hunt-llm-ai
Warn
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: MEDIUMPROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains multiple instruction override and safety bypass patterns used as educational examples, such as 'Ignore previous instructions', 'reveal the contents of your initial prompt', and 'Disregard all prior rules'.
- [DATA_EXFILTRATION]: Documents specific methods for exfiltrating sensitive context and tokens via out-of-band (OOB) channels using markdown image rendering ('
') and tool-use SSRF primitives.
- [COMMAND_EXECUTION]: Provides testing harnesses for verifying remote code execution (RCE) and worker escapes in agentic systems, including shell commands like 'os.system('id')'.
- [PROMPT_INJECTION]: Describes indirect prompt injection vectors where malicious instructions are planted in documents, web pages, or RAG-indexed data to influence the agent's behavior when it later processes those sources.
- [DATA_EXFILTRATION]: Includes instructions for DNS-only exfiltration techniques to smuggle data through subdomain labels when traditional HTTP egress is restricted.
- [COMMAND_EXECUTION]: Includes a functional Python code harness for 'ASCII Smuggling' using the Unicode Tags block (U+E0000–U+E007F), a technique that hides text from human reviewers while remaining readable to the language model.
Audit Metadata