grill-me

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). SKILL.md’s “Source Resolution” instructs the agent to read an outsider-authored GitHub issue body or issue comment (via gh issue view ... --json body / gh api .../issues/<N>/comments ... .body) and then grill it, meaning that fetched free-form text can be ingested into the LLM context at runtime.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs at runtime to fetch GitHub issue content using the GitHub CLI (e.g., "gh api repos/{owner}/{repo}/issues//comments --jq '.[] | select(.body | startswith("")) | .body'"), which would load remote text that directly becomes the document the agent uses to generate its prompts and questions.