Build visual skill workflows
Place skills on a canvas, connect dependencies, and reshape repeatable routines without reconstructing them from memory.
Observe every run.
Turn Claude and Codex skills in your repository into visible, repeatable, controllable development workflows.
Why Circuit
Agent work should be observable while it happens.
Long prompts and terminal streams hide the shape of a process. Circuit keeps skill order, run state, failures, and recovery paths visible in one desktop workspace.
Track run state and logs
See what is running, what completed, where a failure happened, and whether an active run should be cancelled.
Handle loops deliberately
Keep cyclic workflows visible and require confirmation before running paths that can repeat indefinitely.
Use Claude and Codex together
Read both skill directories from the same local repository and choose the execution capability that fits each step.
Visual workflows
Change the routine by changing the graph.
Circuit turns skill order and dependencies into something you can inspect and adjust. Planning, implementation, review, and commit steps can move as the team's process changes.
planning -> implementation -> review -> commit
planning -> implementation -> commit -> review
planning -> check-token -> implementation -> review
Run visibility
The canvas and run log stay in sync.
Active workflows can be cancelled, completed steps remain visible, and failed runs are easier to inspect because the stopping point is still on the graph.
Track execution state without reading every line first.
Circuit keeps the current node, completed nodes, and run output close together so failures are easier to locate and recover from.
Loop handling
Repeated work stays explicit.
Some routines need to repeat after review, validation, or failure. Circuit shows cyclic paths before execution and asks for confirmation when a workflow can run indefinitely.
Claude and Codex
Use multiple agent skill systems in one workflow.
Circuit reads Claude and Codex skill directories from the local project and treats them as execution capabilities that belong in the same workflow, not as competing islands.