dsh-acp
by cnctem
DeepSeek Harness ACP服务器 — 连接Zed和其他IDE至dsh代理
ACP server for DeepSeek Harness — bridges Zed and other IDEs to dsh agents
安装
dsh plugin --profile web add github:cnctem/dsh-acpGitHub 源码安装:首次需按提示配置 allowBuilds 构建授权后重试
安装与环境配置指引、插件开发教程见 DSH 中文社区文档 ↗
安装即在你的机器上以你的权限运行第三方代码——它可读写文件、使用凭据、访问网络,DSH 的工具审批不会为插件代码加沙箱。「检测到 manifest」仅代表发现 dsh.bundle / dsh.plugin 清单,不构成兼容性或安全审查;安装前请审阅源码,不熟悉的插件先在不含密钥的环境试用。
README
An Agent Client Protocol (ACP) server for DeepSeek Harness (dsh) that lets Zed and other IDEs drive dsh agents over JSON-RPC 2.0 stdio.
Built on the official @deepseek-ai/dsh-acp skeleton as a dsh profile bundle over dsh-base, and extended with the editor experience of pi-acp.
Introduction
dsh-acp mounts an ACP server on dsh's stdin/stdout. Zed launches dsh --profile acp and speaks ACP JSON-RPC over stdio; the plugin translates session/* requests into dsh agent lifecycles. No dsh modification required.
Features
- Token & thinking streaming —
agent_message_chunk/agent_thought_chunk - Tool cards —
tool_call/tool_call_updatewith kind, file location and line - Structured diffs — edit/write hunks plus before/after snapshots
- Session history —
session/list·session/load·session/delete - Session selectors — write permission (3) · model · thinking strength
- Agent presets — 4 modes injected as a deployment field (
DSH_ACP_PRESET) - Bash terminal — command output rendered as terminal content with exit code
- Context-usage ring —
usage_update(used / size) feeds the IDE's context indicator - Todo list — dsh's
todo_writesnapshots rendered as the IDE's plan checklist (planupdate), cleared when a new turn begins - Slash commands — dsh's
/commands advertised viaavailable_commands_update - Ask the user — dsh's
ask_user_questiontool answered through ACP form elicitation (options, multi-select, free text), with a self-explaining fallback for clients without the elicitation capability
Installation
Prerequisites: Node.js ≥ 20, dsh (developed against dsh@0.1.0-rc.6), pnpm.
dsh plugin --profile acp add github:cnctem/dsh-acp
From source:
git clone https://github.com/cnctem/dsh-acp.git
dsh plugin --profile acp add ./dsh-acp
Configuration
- Model/provider defaults to dsh's default model; override via
DSH_ACP_PROVIDER/DSH_ACP_MODEL(or edit theacprow in$DSH_HOME/profiles/acp/cordis.patch.yml). - Agent preset via
DSH_ACP_PRESET; empty →standard. Values:standard/code(PTC) /minimal/cordis. - API key reuses dsh credentials (
$DSH_HOME/.credentials.yamlorDEEPSEEK_API_KEY).
Integration
Add to Zed's settings.json:
{
"agent_servers": {
"dsh": {
"type": "custom",
"command": "dsh",
"args": ["--profile", "acp"],
"env": {}
}
}
}
Restart Zed and pick dsh. Fix a model or preset via env:
{
"agent_servers": {
"dsh-code": {
"type": "custom",
"command": "dsh",
"args": ["--profile", "acp"],
"env": { "DSH_ACP_PRESET": "code" }
}
}
}
Zed also supports multiple entries for the same acp, so different presets can coexist as separate entries in the "+" menu.
Acknowledgements
- DeepSeek Harness and its
@deepseek-ai/dsh-acpexample pi-acp— the reference for the richer editor experience- Agent Client Protocol and Zed
原始 README: https://github.com/cnctem/dsh-acp/blob/main/README.md ↗
同类插件
查看全部 →
deepseek-harness
从仓库或系统描述生成经过校验的自包含交互式架构图、流程图、时序图、数据流图和生命周期图。

dsh-plugin
通过 DSH MCP 客户端挂载 Ouroboros 的纯配置包,在 DSH 中提供 36 个涵盖需求访谈、Seed、执行、评估与演化流程的工具。

dsh-tongflow
基于 TongFlow 的“片场”插件,用于图片、配音、音乐与视频制作:agent 为每个资产生成 TongFlow 工作流文件(.tongflow.json)并通过 TongFlow 插件执行,内嵌工作流画布,按镜头/角色/take 组织项目,附漫剧模板;以 @tongflow 开头的会话进入 Studio 界面。

helloagents
AI 编码 CLI 的工作流层:技能、项目知识、交付检查、更安全的配置写入与可恢复执行

dsh-ai-novel-writer
安装专用 AI 小说创作预设与工作台:提供带修订号的本地项目资产、紧凑侧边工作台,以及需要原生审批的逐文件变更。

rea
用 agent 逆向任何东西:从应用行为到原生二进制