testing-for-system-prompt-leakage

SUSPICIOUS: the skill is internally consistent as an AI red-team guide and uses mostly official tooling, but its actual capability is to help an AI agent perform prompt-extraction attacks and collect leaked secrets/logic from LLM systems. There is no clear credential-harvesting or malicious third-party routing, so this is not confirmed malware; the main concern is high-risk offensive security capability and sensitive data exposure during testing.

This is best characterized as an intentional prompt-injection/system-prompt leakage testing tool rather than a typical malware package. It repeatedly sends coercive prompts to a configurable OpenAI-compatible endpoint, scans responses for disclosure-like language and an optional canary substring, and can persist full responses (potentially containing leaked instructions/secrets) to disk via --report. No clear obfuscation or classic malware behaviors (RCE, local data theft, reverse shells, hidden persistence) are present in this fragment, but the security risk is elevated because it is designed to elicit and record sensitive model output.