DSH Workflow — Visual workflows for DeepSeek Harness

[English](./README.en.md) · [简体中文](./README.md) ### DSH-native workflow automation platform Compose triggers, sub-agents, code, conditional branches, loops, and scheduled tasks with drag-and-drop nodes, then run the complete workflow directly in DSH. [![License: MIT](https://img.shields.io/badge/License-MIT-22c55e.svg)](./LICENSE) ![TypeScript](https://img.shields.io/badge/TypeScript-5.5-3178c6.svg) ![React Flow](https://img.shields.io/badge/React%20Flow-12-7c3aed.svg)

![Node palette and workflow canvas](./docs/images/node-palette.png)

_{Add nodes from the left palette and arrange the complete workflow on the canvas}

## Overview DSH Workflow is a native workflow automation platform for [DeepSeek Harness (DSH)](https://github.com/deepseek-ai/deepseek-harness). It combines agent calls, TypeScript / Python code, HTTP, triggers, conditions, loops, and persistent execution into a workflow system that runs directly inside DSH. The repository is named `dsh-workflow`. For compatibility with existing DSH installations, persisted data, and runtime API routes, the npm package, Cordis bundle ID, and host API paths remain `dsh-node-flow`. This compatibility identifier does not change the project’s independent repository identity. ## Features - **Visual orchestration** — drag nodes, connect handles, and inspect data and execution paths directly on the canvas. - **Sub-agent calls** — agent nodes can inherit the current model or select a model route configured in DSH. - **Code nodes** — run TypeScript or Python with a full-screen code editor. - **Flow control** — built-in If, Switch, Loop, and While nodes cover branching and iteration. - **Execution feedback** — nodes display their run state, and the actual path taken is highlighted. - **Failure policies** — Code and Agent nodes support timeouts and either stopping the branch or continuing with an empty value. - **Scheduled tasks** — save workflows with five-field cron expressions, inspect them in the task board, run them immediately, or cancel them. - **Import and export** — save workflows as files for backup, sharing, and reuse. - **Built-in documentation** — the help panel explains node fields, connection rules, result semantics, and workflow examples. ## Screenshots ### Node properties Select a node to configure its label, model, prompt, timeout, and failure policy in the right-hand inspector. ![Node properties inspector](./docs/images/node-inspector.png) ### Scheduled tasks The scheduled-task board displays cron plans, previous trigger times, and latest run results, with actions to inspect, run once, or cancel a task. ![Scheduled task board](./docs/images/scheduled-tasks.png) ## Quick start ### Installation The package has not been published to npm. Install it directly from the public GitHub repository: ```sh dsh plugin --profile web add git+https://github.com/luxi233/dsh-workflow.git ``` When running DSH from a DeepSeek Harness source checkout, use: ```sh pnpm dsh plugin --profile web add git+https://github.com/luxi233/dsh-workflow.git ``` Restart the Web UI after installation: ```sh dsh web ``` For a source checkout, the corresponding command is `pnpm dsh web`. Open **Node Mode** from the DSH sidebar to start building workflows. ### Basic usage 1. Click **Add Node**, then choose a node or a built-in example from the left panel. 2. Connect node handles and configure parameters in the right-hand inspector. 3. Click **▶ Run** and watch the node states and highlighted execution path. 4. Use **Export** to save the workflow as a file, or create a cron schedule from **Scheduled Tasks**. ## Companion AI Skill Install the companion [`dsh-node-flow-skill`](https://github.com/luxi233/dsh-node-flow-skill) to help AI agents in DSH generate JSON workflows that can be imported directly into this plugin. The Skill documents the current workflow schema, node fields, connection handles, loop rules, and complete examples, reducing missing fields and invalid handle names. After installation, prompt the AI with: ```text Use dsh-node-flow-skill to generate an importable dsh-node-flow JSON workflow that … ``` Skill repository: ## Node types | Node | Purpose | | --- | --- | | **Trigger** | Starts the workflow and passes its content downstream. | | **Schedule** | Runs the complete workflow using a five-field cron expression. | | **Agent** | Calls a DSH sub-agent using the inherited model or a selected model route. | | **Code** | Runs TypeScript or Python on the host, reads upstream input, and returns a result. | | **If** | Routes the workflow to `true` or `false` based on a truthy condition. | | **Switch** | Matches a value against cases and falls back to `default`. | | **Loop** | Iterates over an array or repeats N times, optionally collecting each result. | | **While** | Repeats while a condition remains true, guarded by `maxIters`. | | **Output** | Renders the upstream value using a `{{result}}` template. | | **Note** | Adds a canvas annotation and is not executed. | ## Code node API TypeScript programs can use the global `dsh` object: ```ts const response = await dsh.fetch({ url: "https://example.com/api", method: "POST", headers: { "Content-Type": "application/json" }, body: { message: "hello" }, }) const previous = await dsh.input(null) // Read the upstream node output const now = await dsh.now(null) // { iso, ts, local } return response.json ``` Python nodes read upstream input from an environment variable and use standard output as the node result: ```python import os previous = os.environ.get("DSH_INPUT", "") print(previous) ``` > DSH Code Runtime bindings require a lossless JSON argument. Use `dsh.input(null)` and `dsh.now(null)` rather than omitting the argument. ## Runtime requirements The plugin expects the DSH host to provide these services: - `ctx.webServer` - `ctx.codeRuntime` for Code nodes - `ctx.agents` and `ctx.subagents` for Agent nodes - `ctx.llm` for the model-route list React, ReactDOM, and `@deepseek-ai/cordis` are provided by DSH as peer dependencies. ## Local development ```sh npm install npm run dev # Start the standalone Vite preview npm run build # TypeScript checks + client bundle + demo build npm test # Run tests ``` ## License Released under the [MIT License](./LICENSE).