dsh-state-graph
Run the following command in DeepSeek Harness:
dsh plugin install zerosloney/dsh-state-graph
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install zerosloney/dsh-state-graph in DeepSeek Harness to install; the source is available at https://github.com/zerosloney/dsh-state-graph .
About this plugin
dsh's ReAct loop is inherently linear—think, act, repeat—yet real-world multi-step workflows demand conditional routing, retry loops, and parallel fan-out that prompt engineering alone can barely approximate. dsh-state-graph ports the StateGraph directed-graph orchestration engine into dsh, letting you describe arbitrary workflow topologies with declarative nodes, static edges, and conditional edges: multi-branch routing, retry with circuit-breaking, and parallel Fan-out / Fan-in are all handled by the engine rather than fragile prompt tricks.
The headline is that the graph itself becomes the runtime. Each node simply returns a partial state patch; the engine takes care of immutable merging, iteration circuit-breaking (25 rounds by default to prevent infinite loops), cooperative cancellation via AbortSignal, and a full bidirectional event stream on the graph/* bus. A conditional edge that returns a string array triggers automatic parallel fan-out, with all branch patches merged at a join node. For governance, approval-gate nodes request one-time authorization before execution; subagent nodes delegate a step to a real dsh subagent—its own session, tools, and model routing—mapping structured output back into the parent state. The bundled Graph Studio visual designer (drag-and-drop SVG canvas plus one-sentence natural-language graph generation) lets even non-graph-theory teammates build, debug, and iterate on workflows quickly.
Built for teams orchestrating multi-step agent workflows in dsh: quality-gate loops like write-code, run-tests, fix-on-failure with a retry cap; parallel fan-out of subtasks that rejoin; or embedding subagent delegation inside a graph instead of calling it ad hoc. As long as your profile already loads dsh-base, adding this bundle is all you need—no extra injection dependencies.
Use Cases
- Write code, run tests, fix on failure, retry up to 3 times
- Fan out parallel subtasks and join results before the next stage
- Insert approval gates so critical steps require human sign-off before proceeding
Best For
- Teams orchestrating multi-step agent workflows in dsh
- Developers building automation pipelines with branching, retries, and parallelism
- Engineers who prefer visual design over DSL for agent graph orchestration
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.