langchain-embeddings-search
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external data sources (PDFs, Markdown documents) for Retrieval-Augmented Generation (RAG). This process introduces a surface for indirect prompt injection, where instructions embedded in the processed data could potentially influence the agent's behavior. * Ingestion points: Document loading and splitting logic in SKILL.md and references/hybrid-search.md. * Boundary markers: The provided code snippets focus on retrieval and do not include prompt-level delimiters or warnings to ignore embedded instructions. * Capability inventory: The skill is configured to use Bash(python:*), Write, Edit, and Grep tools. * Sanitization: No sanitization or content filtering logic is implemented in the document processing examples.
- [DYNAMIC_EXECUTION]: The documentation advises using the pickle module for persisting and loading BM25 indices to improve performance. * Evidence: references/hybrid-search.md states: "For large corpora (> 100K docs), persist with pickle and reload on startup." * Risk: Deserializing data using pickle can be unsafe if the source file is compromised, potentially leading to arbitrary code execution.
Audit Metadata