learning
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill contains no executable code, scripts, or network requests. It consists of instructional guidance and evaluation criteria for the agent's internal reasoning process.
- [INDIRECT_PROMPT_INJECTION]: The skill involves reading lessons from external sources such as 'durable notes' or a 'memory store'. The risk of indirect prompt injection is mitigated by the instruction that 'memory and heuristics never silently override current accepted authority or direct current evidence'. Evidence Chain: 1. Ingestion points: Durable notes or memory store (SKILL.md); 2. Boundary markers: Explicit instructions to treat memory as advisory and prioritize current evidence; 3. Capability inventory: None (no scripts or tool definitions); 4. Sanitization: Relies on agent reasoning and evidence verification rather than technical sanitization.
Audit Metadata