MrMaii / knotline

Listed

把 Agent 从对话框搬到一张图上——连线即执行 / Orchestrate AI agents by drawing lines

mainSkill View source

Installation

npx -y @deepseek-ai/dsh plugin --profile web add github:MrMaii/knotline

This installation command is an unverified starting point generated from the GitHub repository address.

README

Maintainer-authored documentation snapshot.

View on GitHub ↗
Commit 1775b05Synced Aug 18, 2026
运筹 Knotline — plan, execute, and review real agent work on one operating map



License Node DSH npm version npm downloads

English · 简体中文


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

One line runs one real execution

A request connects to an agent, is classified, executes for real, and the answer grows on the map

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.

Everything a chat gives you, on the map

Live transcript Governance flow

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.

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.

Six root nodes; everything else grows by itself

Request, Agent, Skill, Backlog, Approval Pool, Scheduled Trigger
  • 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/architecture.md for the runtime path, and docs/development.md for local development.

运筹 Knotline

Knotline is licensed under the Apache License 2.0. It descends from the Dashi Taskboard codebase; PROVENANCE.md records what was carried forward, NOTICE carries the required attribution, and bundled third-party code is listed in 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.

Project files and signals

Shown items are public repository signals detected in the directory snapshot.

TestsDetected
Security policyDetected
Contributing guideDetected
DocumentationDetected

Repository information

Language
JavaScript
License
Apache-2.0
Latest release
v0.1.0
Last updated
Aug 18, 2026, 3:31 AM

Install deliberately

Review source code, permissions, lifecycle hooks, dependencies and network access. Test untrusted plugins in an isolated environment.