Declarative task-DAG orchestration for DSH: validates dependency graphs, runs topological task layers in parallel through workflow-backed subagents, and propagates failures deterministically.
Install
# from npm (prebuilt)
dsh plugin --profile web add dsh-tool-orchestrate
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:apheli0os/deepseek-harness-orchestrate
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
English | 中文
Declarative task-DAG orchestration for DeepSeek Harness. This optional Cordis plugin registers a model-facing orchestrate_tasks tool, validates a bounded directed acyclic task graph before anything starts, and executes it through the existing ctx.workflowEngine.
This is an independent community plugin and is not an official DeepSeek package.
The package-owned worker script executes deterministic topological layers. Model-authored ids, prompts, dependency ids, schemas, provider names, model names, and upstream results are passed only as JSON data; none is interpolated into JavaScript source.
Install
dsh plugin --profile web add dsh-tool-orchestrate
Replace web with the profile you want to extend. The package declares a DSH bundle manifest, so installation adds its tool-orchestrate Cordis row automatically. This project distributes prebuilt releases through npm; installing directly from the GitHub repository is not supported.
The plugin expects a DeepSeek Harness deployment that already provides:
ctx.toolsctx.systemPromptctx.workflowEngine, normally supplied by@deepseek-ai/dsh-workflow-worker-thread- a subagent provider selected by the workflow engine
Manual Cordis configuration
Normally no manual configuration is needed after dsh plugin add. For a custom composition, add:
- id: tool-orchestrate
name: 'dsh-tool-orchestrate'
The standard DSH base bundle already loads workflow-worker-thread. A custom composition must provide ctx.workflowEngine; if it disables the workflow engine, re-enable it explicitly in a later patch layer.
Model input
orchestrate_tasks accepts:
meta: normalizednameanddescription;tasks: a non-empty array of tasks with lowercase kebab-caseid,title, and normalizedprompt;- optional
dependsOn,provider,model, and object-rootedoutputSchemaper task.
Each completed task returns its child's text or structured value. A failed child becomes { status: 'failed', error: 'subagent_failed' }; 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, or failed status.
API key ownership
This plugin does not own or store an API key. Child tasks use the host deployment's existing LLM route and credential source: for example DEEPSEEK_API_KEY in the process environment, a host credential provider, or another configured route. Subagents inherit the same credential configuration as the parent composition unless a task selects a different provider/model route.
Do not place API keys in prompts, cordis.yml, source code, or GitHub Actions logs. Prefer environment variables or your deployment's credential mechanism.
Config
| Key | Default | Meaning |
|---|---|---|
toolName |
orchestrate_tasks |
Model-facing tool name. |
maxTasks |
64 |
Maximum tasks per call. |
maxDependencies |
16 |
Maximum direct dependencies per task. |
maxPromptChars |
32768 |
Maximum normalized characters per task prompt. |
maxResultChars |
50000 |
Rendered canonical-JSON ceiling. |
All numeric limits must be positive safe integers; toolName must be a non-empty normalized string.
Development
pnpm install
pnpm run typecheck
pnpm run check
Publishing is performed from a clean checkout through the manually triggered Publish to npm GitHub Actions workflow. Maintainers must configure its npm environment and NPM_TOKEN secret first.
Links
More in this category
Q00/ouroboros#integrations/dsh-plugin★ 6128
Config-only bundle that mounts Ouroboros through the DSH MCP client, exposing 36 interview, Seed, execution, evaluation, and evolution workflow tools in DSH.
loopx-project/loopx#dsh-loopx-plugin★ 6087
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
chuspeeism/dashi-taskboard#deepseek-harness★ 3253
Embeds the active installed Codex Taskboard runtime in the DeepSeek Harness sidebar, using its launcher runtime descriptor instead of a fixed port.
NanmiCoder/dsh-agent-teams★ 1839
AgentTeams multi-agent teams.
EthanYoQ/AI-Novel-Writer#dsh-ai-novel-writer★ 1181
Installs a dedicated AI novel-writing preset and workbench: revisioned local project assets, a compact side drawer, and native approval-gated single-file changes.
tong-io/tongflow#dsh-tongflow★ 1034
TongFlow film-crew studio for image, voice, music and video production: the agent writes per-asset TongFlow workflow files (.tongflow.json) that run through TongFlow plugins, with an embedded workflow canvas, a shot/character/take project layout and a manga-drama template; sessions starting with @tongflow open the Studio view.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.