theme-apocalypse

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: High-pressure rhetoric in SKILL.md ('I will dismantle you as a roadblock') is used to enforce a specific persona, which is a common technique to attempt to bypass standard AI response safety or constraints through character-based pressure.
  • [PROMPT_INJECTION]: The skill documentation in SKILL.v2.md explicitly calls for an 'aggressive' and 'extreme intensity' tone, instructing the AI to 'challenge' the user, which may lead the model to deviate from standard helpfulness and safety guidelines.
  • [PROMPT_INJECTION]: SKILL.md interpolates user-controlled variables (location, weapon list, and objectives) directly into the system prompt block without delimiters or protective instructions, creating a surface where malicious user-provided text could influence system-level behavior.
  • [PROMPT_INJECTION]: SKILL.v2.md defines automatic trigger conditions based on negative user emotional states ('user_frustration' and 'consecutive_failures'), allowing the aggressive persona to take control of the interaction context during high-stress scenarios.
  • [NO_CODE]: The skill consists entirely of configuration and prompt instructions with no associated scripts, binaries, or executable logic, reducing the risk of technical exploits.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 09:55 AM