mem-weekly
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- PROMPT_INJECTION (LOW): Indirect Prompt Injection surface detected. 1. Ingestion points: AI_MEMORY/L1_情境层/ files. 2. Boundary markers: Absent in the processing logic. 3. Capability inventory: Grep shell execution and local file write access to update L2 behavior layers. 4. Sanitization: Absent; the system does not filter or escape content from memory records before analysis.
- COMMAND_EXECUTION (SAFE): The skill utilizes the Grep utility to perform searches within specific local directories. This execution is limited to internal data management and does not involve high-risk parameters or unauthorized paths.
Audit Metadata