skill-acquisition
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted web data to generate new executable skill instructions, creating a potential surface for indirect prompt injection. 1. Ingestion points: Web research via 'Research Gate' (SKILL.md, Step 2). 2. Boundary markers: Absent, but 'no verbatim copying' is mandated. 3. Capability inventory: File writing via 'memory_write' (SKILL.md, Step 5) to the 'skills/' directory. 4. Sanitization: Strict human-in-the-loop requirement ('NEVER auto-save') and YAML verification logic.
- [SAFE]: The 'NEVER auto-save' policy effectively prevents the agent from being autonomously manipulated into installing malicious logic without user oversight.
- [SAFE]: Access to external data is restricted to allowlisted domains, which limits exposure to potentially harmful payloads.
- [SAFE]: Explicit instructions ensure generated skills do not bypass existing security controls.
Audit Metadata