chroma-hybrid-search

Pass

Audited by Gen Agent Trust Hub on Aug 9, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads embedding and reranking models (multilingual-e5-small and bge-reranker-base) from Hugging Face's official repositories using the huggingface_hub library. This is a well-known service and the operation is standard for initializing local semantic search capabilities.
  • [COMMAND_EXECUTION]: Orchestrates the execution of local Python scripts within a managed virtual environment (~/.deep-memory/.venv). These scripts handle indexing of local markdown files and JSONL notes, and provide a search CLI for the agent. All operations are scoped to the user's workspace storage.
  • [INDIRECT_PROMPT_INJECTION]: As a retrieval-augmented generation (RAG) tool, the skill ingests content from local files (knowledge-base/, experience/, and cold-notes/) to provide context. It lacks explicit filtering for instructions embedded in these source files, which is a standard characteristic of search tools rather than a malicious defect.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 9, 2026, 10:08 PM
Security Audit — agent-trust-hub — chroma-hybrid-search