knotline

<div align="center">

<img src="docs/assets/knotline-banner.svg" alt="运筹 Knotline — plan, execute, and review real agent work on one operating map" width="100%" />

<br/><br/>

    

English · 简体中文

<br/>

Knotline (运筹) is a project operating map for the DeepSeek Harness sidebar.<br/> Requests, agents, execution, review, and delivery live on one canvas — a line is a command: it runs real agent work.

</div>

<img src="docs/assets/divider.svg" width="100%" alt="" />

One line runs one real execution

<div align="center"> <img src="docs/assets/demo-workflow.svg" width="92%" alt="A request connects to an agent, is classified, executes for real, and the answer grows on the map" /> </div>

Drag a Request onto the canvas and connect it to an Agent. Classification derives an Answer for questions, Review Feedback then a Plan for complex work, or a live Task Bench for Debug — then a real, resumable DSH conversation does the work. There is no fake execution: status only advances through structured lifecycle tools.

<img src="docs/assets/divider.svg" width="100%" alt="" />

Everything a chat gives you, on the map

<table> <tr> <td width="50%" valign="top"> <img src="docs/assets/demo-transcript.svg" width="100%" alt="Live transcript" /> </td> <td width="50%" valign="top"> <img src="docs/assets/demo-governance.svg" width="100%" alt="Governance flow" /> </td> </tr> <tr> <td valign="top">

Chat-grade visibility. Running Task Benches embed a live transcript (you / agent / tools); finished work carries the agent's full reply, delivery summary, and validation evidence — nothing is lost.

</td> <td valign="top">

Governed by design. Execution flows through a pre-review artifact and an independent reviewer — self-approval is rejected by the backend. Approval pools let a trusted agent execute only approved plans.

</td> </tr> </table>

<img src="docs/assets/divider.svg" width="100%" alt="" />

Six root nodes; everything else grows by itself

<div align="center"> <img src="docs/assets/demo-nodes.svg" width="100%" alt="Request, Agent, Skill, Backlog, Approval Pool, Scheduled Trigger" /> </div>

• Agent + Agent forms a Team that keeps both conversations and holds an internal discussion before assignments;
• Backlog pools queue work that idle agents pull automatically;
• Scheduled Triggers run on a Host-owned timer that survives restarts;
• Skills dragged onto an Agent or Team bind capabilities for every later run.

Reports read like posts

Click a Work Report, Answer, or Plan to open a fullscreen detail page: the producing agent leads, the body renders Markdown (tables included), and a status rail shows whether that agent is busy and what is queued. Select any passage to raise a floating annotate button; comments are relayed to the producing agent, whose status reply types out beneath your comment.

Quick start

npm install
npm run build
npx @deepseek-ai/dsh@0.1.0-rc.6 plugin --profile web add (Resolve-Path .).Path
npx @deepseek-ai/dsh@0.1.0-rc.6 web

Open the URL printed by DSH and pick 运筹 / Knotline in the sidebar: choose a workspace on the fullscreen picker and hit "Enter Knotline" at the bottom right. A three-step guide greets first-time users.

Verify

npm run check     # lint + types + 55 tests + both bundles
npm run pack:check

See the Chinese [product requirements document](docs/PRD.md), [docs/architecture.md](docs/architecture.md) for the runtime path, and [docs/development.md](docs/development.md) for local development.

<img src="docs/assets/divider.svg" width="100%" alt="" />

<div align="center"> <img src="docs/assets/knotline-wordmark.svg" width="300" alt="运筹 Knotline" /> </div>

Knotline is licensed under the [Apache License 2.0](LICENSE). It descends from the Dashi Taskboard codebase; [PROVENANCE.md](PROVENANCE.md) records what was carried forward, [NOTICE](NOTICE) carries the required attribution, and bundled third-party code is listed in [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md). Knotline is an independent community project, not affiliated with, sponsored by, or endorsed by DeepSeek or the upstream Dashi Taskboard authors. Product names are used only to describe interoperability.