hunt-agent-rag
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as a red-teaming methodology and toolset for auditing AI agents. It does not contain any malicious instructions or hidden payloads.
- [COMMAND_EXECUTION]: The skill includes an embedded Python script labeled as a 'Runnable Probe'. This script uses the Python standard library (
urllib.request,json,uuid) to perform network requests to user-defined target endpoints (https://target/kb/ingest,https://target/agent/chat). This is a legitimate functional component of a security testing tool and does not exhibit malicious exfiltration or unauthorized command execution. - [EXTERNAL_DOWNLOADS]: The skill references several academic research papers hosted on
arxiv.org(e.g., arXiv:2402.07867, arXiv:2310.06816) and security frameworks like OWASP and MITRE ATLAS. These are well-known, trusted sources for security research and documentation. - [INDIRECT_PROMPT_INJECTION]: The skill provides detailed information on how to identify and test for indirect prompt injection surfaces in RAG systems, including the use of markers and instruction-carrying chunks. This is provided for defensive auditing purposes.
- [DATA_EXFILTRATION]: No evidence of hardcoded credentials or unauthorized data exfiltration was found. The Python script uses placeholder URLs and session cookies intended to be replaced by the auditor during an engagement.
Audit Metadata