nemoclaw-maintainer-cross-issue-sweep

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly searches and fetches public GitHub issue bodies and comments (scripts/search-candidate-issues.sh uses gh issue view and the relationship-judgment.md lists "candidate issue ... body" and "first ~5 comments" as LLM inputs), so untrusted user-generated content from open issues is read and used to drive LLM classifications and downstream decisions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill invokes the GitHub CLI at runtime to fetch PR diffs, bodies, and issue content from the GitHub API (e.g., https://api.github.com) which are then injected verbatim into the LLM prompts, so external content from that URL directly controls agent inputs.