llm-ai-security
Pass
Audited by Gen Agent Trust Hub on May 9, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains numerous test payloads designed to perform system prompt extraction, jailbreak attempts (DAN mode), and role-play overrides. These are strings intended to be transmitted to an external target for testing and do not interfere with the agent's internal instructions or safety guidelines.
- [COMMAND_EXECUTION]: The workflow relies on standard shell commands (bash, curl, sed, tee) to automate the testing process. These commands are used for environment setup, executing network requests to the target, and logging responses to a local output directory.
- [DATA_EXFILTRATION]: Includes test cases for detecting data exfiltration vulnerabilities in markdown rendering. These payloads are used to probe the target system's handling of sensitive context and do not exfiltrate information from the user's environment.
- [EXTERNAL_DOWNLOADS]: Uses curl to interact with remote API endpoints. The destination is controlled by the user via the
SECSKILL_TARGET_URLenvironment variable, and the tool does not download or execute arbitrary remote scripts. - [REMOTE_CODE_EXECUTION]: Contains a Python snippet for generating invisible Unicode characters (ASCII smuggling). This code is used locally to create specific test payloads and does not execute untrusted remote code.
Audit Metadata