gh-address-comments

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The skill explicitly instructs the agent to escalate network/auth permissions and to rerun with sandbox_permissions=require_escalated, which are instructions to override environment/sandbox constraints and escalate privileges beyond the normal PR-comment-handling scope.

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

• Third-party content exposure detected (high risk: 0.85). Outsider-authored PR conversation text (comments/reviews/review thread bodies) is fetched at runtime via gh api graphql in scripts/fetch_comments.py and then printed as JSON, which the workflow uses to summarize and select comments—so that free text can be fed into the agent/LLM context.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The prompt explicitly asks to run gh commands with "elevated network access" and to rerun auth checks with "sandbox_permissions=require_escalated", i.e. to escalate/bypass sandbox/security permissions, which pushes the agent to request elevated privileges.