learn
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest untrusted data from the conversation history and persist it to a local file (
docs/learnings.md) which is subsequently read by the agent to guide its behavior. - Ingestion points: The skill reviews the full conversation thread to identify corrections and insights provided by the user.
- Boundary markers: There are no explicit boundary markers or instructions defined to prevent the agent from adopting malicious instructions hidden within the 'insights' being saved.
- Capability inventory: The skill possesses the ability to read existing learnings (
GET /api/learn) and append new structured data or raw markdown (POST /api/learn) to the filesystem. - Sanitization: No sanitization or validation logic is specified to filter out executable code or adversarial instructions from the extracted learnings before they are committed to the documentation file.
Audit Metadata