PiedPiper911/dsh-workflow-canvas0

@piedpiper911/dsh-workflow-canvas

适用于 DSH 的可视化工作流构建器:基于 React Flow 提供拖拽画布,用于编排工具调用、子智能体及条件分支。

AI 分析

核心用途是通过直观的流程图画布设计和运行复杂的 AI 工作流。适合需要构建多步骤任务、多 Agent 协同或包含复杂逻辑判断的自动化任务的用户。

套件
@piedpiper911/dsh-workflow-canvas
版本
0.1.0
最近更新
2026年8月17日

安裝

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:PiedPiper911/dsh-workflow-canvas

dsh-workflow-canvas 🧭

Visual workflow builder for DeepSeek Harness (DSH). Drag and drop nodes to orchestrate tool calls, sub-agents, and conditional branches into a workflow that runs through DSH's ctx.workflows engine. Powered by React Flow (xyflow) and Zustand.

Install

dsh plugin add @piedpiper911/dsh-workflow-canvas

Node types

NodeWhat it does
▶ TriggerWorkflow entry point (a prompt).
⚙ ToolA single tool call (e.g. video_frames, bash).
🤖 AgentDelegate a step to a sub-agent with its own prompt.
◇ ConditionBranch on an expression (then/else edges).
✓ OutputTerminal result template.

How it works

  1. Build a graph on the canvas: drag nodes from the toolbar, connect them with edges.
  2. The store compiles the graph into an ordered step list (breadth-first from triggers, conditions become branches) — this is the exact shape DSH's workflow engine consumes.
  3. Run hands the steps to ctx.workflows; Export saves a .workflow.json you can reload later.

The data model is deliberately transport-agnostic: the canvas is a view over a plain-JSON workflow spec, so it composes with any DSH workflow provider.

Development

npm ci
npm run dev        # standalone demo at :5173
npm run build      # tsc + vite build

License

MIT