agentmemory-architecture
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill instructs the agent to read authentication secrets from local files located at
~/.agentmemory/.envand~/.agentmemory/secretto interact with the local REST API. - [INDIRECT_PROMPT_INJECTION]: The architecture described involves capturing untrusted data that could contain malicious instructions.
- Ingestion points: Captured prompts and tool inputs/outputs are indexed and stored in
SKILL.md. - Boundary markers: No specific delimiters or "ignore instructions" markers are mentioned for processing ingested data.
- Capability inventory: The system provides capabilities to store and search data through a REST API as detailed in
TROUBLESHOOTING.md. - Sanitization: There is no documentation regarding the sanitization or validation of the stored memory content before retrieval.
Audit Metadata