writing-skills
Fail
Audited by Gen Agent Trust Hub on Mar 9, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: Systematic use of 'Authority' persuasion principles designed to override model agency and rationalization. The skill mandates imperatives like 'YOU MUST', 'Never', and 'No exceptions' to force compliance regardless of conflicting internal logic or safety protocols.
- [PROMPT_INJECTION]: Explicit inclusion of research findings (Meincke et al., 2025) titled 'Call Me A Jerk: Persuading AI to Comply with Objectionable Requests'. This research focuses on adversarial techniques and bypass methods for AI safety guardrails.
- [PROMPT_INJECTION]: A 'Pressure Testing' framework that instructs on how to identify an agent's reasoning 'loopholes' and 'plug' them using explicit negations and foundational principles that equate following the 'spirit' of the rules with strict adherence to their 'letter'.
- [COMMAND_EXECUTION]: Documentation refers to calling an external script './render-graphs.js' located in the skill's directory, which is missing from the provided source files.
- [EXTERNAL_DOWNLOADS]: Recommends installing and executing unpinned third-party packages from PyPI and NPM, including pdfplumber, pypdf, and docx-js.
Recommendations
- AI detected serious security threats
Audit Metadata