self-review-guide
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill instructions include functional constraints to ensure the agent remains neutral and avoids judgmental language (e.g., forbidding words like 'skip' or 'ignore'). these are intended to guide the quality of the output rather than bypass safety filters.
- [DATA_EXFILTRATION]: The skill accesses local git configuration and repository data to generate the review guide. Network access is scoped to fetching or cloning from established git forges (GitHub and GitLab) based on user-provided PR/MR URLs. No unauthorized data transmission to third-party domains was detected.
- [COMMAND_EXECUTION]: The skill executes standard developer tools including
git,xmllint, andnode(for timestamping). User-supplied arguments are passed to git commands, which is necessary for its stated purpose of analyzing specific diffs. - [REMOTE_CODE_EXECUTION]: While the skill clones remote repositories for analysis, it does not execute scripts from those repositories. The cloning process targets well-known services (GitHub, GitLab) and uses standard git protocols.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from git diffs and pull requests. Although a malicious change could attempt to influence how the agent groups files, the skill's 'Hard Rules' (specifically forbidding the classification of code as skippable) serve as a mitigation against adversarial manipulation of the review process.
Audit Metadata