reflect
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes the conversation history, which serves as a source of untrusted data.
- Ingestion points: Analyzes the full conversation history to extract tasks, errors, and feedback as described in SKILL.md.
- Boundary markers: There are no instructions or delimiters provided to the agent to treat historical content as data rather than instructions during the analysis process.
- Capability inventory: The skill identifies and proposes actions such as creating or updating other skills and modifying the agent's memory based on the analyzed content.
- Sanitization: The instructions do not specify any validation, filtering, or escaping of the history text before it is analyzed for learning opportunities.
Audit Metadata