bot-panel-review-loop
Pass
Audited by Gen Agent Trust Hub on Jun 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell scripts (
select-prs.sh,reserve.sh, andpr-actions.sh) to perform repository management and GitHub API operations. These scripts are bundled with the skill and utilize the GitHub CLI (gh) andjqfor secure data handling. Analysis of the scripts shows proper input validation and safe command construction patterns to prevent shell or SQL injection. - [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes untrusted data from Pull Request diffs, descriptions, and comments. This is inherent to its primary purpose as a code reviewer.
- Ingestion points: External data enters the context via
gh pr viewandgh pr diffcalls managed by thepr-actions.shscript. - Boundary markers: The skill instructions specify providing a clear "brief" to sub-agents and enforcing "Gather-only" constraints to prevent unauthorized actions.
- Capability inventory: Agents possess tools for executing bundled scripts, dispatching sub-agents, and invoking the
panel-reviewskill for deeper diff analysis. - Sanitization: The instructions mandate an "adversarial re-read" where the agent must actively attempt to refute its own findings before recommending fixes, particularly on sensitive surfaces like authentication, money movement, or database schemas.
Audit Metadata