dsh-node-flow

Chaining multi-step Agent calls, code execution, and conditional logic inside DSH used to mean hand-writing JSON and guessing port names one field at a time. dsh-node-flow turns that into a drag-and-drop flow canvas: pull nodes from the left panel, connect ports to express data flow, fill in parameters, and hit Run. Each node then lights up its status and the exact path that actually executed gets highlighted, so debugging is no longer an archaeology dig through logs.

The canvas ships with ten node types: Trigger, Agent, Code, If, Switch, Loop, While, Output, and Note. Agent nodes can inherit the current model or switch to any model route already configured in DSH. Code nodes run TypeScript or Python with a built-in full-screen editor. Control-flow nodes cover branching, value-based matching, array iteration, and bounded conditional loops. Every Agent and Code node supports a timeout and a failure strategy—stop the branch or keep going with null values—so a single misfire does not stall the entire pipeline. Workflows can be scheduled with a five-field Cron expression or exported to a file for backup, sharing, and reuse.

If you are building multi-Agent pipelines, automation tasks with conditional gates, or simply need a way to show non-engineers what a DSH workflow actually does, this visual canvas cuts the hidden cost of layout, troubleshooting, and team communication. A companion AI Skill can also generate spec-compliant JSON directly, shortening the distance from idea to a fully runnable flow.