Starting a Playground Session
To start a playground session, navigate to a scenario and select Playground. You’ll configure:- Build — Which version of your game to test
- Device — A single device to run on
- Agent Template — Which agent configuration to use
- Skills — Behavioral instructions for the agent (editable before starting)
- Knowledge — Game context (can be overridden inline)
- Version — Optionally start from a specific scenario version instead of the current draft
The Playground Loop
The core workflow is an iterative loop:1. Watch the Agent
Once the session starts, you see the live device screen streamed to your browser. The agent begins working through the scenario’s checkpoints. You can follow along with:- The live video feed
- Checkpoint progress in the sidebar
- The agent’s reasoning and actions
2. Pause and Adjust
At any point, you can pause the agent. While paused:- Completed checkpoints are locked — checkpoints the agent has already passed are frozen with their results (timestamp, screenshot, duration). You can’t edit or remove them.
- Pending checkpoints are editable — you can modify goals and instructions, reorder them, or remove them entirely.
- Add new checkpoints — insert new milestones for the agent to work toward.
- Add verifications — attach assertions to any checkpoint to check game behavior.
3. Resume
When you’re ready, click Resume to send the agent back to work with your updated checkpoints. If you made changes while paused, you’ll be asked to confirm:- Save & Resume — Apply your checkpoint edits and continue
- Discard & Resume — Throw away edits and continue with the previous checkpoints
4. Repeat
Continue the watch-pause-adjust-resume cycle until you’re satisfied with the scenario. The agent will keep working through checkpoints as you refine them.Auto-Sync
You don’t always need to pause. When you edit checkpoints while the agent is running, your changes are automatically synced to the agent after a short delay. The agent picks up the updated instructions on its next decision cycle without interruption.Saving to Scenario
When the agent finishes (or you’re happy with the results), click Save to Scenario. This:- Saves checkpoints — Writes the current checkpoint list back to the scenario
- Syncs skills — Updates, creates, or links any skills that changed during the session
- Promotes verifications — Converts any assertions you added during the session into persistent scenario verifications
- Creates a new version — Snapshots the saved state as a new scenario version (e.g., v2, v3)
- Links the run — Associates this playground run with the new version
When to Use Playground
- Building a new scenario from scratch — Start with a rough idea, watch the agent, and add checkpoints as you discover what matters
- Debugging a failing scenario — Run the failing scenario in playground, watch where it goes wrong, and adjust checkpoints on the spot
- Exploring a new build — Use playground to poke around a new version of your game and see how the agent handles changes
- Refining checkpoints — Watch the agent struggle with a vague goal, then rewrite it with better instructions while the session is still live
- Adding verifications — Watch the game behavior during a run and add assertions for things you want to check automatically in future runs
Quick Reference
| Concept | What it is |
|---|---|
| Playground session | An interactive run where you can edit checkpoints live |
| Locked checkpoint | A completed or failed checkpoint — frozen, can’t be edited |
| Pending checkpoint | A checkpoint the agent hasn’t reached yet — fully editable |
| Auto-sync | Checkpoint edits are sent to the agent automatically |
| Save to Scenario | Writes your playground changes back to the scenario and creates a new version |