qdrant-vector-search
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The troubleshooting and installation sections provide standard system commands for managing local database instances, such as
docker run,sudo systemctl start docker, andlsof -i :6333. These are documented for diagnostic and setup purposes. - [INDIRECT_PROMPT_INJECTION]: The skill demonstrates RAG (Retrieval-Augmented Generation) patterns where data retrieved from the vector database is interpolated directly into prompts (e.g.,
f"Context: {context}\n\nQuestion: {query}"). This represents a vulnerability surface for indirect prompt injection if the stored document payloads contain malicious instructions designed to influence the LLM's behavior at runtime. - Ingestion points: Data enters the context via the
client.searchandretrievefunctions inSKILL.md. - Boundary markers: The code examples show basic text interpolation without explicit boundary markers or XML-style delimiters to separate context from instructions.
- Capability inventory: The skill includes file writing (via Docker volume mounts), network operations (API calls to Qdrant), and subprocess execution (standard developer workflow commands).
- Sanitization: No explicit sanitization or filtering of retrieved payload text is demonstrated in the RAG code snippets.
Audit Metadata