Skills Plugins MCP Prompt Model 导航 博客 资讯 我的中心
开发与运行时 #agent-harness#dsh#dsh-bundle#dsh-plugin#dsh-plugin-market#dsh-plugins

better-dsh (dashr)

能胜任严肃编程任务的 dsh。(Tools x Schemas)^REPL。`skill://`,`ctx://`,`agent://`,`dvc://`,`dsh://`,IPython REPL,Context as Variables,抗压缩可回溯、完整上下文重构。`hash-edit`、`dvc://browser`、subagent as function、workflow as function。

pgmi-builds @pgmi-builds ⬇ 1 ★ 5 main

安装

dsh plugin --profile web add github:pgmi-builds/better-dsh
下载安装清单

需要可复现安装时,可在仓库后追加 #commit 固定提交。

能胜任严肃编程任务的 dsh。(Tools x Schemas)^REPL。`skill://`,`ctx://`,`agent://`,`dvc://`,`dsh://`,IPython REPL,Context as Variables,抗压缩可回溯、完整上下文重构。`hash-edit`、`dvc://browser`、subagent as function、workflow as function。

该插件未提供要点说明,请参考仓库 README。

agent-harnessdshdsh-bundledsh-plugindsh-plugin-marketdsh-plugins
  1. 安装并启动 DeepSeek Harness:npx @deepseek-ai/dsh web
  2. 在终端执行上面的安装命令(CLI 会解析插件并核验来源)
  3. 用 dsh plugins list 确认已安装,必要时重启 Harness 生效

插件以当前 dsh 进程的权限运行,安装时可能执行代码。请先通读仓库源码与许可证,确认无破坏性命令与越权访问;本站只做索引,不对第三方插件安全性作担保。

代码仓库github.com/pgmi-builds/better-dsh/tree/main/dashr
许可证MIT
主要语言main
下载量1
GitHub 星标5
最近推送2026-09-14
收录日期2026-09-19
分类开发与运行时

事实信息来自公开插件目录快照(2026-10-01),介绍文案由本站再加工。

以下为插件仓库 README 全文(原始内容,由公开目录抓取整理)。

# Better Dsh — Dashr

  [图片: dsh plugin]

  [图片: npm package]

  [图片: Repository]

  [图片: License]

**Dashr** (*/ˈdæʃ.ɚ/*) makes your native [`dsh`](https://github.com/deepseek-ai/deepseek-harness) ready for serious coding.

---

## Why

- **$(\text{Tools} \times \text{URL schemas})^{\text{REPL}}$** - a concise, cognitively frictionless setup delivering maximal capacity.
- Universal **Tools** - `read`, `write`, `grep`, `glob`.
- **URL schemas** - `skill://`, `ctx://`, `agent://`, `dvc://`, `dsh://`, an intuitive, unified interface for runtime resources.
- **Persistent REPL Kernel** - IPython (compared with the native ephemeral TS REPL).
- (20+ tools × 5 schemas) ^ REPL free composition = **∞**

- **Context as Variables** - full session transcripts are accessible as variables, even cross compactions.
- **Full Context Revive** — full-namespace `dill` snapshots save and restore kernel state across restarts.

```json
{ "cell": "
  scout = await agent('agent://ralph', 'graph-analyze this codebase; scout the breaking points to add tests')
  ui    = await read('dvc://browser/127.0.0.1:3080')
  me    = await read('ctx://transcript')",
  "description": "one cell: agents × devices × session" }
```

---

## ✨ Other Features

- **Agent swarm as resources** — sub-agents, workflows, and Ralph loops are first-class calls (`subagent`, `workflow`, `ralph`, `message`).
- **KeepRecency context compactions** - preserve recent relevancy. Resets the default 80% compaction threshold to 50% as "usable" context max.
- A simplistic global LLM-endpoint failover, absent in the native build (a safeguard for long-range, unattended tasks).
- keeps the system prompt to ~3–4K tokens.
- **`dvc://browser`** - browser use ready.
- All `host plane` wired — you do not even see any difference from your `dsh` UI; it just silently fuels up the `dsh` base.
- **iOS PWA Ready** - minor UI twists that make the webUI a decent PWA for iOS `Add to Home Screen`.

## Recommended Companion Plugin(s)

- [`dsh-better-sidebar`](https://github.com/omdsh-dev/DSH-better-sidebar)
- [`corti`](https://github.com/pgmi-builds/corti) multi-agent memory plugin

---

## ⚡ Quick Install

```bash
dsh plugin --profile web add --config.auto-install-peers=false better-dsh@latest
```

---

## 📚 Inspiration & Attribution

`better-dsh` is an open-source plugin for [DeepSeek Harness (`dsh`)](https://github.com/deepseek-ai/deepseek-harness) (MIT). It stands on the shoulders of earlier open-source work:

- **[BetterEdit](https://github.com/Rianico/dsh-better-edit)** (`dsh-better-edit`, by Rianico) — the hash-anchored `read` → `edit` → `undo_last_edit` editing model. Dashr adopts this toolchain (itself a lineage of `pi-hashline-edit` → `pi-hashline-edit-lsz` → `dsh-better-edit`) with attribution, and re-wires it natively.
- **[omp-agent](https://github.com/can1357/oh-my-pi)** (oh-my-pi, by Can Bölük) — the `dvc://` device framework and its `ast_edit` / `ast_grep` / `browser` / `lsp` devices, adopted under its MIT license.
- **[Prime Agent](https://github.com/primeintellect-ai/prime)** (Prime Intellect) — the *Context as Variables* paradigm that shapes the persistent-kernel model.

All upstream dependencies carry their original MIT licenses, recorded in the source tree.

---

## 📄 License

MIT. See [LICENSE](./LICENSE).

数据来源:公开的 DeepSeek Harness 插件目录与各插件 GitHub 仓库。本站为独立第三方目录,与 DeepSeek、幻方(High-Flyer)及插件作者均无隶属或背书关系。

每日精选 Skill 推荐,免费送到你邮箱

输入邮箱,每天接收一个精选 AI Agent 技能推荐。完全免费,持续更新。

提交后我们会发送一封确认邮件,点击邮件里的链接才会开始收信。

完全免费,取消任意时间。我们不会发送垃圾邮件。