DeepSeek Harness Plugin

ccch1mneyyy/dsh-TUI

Stars ★ 829 Category UI Enhancements Added 2026-08-14

Claude Code-style full-screen terminal UI: pixel-whale header, live status line, and streaming thought expansion.

Install

# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)

dsh plugin --profile web add github:ccch1mneyyy/dsh-TUI

GitHub-sourced plugins run build scripts on your machine at install time. Only install sources you trust, and pin a commit (github:owner/repo#sha).

README

dsh-TUI

dsh-TUI is an interactive terminal front door for DeepSeek Harness. It is mounted as a Cordis plugin and provides a Claude Code-style conversation, tool, session, and fullscreen terminal experience while continuing to use the official DSH agent, model, tool, session, and persistence services.

The project does not patch DeepSeek Harness core. Installing the plugin enables the interface, and removing it leaves no core modifications behind.

Status: public beta. It is suitable for daily use and extension work. Read Architecture and limitations before relying on its permission model or terminal-specific behavior.

Highlights

  • Terminal-native interaction: streaming Markdown, structured tool cards, command and file completion, @ file references (complete anywhere in the message; sending attaches the file content or directory listing), history search, message selection, inline or alternate-screen rendering, and /lang zh/en UI language switching.
  • Visible agent state: live activity, segmented context usage, TPS, cache hit rate, reasoning effort, input/output tokens, and Git/session metadata.
  • Complete session workflow: /resume, /new, /compact, /export, model switching, and double-Esc rewind through a session fork.
  • Official DSH integrations: agent presets, skills, MCP, goals, todos, subagents, and ask_user_question are connected through existing services and registries.
  • Designed for long sessions: event-driven projection, differential output, message virtualization, replay coalescing, and bounded caches prevent render cost and memory from growing without limit.

Preview

Live activity, goal/todo state, and context metrics:

Quick Start

Prerequisites: an interactive terminal TTY, the official dsh CLI, and pnpm 10+. Model requests also require DEEPSEEK_API_KEY.

# 1. Install the DeepSeek Harness CLI
npm install -g @deepseek-ai/dsh

# 2. Add the dsh-TUI profile plugin
dsh plugin --profile cc-tui add dsh-cc-tui

# 3. Start it
dsh --profile cc-tui

The repository's sh install.sh wraps step 2 and checks the required commands. Windows users can also launch with dsh-cc.cmd; passing --resume restores the most recently selected session.

See Getting started for profile composition, source builds, and troubleshooting.

Inside the TUI, /update updates the installed dsh-cc-tui package and automatically restarts into the current session.

The TUI also checks npm for updates in the background after startup. The check never blocks the first frame and silently ignores offline or registry errors.

Documentation

Topic Contents
Getting started Prerequisites, installation, startup, profile lifecycle, source development
Configuration Cordis overrides, fields, agent presets, MCP, environment variables
Themes Built-in themes, background detection, custom JSON themes, validation
Interaction and commands Keyboard, mouse, questionnaires, slash commands, session workflows
Architecture and limitations Runtime path, rendering, persistence, security boundary, known limitations
Contributing Contribution workflow, repository map, build artifacts, verification matrix, change rules

The complete bilingual index is docs/README.md.

How It Works

dsh profile
  -> dsh-base
  -> dsh-TUI Cordis patch
  -> agent preset + DSH services
  -> session/event
  -> Channel projection
  -> React components
  -> ported Ink/Yoga renderer
  -> terminal

The TUI owns interaction and presentation only. The session log remains the conversation source of truth, while model calls, tool execution, fork/resume, compaction, and persistence remain owned by DSH services. See the architecture guide for module boundaries and performance details.

Development

CI uses Node 24 and pnpm 11. The package supports Node ^22.19 || >=24.

pnpm install --frozen-lockfile
pnpm build
pnpm smoke

pnpm build compiles src/ into the checked-in lib/types/ output. Source changes must include regenerated artifacts, and rendering, questionnaire, or tool-card changes require the relevant regression scripts.

Permissions and Security Boundary

dsh-TUI does not implement a separate sandbox. It uses the filesystem, shell, sandbox, and approval policies of the active DSH profile. The supplied profile uses workspace confinement and approvals by default on non-Windows platforms. Windows currently has no corresponding sandbox backend, so the composition falls back to danger-full-access without approval prompts. Inspect the profile before starting it around sensitive credentials or an untrusted repository.

See Permissions and security boundary for details.

Featured by DeepSeek Harness

The DeepSeek Harness official WeChat account featured this plugin among its early user-built extensions. View the feature screenshot.

License

MIT

Content from the project README on GitHub ↗

Links

More in this category

View the whole category →