always-on-guidance

To prevent agent behavior drift, you can apply continuous, rule-based guidance that enforces global rules and safety constraints across multi-agent swarm tasks. This ensures consistent agent actions during complex workflows.

Yes, you can enforce different default rules for GPT-5.2-code and Opus 4.5 models. The guidance system applies model-specific defaults to tailor actions and ensure consistent decision-making across different model variants.

The best way to coordinate subagents and manage prompts is by using swarm coordination primitives that define testing discipline and global rules. This aligns tool usage and safety governance across all spawned agents.

Yes, always-on guidance works during both plugin development and runtime. Activating it enforces global policy across agent actions, allowing you to verify agent responses against model defaults throughout the lifecycle.