Read the dashboard#

Prerequisites: the image-only tutorial. Outcome: know where each output comes from and where to send feedback.

Follow the stages#

Panel	What it produces	What to inspect
Task Decomposer	Ordered subtasks	Object choice, sequence, missing steps
Skill Generator	Prioritized controller recipes and keypoint definitions	Constraints, axes, grasp timing
Parameter Setter	Numeric offsets and units	Distances, force magnitudes, angle units
Vision Module	Image keypoint locations and annotations	Visible surfaces and exact label placement
Robot Script	Generated Python	Geometry, recipe order, execution prerequisites

Enter the task in the task input.
Wait until the current operation finishes.
Select stage tabs to inspect the outputs.
Send feedback to the stage responsible for the issue.
Inspect downstream changes before generating or running a script.

Messages and console events help explain progress, but the latest visible artifact is not necessarily current. Requery and refinement invalidate grounding/script readiness while work is in progress or after failure.

The script panel#

The script panel contains Generate Script and Save Workflow, followed by execution controls. The library’s Prepare view also provides Save Checkpoint; its Run view provides the working-copy controls.

Saving, generating, and running are different actions. A checkpoint preserves agent context; a workflow export preserves artifacts; Run starts supervised hardware processes after checks.

Two servers, separate state#

main.py defaults to port 8000. library.py defaults to port 8001 and has Prepare, Library, and Run navigation. They do not share an in-memory workflow. For example, a task prepared in main.py does not automatically appear in the library’s Prepare workspace.

The applications coordinate managed robot ownership through a lock in the same checkout. See execution for process ownership and controls.

Verify: you can identify the stage that owns each part of a recipe. Continue with feedback or library workflows.