apheli0os/deepseek-harness-orchestrate
Declarative task-DAG orchestration for DSH: validates dependency graphs, runs topological task layers in parallel through workflow-backed subagents, and propagates failures deterministically.
dsh-tool-orchestrate is declarative task-DAG orchestration for DeepSeek Harness: it registers a model-facing orchestrate_tasks tool that validates a bounded directed acyclic task graph before anything starts and executes it through the existing ctx.workflowEngine. The model-facing input takes meta plus a non-empty tasks array (lowercase kebab-case id, title, normalized prompt) with optional dependsOn, provider, model and object-rooted outputSchema per task. The package-owned worker script executes deterministic topological layers: a completed task returns its child's text or structured value, a failed child becomes {status:'failed', error:'subagent_failed'}, and a task blocked by failed or skipped direct dependencies becomes skipped with deterministic blockedBy ids. Results keep the original task-array order with aggregate counts and a top-level completed/partial/failed status. Model-authored ids, prompts, schemas, provider/model names and upstream results pass only as JSON data — nothing is interpolated into JavaScript source. The plugin owns no API key: child tasks use the host deployment's existing LLM route and credential source.
Install
dsh plugin --profile web add dsh-tool-orchestratenpm dsh-tool-orchestrate 0.1.0 verified 2026-09-02 (repository field → github.com/apheli0os/deepseek-harness-orchestrate; README EN primary with zh edition). Install: dsh plugin --profile web add dsh-tool-orchestrate (replace web with the profile to extend). The package declares a DSH bundle manifest, so installation adds its tool-orchestrate Cordis row automatically. Prebuilt releases are distributed through npm — installing directly from the GitHub repository is not supported. Expects a deployment already providing ctx.tools, ctx.systemPrompt, ctx.workflowEngine (normally from @deepseek-ai/dsh-workflow-worker-thread) and a subagent provider selected by the workflow engine; the standard DSH base bundle already loads workflow-worker-thread.
Compatibility
DSH profile with ctx.workflowEngine (base bundle provides it); limits: maxTasks 64, maxDependencies 16, maxPromptChars 32768, maxResultChars 50000.
Details
- Repo: apheli0os/deepseek-harness-orchestrate
- Category: Workflow & Automation
- Stars: 1
- Version: npm dsh-tool-orchestrate 0.1.0
- Last push: 2026-08-15
- First seen: 2026-08-15
Recent updates
orchestrate_tasks tool; bounded DAG validation; deterministic topological layers via ctx.workflowEngine; failed/skipped semantics with blockedBy; JSON-data-only execution; no API-key ownership.
FAQ
- What does the model give it?
- A meta name/description and a non-empty tasks array — each task has a lowercase kebab-case id, title and normalized prompt, plus optional dependsOn, provider, model and outputSchema.
- How are failures represented?
- A failed child becomes {status:'failed', error:'subagent_failed'}; tasks blocked by failed or skipped direct dependencies are skipped with deterministic blockedBy ids; results keep task order with a top-level completed/partial/failed status.
- Does it need its own API key?
- No — child tasks use the host deployment's existing LLM route and credential source (for example DEEPSEEK_API_KEY or a host credential provider).
Alternatives
Saktawdi/dsh-ha-orchestrator · jiruidai/dsh-meta-orchestrator · leemancheung/dsh-task-dag