review-loop

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

• Third-party content exposure detected (high risk: 0.85). The skill reads the PR body via gh pr view --json body -q .body (outsider-authored free text from the PR author) and then includes it in the runtime prompt sent to the LLM via the trigger API.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill makes runtime curl requests to https://command-center.aigora.ai (e.g., /api/orchestration/trigger and /api/orchestration/status) and parses the returned JSON (state.lastMergeResult.blockingIssues, summaries, etc.) which directly instructs the agent what fixes to perform, so this external endpoint controls the agent's runtime instructions.