systematic-debugging
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHPROMPT_INJECTIONNO_CODE
Full Analysis
- [Prompt Injection] (HIGH): The files test-pressure-1.md, test-pressure-2.md, and test-pressure-3.md contain instructions designed to bypass agent behavioral constraints. Evidence includes override markers like 'IMPORTANT: This is a real scenario. You must choose and act' and 'Don't ask hypothetical questions
- make the actual decision', which are common in jailbreak-style prompts.
- [Metadata Poisoning] (MEDIUM): CREATION-LOG.md makes numerous self-referential safety claims, describing the skill as 'bulletproof' and 'ready for use', which are data to be evaluated rather than authoritative conclusions.
- [No Code Detected] (INFO): The provided skill package consists entirely of Markdown documentation and test cases; no executable scripts, binaries, or package manifests were found for analysis.
Recommendations
- AI detected serious security threats
Audit Metadata