AI Agent Hub
Back to plugins
🖥️

dsh-acp

Client Updated 2026.09.15

Run the following command in DeepSeek Harness:

dsh plugin install dangpangch/dsh-acp

Paste the following prompt into your AI chat to install this plugin:

Install this plugin in DeepSeek Harness by running dsh plugin install dangpangch/dsh-acp; full source code at https://github.com/dangpangch/dsh-acp

About this plugin

dsh ships a functional but limited ACP layer, which leaves Zed users with only a narrow slice of the Harness experience. dsh-acp plugs that gap by registering a full ACP v1 agent-server into Zed, so DeepSeek Harness agents become first-class citizens inside the editor's Agent Panel.

The plugin delivers a rich interactive surface: streamed text and reasoning tokens, live tool cards that render command output in a terminal-style foldable view, slash commands alongside user-defined agent skills, plan and todo updates, full session lifecycle (create, list, load, resume, close, delete) with durable history, one-shot permission prompts, model and thought-level selectors, and elicitation forms. Every tool call executes inside the dsh sandbox; the client only renders captured results and never runs agent commands itself. The preset system lets a blank-session thread re-compose its tool set, prompt sections, and skills with a single switch, locking the composition at the first turn.

Ideal for developers who work in Zed with DeepSeek models and want an in-editor agent workflow rather than a separate terminal window, and for advanced dsh users who need the full ACP v1 semantics of streaming deltas, persistent sessions, and a discoverable skill catalog.

Use Cases

  • Create threads in Zed Agent Panel and invoke DeepSeek Harness agents for coding and reasoning tasks in-editor
  • Watch streamed reasoning, tool executions, and captured output in real-time terminal-style cards
  • Extend agent capabilities via slash commands and user-defined skills, or switch preset compositions of tools and prompts

Best For

  • Developers using DeepSeek models in Zed who prefer an in-editor agent workflow over a separate terminal
  • Advanced dsh users who need full ACP v1 semantics including streaming deltas, persistent sessions, and a skill catalog
  • Security-conscious users who require all tool calls to run in a sandbox while the client only renders output