add-agent-playground

You can test agent configuration changes without disrupting live classroom workflows by creating draft agent groups that isolate configuration edits from production users, then applying approved changes via a side-by-side diff. This prevents active student and instructor workflows from being affected during agent persona iteration.

Yes, you can toggle between Claude and Codex providers mid-session while testing draft agent settings. The playground supports provider and skill testing, allowing you to browse a skill library with compatibility badges and evaluate different provider configurations before applying them to live agent groups.

Real-time test chat for draft agent groups works via a web UI that streams responses using Server-Sent Events. It routes interactions through the same container pipeline as live channels, ensuring your testing environment accurately reflects how draft agents will perform once deployed to production.

The best way to compare draft and live agent configurations is using side-by-side diff functionality. This feature highlights differences between your draft edits and the live agent group settings, allowing you to review persona, skill, and provider changes before applying them with a single action.

You need an academic NanoClaw environment to set up an agent config testing playground. The setup installs a lazy-start web-based channel adapter secured with magic-link authentication, designed specifically for instructors and administrators iterating on agent personas and provider configurations.

Testing agent personas directly on live agents risks disrupting students because configuration changes instantly alter active classroom workflows. Modifying provider settings or personas in production affects real-time interactions, making safe iteration difficult without an isolated draft environment for testing first.