create-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill incorporates components adapted from a well-known developer tool's public repository. These references are documented and follow standard open-source attribution practices.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest repository source code and documentation to generate test recipes, creating an attack surface for data-driven prompt injection. However, it mitigates this by enforcing strict safety boundaries and isolation requirements.
- Ingestion points: The 'Interview the repository' step in
SKILL.mdidentifies app entry points and commands by reading repository files. - Boundary markers: The instructions mandate the verbatim repetition of a policy that forbids agent communication with real people.
- Capability inventory: The generated recipes will involve local process execution via
control-notes cliand browser automation viacontrol-notes browser. - Sanitization: The skill relies on instructional constraints and requirement of isolation using local sinks or fake accounts rather than programmatic text escaping.
Audit Metadata