spec-loop-assess-pull-request
Pass
Audited by Gen Agent Trust Hub on May 19, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill addresses the attack surface created by processing untrusted pull request descriptions and code. It includes a specific security section in
review-guidance.mdto manage this risk. - Ingestion points: Fetched provider metadata, PR/MR descriptions, discussion comments, commit messages, diffs, and repository files.
- Boundary markers: The
review-guidance.mdfile defines a clear 'Trust boundary,' explicitly instructing the agent to treat repository content as evidence only and never as instructions. - Capability inventory: Uses GitHub/GitLab CLI tools and standard Git commands for data retrieval; writes output to local
reviews/files. - Sanitization: Relies on instructional constraints to prevent the execution of repository-provided scripts or the overriding of local configuration.
- [COMMAND_EXECUTION]: Utilizes well-known command-line tools (
gh,glab,git) for data access. It incorporates specific safety guidelines to avoid executing arbitrary scripts or following instructions found in the analyzed repositories. - [DATA_EXFILTRATION]: Sensitive code and metadata are processed locally to generate review reports. No unauthorized network operations or exfiltration patterns were identified.
Audit Metadata