Privilege
The skill does not request persistent 'always' privileges (always: false), which is appropriate. However, the metadata sets disable-model-invocation: true while the instructions explicitly rely on launching multiple model-driven sub-agents (Agent tool with model: 'opus'). This mismatch affects whether the skill can perform its stated behavior and is a governance/privilege inconsistency that should be resolved before trusting the skill.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install review-github-pr - After installation, invoke the skill by name or use
/review-github-pr - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.2.1
Updated review-github-pr from 0.2.0 to 0.2.1.
Changes:
- modified `SKILL.md`
v0.2.0
Initial publish of review-github-pr 0.2.0.
Changes:
- added `LICENSE.txt`
- added `SKILL.md`
- added `evals/evals.json`
- added `evals/fixtures/clean/setup.sh`
- added `evals/fixtures/correctness-convention/setup.sh`
- added `evals/fixtures/efficiency-design/setup.sh`
Metadata
Frequently Asked Questions
What is PR Review?
GitHub PR code review - fetches the diff, runs automated checks, launches 3 parallel review agents (correctness, convention compliance, efficiency) to analyz... It is an AI Agent Skill for Claude Code / OpenClaw, with 121 downloads so far.
How do I install PR Review?
Run "/install review-github-pr" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is PR Review free?
Yes, PR Review is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does PR Review support?
PR Review is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created PR Review?
It is built and maintained by Misha Kolesnik (@tenequm); the current version is v0.2.1.
More Skills