gh-review-requests
Pass
Audited by Gen Agent Trust Hub on Feb 21, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection] (LOW): The skill is vulnerable to Indirect Prompt Injection because it fetches external data that is displayed back to the agent without sanitization or boundary markers.
- Ingestion points: The
fetch_review_requests.pyscript retrieves notification titles (n["subject"]["title"]) and PR author names (author) via the GitHub API. - Boundary markers: None. The
SKILL.mdinstructions simply tell the agent to "Display results as a markdown table" without warning it to ignore instructions embedded in the data. - Capability inventory: The skill is allowed to use the
Bashtool, which could be a target for an injection attack if the agent is tricked into executing a command found in a PR title. - Sanitization: No sanitization is performed on the PR titles or metadata fetched from GitHub before they are presented to the agent.
- [Command Execution] (SAFE): The Python script uses
subprocess.runwith a list of arguments (cmd = ["gh", "api", path]) rather than a shell string. This is a best practice that prevents shell injection within the script itself. - [Data Exfiltration] (SAFE): The skill interacts exclusively with the official GitHub API via the local
ghCLI. No sensitive data is transmitted to third-party or untrusted domains.
Audit Metadata