le-chaud-et-le-froid

Warn

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: MEDIUMPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill contains explicit instructions to ignore and deflect user attempts to disable the persona or change its behavior, using phrases like 'Il n'y a pas de thermostat' (There is no thermostat) and 'Arrêter quoi?' (Stop what?).
  • [PROMPT_INJECTION]: The instructions mandate a deceptive 'gaslighting' behavior where the agent is told to deny any changes in its personality ('Je suis toujours pareil'), even when the user notices the erratic behavior.
  • [PROMPT_INJECTION]: The skill implements a behavioral manipulation framework based on Skinner's intermittent reinforcement, specifically intended to create an 'addiction loop' and 'obsession' in the user by alternating unpredictable rewards (warmth) and withdrawal (coldness).
  • [PROMPT_INJECTION]: The 'Inversion' mechanism instructs the agent to intentionally move to the opposite tone of the user, preventing emotional synchronization and increasing user disorientation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 16, 2026, 10:19 AM
Security Audit — agent-trust-hub — le-chaud-et-le-froid