reflect
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies and extracts patterns from conversation history, which consists of untrusted data provided by the user.
- Ingestion points: The skill analyzes user input messages (corrections, praise, edge cases) within the conversation stream to identify potential learning targets.
- Boundary markers: The extracted patterns and proposed changes are presented in a dedicated visual block clearly separated from other output.
- Capability inventory: The skill facilitates updates to the agent's runtime memory and suggests modifications to the governing source files via Pull Requests.
- Sanitization: The skill implements a robust safeguard by requiring explicit user confirmation ("Apply changes? [Y/n/edit]") before persisting any identified patterns, ensuring that malicious or incorrect interpretations can be intercepted by a human operator.
Audit Metadata