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

What to consider before installing: - Understand the privilege: always: true forces this policy into every agent run. If you don't want a behavior-altering policy active globally, do not install or ask the publisher to remove the always flag. - Test in a sandbox: run this skill in an isolated agent or staging environment and exercise interactions with other installed skills to catch formatting breaks, conflicts, or unexpected behavior. - Confirm owner/trust: the skill has no homepage and an unknown source owner. Prefer skills from known authors or request provenance/commit history before deployment. - Ask for limits: request that the skill be made user-invocable or togglable, or that its scope be constrained (e.g., only apply in sessions explicitly opting in), rather than always-on. - Audit subjective triggers: ask the author to clarify trigger definitions (what constitutes a 'strategic pivot'), and to provide examples and automated tests so you can verify behavior. - Verify exemptions: the SKILL.md claims a 'SYSTEM EXEMPTION' for raw code/JSON/CSV — ensure the platform reliably enforces that exemption for tool integrations. If the author can remove the always: true flag or provide stronger justification and clearer, testable trigger definitions, this would reduce the concern and could move the assessment toward benign.