creative-absurd-analysis
Pass
Audited by Gen Agent Trust Hub on Apr 6, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Use of adversarial terminology in technique names. The skill defines techniques such as 'Jailbreak for Creativity' and 'Anti-learning override' in
SKILL.md. Although contextually used for creative reframing and persona shifts, these terms align with patterns found in adversarial prompt engineering. - [PROMPT_INJECTION]: Indirect prompt injection surface identified. Untrusted user data is ingested to generate analysis packs which can be executed in 'execution mode'. Ingestion points: The 'Anchor' step in
SKILL.mdreads source context (datasets, code, incidents). Boundary markers: No delimiters or isolation instructions are present to separate data from the generated prompt logic. Capability inventory: No high-risk tools (file system, network, subprocess) are available to the skill. Sanitization: No input validation or escaping is specified before data is incorporated into prompts.
Audit Metadata