project-focus-first
Pass
Audited by Gen Agent Trust Hub on Aug 4, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is purely instructional and governs interaction flow without using executable code or external resources.
- [PROMPT_INJECTION]: The skill contains rules for agent persona and behavior, but these are for organizational purposes and explicitly yield to safety and system guidelines. No bypass attempts were found. Additionally, while the skill handles untrusted user input describing projects, it acts as a restrictive gate to ensure alignment. Ingestion points: User-supplied project descriptions in SKILL.md trigger conditions; Boundary markers: Explicit categorical responses are required before the agent proceeds; Capability inventory: No tools, subprocess calls, or file-system writes are present; Sanitization: The restrictive gate logic inherently prevents arbitrary user content from being treated as executable instructions.
- [DATA_EXFILTRATION]: No network activity or sensitive data access was identified.
- [REMOTE_CODE_EXECUTION]: The skill does not download or execute any external scripts or packages.
Audit Metadata