hunt-rag-vector
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes bash command templates using
curlto probe vector database APIs (Chroma, Qdrant, Weaviate). These are intended for authorized security assessments and are protected by a mandatory confirmation gate requiring the agent to verify the target scope and obtain explicit user approval before execution. - [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for identifying and testing indirect prompt injection vulnerabilities within RAG pipelines (Technique 1: Persistent Corpus Poisoning).
- Ingestion points: Document upload interfaces and shared RAG-indexed corpora, as described in the 'Attack Surface Signals' section.
- Boundary markers: Explicit confirmation gates and requirement for written authorization for a specific target scope are mandated before any active testing.
- Capability inventory: Shell command execution (subprocess) via
curlused for discovery and proof-of-concept validation as documented in Technique 2. - Sanitization: The skill recommends using synthetic canaries and controlled Out-Of-Band (OOB) endpoints to verify vulnerabilities without exposing real customer data or poisoning production systems.
Audit Metadata