stacking-prs
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from GitHub pull requests, including titles, bodies, and commit history.
- Ingestion points: Data is ingested through
gh pr list,gh stack view, andgit logcommands. - Boundary markers: The instructions explicitly state that PR comments, titles, and bodies are to be treated as data and never as instructions. It references a dedicated document (
shared/external-data.md) outlining strict isolation rules. - Capability inventory: The skill utilizes
git,ghCLI tools, and a local Node.js helper script (stack-guard.mjs) to manage branch state and PR metadata. - Sanitization: Untrusted content is handled by writing to temporary files and passing paths to CLI tools (e.g., using
--body-file), effectively preventing shell interpolation vulnerabilities. - [SAFE]: The skill recommends the installation of the
github/gh-stackextension, which is an official tool from a well-known service (GitHub). This is a manual recommendation for the user to execute and does not involve automated remote code execution by the agent.
Audit Metadata