agent-semantic-memory
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill explicitly recognizes that data retrieved from persistent memory stores acts as an external untrusted input. It incorporates specific guardrails (referencing OWASP ASI06) requiring that retrieved memories be treated as hypotheses and validated against the live codebase before any action is taken to prevent context poisoning.
- [CREDENTIALS_UNSAFE]: To prevent data exposure, the skill mandates a 'Zero Secret Ingestion' policy. It instructs the use of regex and entropy-based secret scanners on all payloads before they are persisted to the semantic store, specifically prohibiting the storage of credentials, PII, or private keys.
- [COMMAND_EXECUTION]: While the skill includes powerful tools such as
execute_commandin its allowed list, its instructions are focused on using these tools for verifying the current state of the codebase against stored memories rather than performing unauthorized system operations.
Audit Metadata