jmb-review-pr
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes multiple Git and GitHub CLI (
gh) commands to retrieve diffs, commit logs, and PR information. It also runs a local bash script,pending-review.sh, which is included in the skill package to manage the state of PR reviews usingjqfiltering. All commands are standard for repository management workflows.- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external, untrusted sources including Pull Request titles, bodies, and Issue comments, which are then passed into the prompts of AI subagents. - Ingestion points: Untrusted data enters the agent context via
gh pr view,gh issue view --comments, andgit show(reading branch-specific files and commit messages). - Boundary markers: The skill does not explicitly instruct the agent to use delimiters (such as XML tags) or specific boundary markers to isolate untrusted external content from its internal review instructions.
- Capability inventory: The skill has read access to the entire repository and the ability to post comments to GitHub Pull Requests.
- Sanitization: There is no defined process for sanitizing or escaping the fetched text before it is interpolated into the prompts for the 'sonnet' model.
Audit Metadata