Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed reasoning-mode helper with no code execution or data access, though its automatic activation and corrupted thresholds may surprise users.
Install this only if you want the agent to automatically decide when to use deeper reasoning. Expect possible extra latency or token use, and consider editing or avoiding it if you need precise control over reasoning mode or dislike visible reasoning icons in responses.
The skill is explicitly designed to run on every user message as a preprocessing step, which creates an overly broad activation scope. Even though it does not appear overtly malicious, always-on behavioral modification can silently change response generation across all contexts, increase unnecessary internal processing, and interfere with user intent or system controls.
The automatic activation logic is ambiguous and insufficiently constrained: several thresholds are corrupted or unspecified, and the skill instructs the agent to 'Do not ask. Just activate.' This can cause inconsistent or silent escalation of reasoning behavior, making model behavior less predictable and potentially overriding user preferences for speed, cost, or minimal processing.
No suspicious patterns detected.
