Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture to extend the behavior of agents. When developing applications that involve multiple steps, external API calls, and complex logic, simple scripts can often be difficult to maintain. DSH Workflow is a native workflow automation platform designed to provide a visual way to organize triggers, HTTP requests, code execution, conditional branching, and scheduled tasks.

Core Features

Visual Orchestration and Node Testing

The plugin provides a canvas-based interface, supporting intuitive definition of execution paths by dragging nodes and connecting ports. Each executable node (except persistent Wait) supports standalone testing. During testing, the system generates a JSON field tree from the test input or an upstream sample, allowing nested field expressions to be inserted by clicking or dragging.

Code Execution and HTTP Requests

Code nodes support TypeScript and Python and provide a full-screen editor. HTTP nodes support custom methods, request headers, request bodies, response parsing, timeouts, and templates. Both run in the host process, and code nodes can access DSH runtime capabilities through the global dsh object.

Triggering and Scheduling

Workflows support multiple startup methods. Webhook triggers use independent tokens to receive external requests and support HMAC-SHA256 signature verification. Scheduled tasks use 5-field Cron expressions. Application event triggers support continuous polling of services such as GitHub and Slack, and new events can trigger independent run instances.

Reliability and Fault Tolerance

The system includes a built-in failure retry mechanism, supporting 0–10 retries with exponential backoff. Executable nodes can be routed to an error branch after retries are exhausted, and error events are redacted. The server persists checkpoints at node boundaries and supports recovery from a failure point. The execution budget can limit the maximum runtime and maximum number of steps; when exceeded, a safe checkpoint is retained.

Security and State Management

Connection management supports Bearer Token, API Key, Basic Auth, and OAuth2, with sensitive information encrypted using AES-256-GCM. Project variables and Secrets are centrally stored in the host variable store and are automatically redacted in run logs. Data Store provides an encrypted lightweight state store.

Installation and Enablement

This project has not been published to npm. Please install it directly from the GitHub repository.

dsh plugin --profile web add git+https://github.com/luxi233/dsh-workflow.git

After installation, restart the Web UI (dsh web). Once inside DSH, open Node Mode from the sidebar to start using it.

Typical Usage

  1. Create a workflow: Create a new workflow from the Node Mode home page, or filter from the folders on the left.
  2. Add nodes: Open the node library from the left tool rail, search for the needed node, and add it (for example, Agent, Code, HTTP).
  3. Connect nodes: Drag node ports to create connections, or click the + in the middle of a connection line to insert a step.
  4. Configure parameters: Edit parameters in the Configuration or Output tab of the right-side inspector.
  5. Test a node: Fill in JSON test input in the inspector, insert expressions through field mapping, and click Test This Node.
  6. Publish the flow: Click Publish to update the snapshot used by actual runs and Cron scheduling.
  7. Manage credentials: Create credentials such as Bearer Token or OAuth2 in Connections on the home page, then select them in a node.
  8. Configure policies: Set concurrency mode, maximum runtime, and maximum execution steps in More Actions → Run Policy.
  9. Error handling: Select On Failure → Route to Error Branch for a node and connect the red error port.
  10. Version control: Click Version History to compare versions, or restore and publish.

Notes

  • Compatibility marker: The npm package name, Cordis bundle ID, and host API path still retain dsh-node-flow. This is only a compatibility marker and does not affect independent repository maintenance.
  • Secret security: Secrets are not returned to the browser and are automatically redacted in run records, errors, and API responses.
  • Replay mechanism: Replays always use the workflow snapshot saved at that time, not the current draft.
  • Execution budget: The budget does not accumulate across runs. A resumed run reuses the budget stored in the snapshot.
  • Folder ownership: Folders only manage control-plane ownership, such as list filtering, and do not modify the content of drafts, published snapshots, run records, or version history.

Summary

DSH Workflow provides a complete visual workflow solution that combines Agent calls, code execution, HTTP requests, and conditional logic into directly runnable flows. Through node testing, error branches, checkpoint recovery, and version history, it provides a foundation for building reliable automation tasks.

  • GitHub: https://github.com/luxi233/dsh-workflow
  • Skill Directory: https://www.skillhub.cn/plugins/luxi233/dsh-workflow