Senior-level code review for GitLab merge requests. Use when: reviewing MRs, providing feedback on code quality, security, performance, maintainability, or p...
Security Analysis
medium confidenceThe skill largely does what it claims (review GitLab MRs and post inline comments) but contains a few coherence issues you should be aware of before installing.
Name/description match the included code and instructions: the skill reads a GitLab token file, fetches MR diffs, filters, analyzes, and can post inline comments. However, the registry metadata did not declare the required credentials/config path (~/.openclaw/credentials/gitlab.json), which is a meaningful omission the user should know about.
Runtime instructions explicitly tell the agent to read credentials from ~/.openclaw/credentials/gitlab.json, fetch diffs from the MR host, and write temporary JSON to /tmp for posting comments. The code follows these instructions, but there is a functional mismatch: the token-scope check uses the host from the credentials file (creds['host']) while subsequent API calls use the host parsed from the MR URL. That can produce incorrect 'can_write' results when the token's host differs from the MR host. The instructions also rely on a credentials file that the registry metadata did not declare.
No install spec (instruction-only with associated scripts). Nothing is downloaded or written during install; the code files are included in the skill bundle. Risk from install mechanism itself is low.
The skill requires access to a GitLab personal access token stored in ~/.openclaw/credentials/gitlab.json (token and host). This is proportionate to its purpose. The registry did not declare this config path as required, and the credential is a powerful secret (api scope enables comment posting) — the user should ensure the file is stored securely and only contains the intended token.
The skill does not request permanent 'always' installation and does not attempt to modify other skills or system-wide settings. It writes a temporary comments JSON to /tmp only when instructed to post comments (per SKILL.md).
Guidance
This skill appears to implement a legitimate GitLab MR reviewer, but review the following before installing/using it: - Credentials file: The skill reads ~/.openclaw/credentials/gitlab.json (token + host + ignore_patterns). The registry metadata did not list any required config paths — verify you are comfortable storing your GitLab token at that path and that it is the correct token for the MR host. - Token-host mismatch: The CLI's 'check-token' step verifies the token against the host stored in the credentials file, but later API calls use the host parsed from the MR URL. If you store a token for gitlab.com but review an MR on gitlab.example.com (or vice versa), the 'can_write' result may be wrong. Consider updating the credential host to match the MR host or adjust the code. - Token scope: The token may need 'api' scope to post comments (read_api is insufficient). Only grant the minimum scope you need and rotate tokens if shared. - Review included scripts: The bundle contains Python code that makes network requests and reads your home config. Inspect the credential file contents and the scripts locally if you have concerns. The code does not appear to exfiltrate data to unexpected endpoints, but it will send diffs and comment payloads to the GitLab host derived from the MR URL. - Test first: Run in a safe environment with a test token and test MR to confirm behavior (especially token check vs MR host) before pointing it at production repositories or using a high-privilege token. If you can get the publisher to update the registry metadata to declare the required credentials/config path and fix the token-host check, that would increase confidence.
Latest Release
v1.0.2
- Overhauled workflow to use new modular scripts: gitlab_client.py (API access), ignore_matcher.py (file filtering), and post_comments.py (posting review feedback). - Adds comprehensive, step-by-step review process: credential check, diff fetch, ignore filtering, diff analysis, structured summary grouping, and conditional comment posting. - Comment format, severity criteria, and review rules are now externalized in references/review-guidelines.md for consistency. - Stronger engineering constraints: analyzes only diff changes, enforces strict ignore patterns, no generic praise, no speculative feedback. - Removed old monolithic scripts (gitlab_code_review.py, diff_parser.py, security_scanner.py) in favor of single-responsibility modules. - Now supports granular token capability detection and graceful error handling for API scope and diff-posting issues.
More by @Neuyazvimyi
Published by @Neuyazvimyi on ClawHub
