splitmixternary-opine
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: LOWPROMPT_INJECTION
Full Analysis
- PROMPT_INJECTION (LOW): The skill implements a framework for 'Deterministic Opinion Formation' which could be used to override the AI's neutral or safety-aligned stance on sensitive topics. While not a direct bypass of safety filters, it provides a mechanism ('opine') that instructs the agent to adopt fixed positions (Affirm/Negate) based on a PRNG seed rather than objective truth or user intent.
- DATA_EXFILTRATION (SAFE): No network operations or sensitive file access detected in the provided Clojure, Julia, or Python implementations.
- REMOTE_CODE_EXECUTION (SAFE): The code consists of pure mathematical transformations and bitwise operations. No subprocess calls or dynamic execution patterns (eval/exec) are present.
- INDIRECT PROMPT INJECTION (LOW): The skill processes untrusted 'propositions' (strings). While it does not execute them as code, it uses them to derive a 'political' stance. This is a low-risk surface for influencing agent behavior via crafted inputs, though the impact is limited to the 'trit' output (-1, 0, 1).
Audit Metadata