agent-memory-systems
Pass
Audited by Gen Agent Trust Hub on Jun 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides implementation patterns for Retrieval-Augmented Generation (RAG) which involves processing potentially untrusted data from external documents and conversation histories.
- Ingestion points: Document content in 'CHUNKING STRATEGIES' and conversation histories in 'BACKGROUND MEMORY FORMATION' (SKILL.md).
- Boundary markers: Snippets do not explicitly show context delimiters or 'ignore' instructions for the LLM.
- Capability inventory: Database read and write operations across various vector stores and PostgreSQL.
- Sanitization: No explicit sanitization or validation logic is demonstrated for retrieved content before it is interpolated into the model's context window.
- [EXTERNAL_DOWNLOADS]: The skill references several well-known industry-standard libraries and services for vector search and machine learning.
- Dependencies: Mentions Python packages including 'langmem', 'langchain', 'pinecone', 'qdrant-client', and 'chromadb'.
- Well-Known Services: References Pinecone, Qdrant, Weaviate, and Anthropic's contextual chunking methodology.
Audit Metadata