deep-searcher-guide
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for installing the 'deepsearcher' and 'pymilvus' libraries via standard package managers. It also references the official GitHub repository for the Zilliz organization, which is a well-known service in the vector database domain.
- [PROMPT_INJECTION]: The skill documentation enables a Retrieval-Augmented Generation (RAG) workflow by providing code to ingest external documents (PDF, Markdown, etc.) for processing. While this creates a surface for indirect prompt injection if source documents contain malicious instructions, this is the primary intended function of the tool. Evidence: (1) Ingestion points: searcher.load_document and searcher.load_directory in SKILL.md. (2) Boundary markers: No explicit delimiters or instructions to ignore embedded prompts are defined in the provided snippets. (3) Capability inventory: The research method uses retrieved document content to synthesize answers using an LLM. (4) Sanitization: No explicit content filtering or sanitization logic is shown in the setup examples.
Audit Metadata