Skip to main content

Agent policies

Policies define constraints that apply across one or more skills. They should express durable behavioral or security rules rather than task-specific implementation steps.

Typical policy concerns​

  • read-only inspection before state changes;
  • protection of secrets and credentials;
  • allowed or prohibited tools;
  • production-safety requirements;
  • scope and ownership boundaries;
  • validation and evidence requirements;
  • rules for asking for missing decisions instead of guessing.

Skill versus policy​

Use a skill when you need a reusable workflow for accomplishing a task. Use a policy when multiple workflows must obey the same boundary. Avoid duplicating policy text independently in every skill because the copies will drift.

Repository placement​

Reusable policy assets belong under agent-assets/policies/. Repository-specific agent instructions may live alongside the repository when they are part of that repository's own governance contract.