AI Control Protocol (Anti-Sycophancy & Zero-BS)

Before enabling this skill, consider the following: - Understand always:true: this will force the skill to run on every conversation. If you want this behavior only in specific contexts, prefer a user-invocable skill or require explicit opt-in. - Test in a sandbox: deploy the skill in a non-production environment to observe how its mandatory labels, deconstruction boxes, and defense panels interact with your tooling (APIs, JSON outputs, code-generation flows). The exemption for raw code/JSON may not catch every case. - Watch for speculative content: the rules ask the agent to infer motives and urge 'extreme' alternatives to consensus. That can produce sensitive, inaccurate, or risky recommendations — verify outputs manually and add guardrails if needed. - Check interactions with other skills: because it runs persistently, it may conflict with other skill behaviors or system-level policies. Confirm ordering/precedence or disable always:true if you need predictable composition. - Verify provenance: the skill metadata lists a GitHub homepage but the registry owner is unknown. Review the upstream repository (code, issues, maintainer identity) to ensure there are no hidden instructions or additional runtime files. - If in doubt, decline or limit scope: prefer enabling the skill with always:false or user-invocable only, and require human review for strategic recommendations that trigger psychological inferences or extreme alternative proposals.