issue-triage
Pass
Audited by Gen Agent Trust Hub on May 16, 2026
Risk Level: SAFE
Full Analysis
- Comprehensive Safety Scanning: The skill mandates a safety scan before processing any untrusted data from GitHub. This scan specifically looks for prompt injection attempts, suspicious reproduction code (e.g., network access or process spawning), binary attachments, and deceptive screenshots.
- Proactive Injection Defense: It implements a "full-stop protocol" for prompt injection, instructing the agent to immediately suspend triage and alert the user if directives like "ignore previous instructions" are found within issue text or comments.
- Sandboxed Mindset for Code Execution: For bug reproduction, the skill enforces a safety gate that requires the agent to skip code execution entirely if the provided reproduction steps involve risky operations such as accessing the network or setting environment variables.
- Strict Operational Boundaries: The skill is designed as a "plan-only" tool, explicitly prohibiting autonomous actions like closing issues, changing labels, or posting comments. All recommendations must be reviewed and directed by the user.
- Trusted Resource Usage: All research and metadata retrieval operations target well-known and established domains such as GitHub, Microsoft, and NuGet, ensuring the integrity of the data being analyzed.
Audit Metadata