@hawk2048/oh-my-dsh 0.2.0 → 0.2.1

package/README.md

@@ -1,77 +1,106 @@
 # oh-my-dsh
 
-**面向 DeepSeek Harness (DSH) 的多智能体编排层** —— 把 [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode) 的编排模式，映射到 DSH 原生原语（`workflow` / `subagent` / `ralph` / `goal`）之上。
+[![npm version](https://img.shields.io/npm/v/@hawk2048/oh-my-dsh?color=cb3837)](https://www.npmjs.com/package/@hawk2048/oh-my-dsh)
+[![GitHub stars](https://img.shields.io/github/stars/hawk2048/oh-my-dsh?style=flat&color=yellow)](https://github.com/hawk2048/oh-my-dsh/stargazers)
+[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)
+[![dsh-plugin](https://img.shields.io/badge/dsh-plugin-orange)](https://github.com/topics/dsh-plugin)
 
-> 两种触发方式：**自然语言**（29 个随 preset 打包的编排 skill，代理自动加载）+ **斜杠命令**（12 条，敲 `/omd` 即弹出候选）。启动一次会话，编排能力就位。
+**A multi-agent orchestration layer for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (DSH)** — ports [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode)'s orchestration modes onto DSH's native primitives (`workflow` / `subagent` / `ralph` / `goal`).
 
-## 快速开始
+> Two ways to trigger: **natural language** (29 skills, auto-loaded by the agent) + **slash commands** (12 `/omd-*` commands). Start a session, orchestration is ready.
 
-一条命令装好全部 29 个 skill + 12 条 `/omd-*` 命令（等价 OMC 的插件市场安装）：
+[中文](README.zh.md)
+
+## Quick Start
+
+One command installs all 29 skills + 12 `/omd-*` commands (the OMC plugin-marketplace equivalent):
 
 ```sh
 dsh plugin --profile web add github:hawk2048/oh-my-dsh
 ```
 
-装完重启 profile，敲 `/omd` 弹出命令候选；或直接用自然语言（「先规划一下」「autopilot 搭个 API」）触发对应 skill。完整说明见下方[安装 / 使用](#安装--使用)。
+Or from npm (the `npm i -g` equivalent):
+
+```sh
+dsh plugin --profile web add @hawk2048/oh-my-dsh
+```
+
+Restart the profile, type `/omd` to see command candidates, or use natural language ("plan this", "autopilot, build a REST API").
+
+## Requirements
+
+- [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (`dsh` CLI)
+- [pnpm](https://pnpm.io) — `dsh plugin` forwards to pnpm
+
+## Why oh-my-dsh
+
+- **Zero configuration** — one `dsh plugin add` installs everything; the bundle self-registers, no setup wizard.
+- **Team-first orchestration** — Team is the canonical staged pipeline (`team-plan → team-prd → team-exec → team-verify → team-fix`), built on DSH's native `workflow`.
+- **Natural language interface** — describe what you want; the 29 `omd-*` skills auto-load, nothing to memorize.
+- **Automatic parallelization** — complex tasks fan out across parallel `subagent`s.
+- **Persistent execution** — Ralph / Autopilot / Ultragoal keep going until verified complete (native `ralph` loops + `goal` cross-turn objectives).
+- **Zero runtime overhead** — adds no new host tools or runtime, just skills over DSH's native primitives, so you keep DSH's own model routing and sandbox.
+- **Learn from experience** — `omd-skillify` extracts reusable skills; `omd-remember` persists durable notes across sessions.
+- **Real-time visibility** — the inline HUD renders live `dsh-ui` progress dashboards.
 
-## 与 oh-my-claudecode 的对应关系
+## OMC correspondence
 
-对齐 OMC **v5.3.0** 的命令面（核心流水线 `plan → execute → review → verify` + 保留模式）：
+Aligned to OMC **v5.3.0** (the `plan → execute → review → verify` pipeline + retained modes):
 
-| OMC 工作流 | oh-my-dsh skill | DSH 原生载体 |
+| OMC workflow | oh-my-dsh skill | DSH primitive |
 |-----------|-----------------|--------------|
-| Plan（规划） | [`omd-plan`](skills/omd-plan/SKILL.md) | 只读探索 + `subagent` 并行调研 |
-| Execute（执行） | [`omd-execute`](skills/omd-execute/SKILL.md) | `subagent` 并行 + `todo_write` |
-| Review（评审） | [`omd-review`](skills/omd-review/SKILL.md) | `subagent_fork` 独立评审 |
-| Verify（验证） | [`omd-verify`](skills/omd-verify/SKILL.md) | `pwsh`/`bash` 测试 + `subagent` 复核 |
-| Team（分阶段流水线） | [`omd-team`](skills/omd-team/SKILL.md) | `workflow` + `subagent` |
-| Autopilot（自动驾驶） | [`omd-autopilot`](skills/omd-autopilot/SKILL.md) | `goal` + `todo` + `subagent` |
-| Ralph（持久验证循环） | [`omd-ralph`](skills/omd-ralph/SKILL.md) | `ralph` + `goal` 验证循环 |
-| Deep Interview（需求澄清） | [`omd-deep-interview`](skills/omd-deep-interview/SKILL.md) | `ask_user_question` |
-| Ralplan（共识规划） | [`omd-ralplan`](skills/omd-ralplan/SKILL.md) | 并行 `subagent` 草案 + 仲裁收敛 |
-| Research（有据调研） | [`omd-research`](skills/omd-research/SKILL.md) | `web_search`/`web_fetch` + 并行 `subagent` |
-| Autoresearch（评估驱动改进） | [`omd-autoresearch`](skills/omd-autoresearch/SKILL.md) | `goal` + 评估子代理循环 |
-| Ultragoal（持久多目标） | [`omd-ultragoal`](skills/omd-ultragoal/SKILL.md) | `goal` + `todo_write` + checkpoint 工件 |
-| HUD（状态栏） | [`omd-hud`](skills/omd-hud/SKILL.md) | `dsh-ui` 内联仪表盘（genui） |
-
-**提示词触发 + 工具 skill**（自然语言触发，无独立斜杠命令）：
-
-| OMC 触发 | oh-my-dsh skill | DSH 原生载体 |
+| Plan | [`omd-plan`](skills/omd-plan/SKILL.md) | read-only exploration + parallel `subagent` |
+| Execute | [`omd-execute`](skills/omd-execute/SKILL.md) | parallel `subagent` + `todo_write` |
+| Review | [`omd-review`](skills/omd-review/SKILL.md) | `subagent_fork` independent review |
+| Verify | [`omd-verify`](skills/omd-verify/SKILL.md) | `pwsh`/`bash` tests + `subagent` re-check |
+| Team | [`omd-team`](skills/omd-team/SKILL.md) | `workflow` + `subagent` |
+| Autopilot | [`omd-autopilot`](skills/omd-autopilot/SKILL.md) | `goal` + `todo` + `subagent` |
+| Ralph | [`omd-ralph`](skills/omd-ralph/SKILL.md) | `ralph` + `goal` verify loop |
+| Deep Interview | [`omd-deep-interview`](skills/omd-deep-interview/SKILL.md) | `ask_user_question` |
+| Ralplan | [`omd-ralplan`](skills/omd-ralplan/SKILL.md) | parallel `subagent` drafts + arbitration |
+| Research | [`omd-research`](skills/omd-research/SKILL.md) | `web_search`/`web_fetch` + parallel `subagent` |
+| Autoresearch | [`omd-autoresearch`](skills/omd-autoresearch/SKILL.md) | `goal` + evaluator loop |
+| Ultragoal | [`omd-ultragoal`](skills/omd-ultragoal/SKILL.md) | `goal` + `todo_write` + checkpoint artifacts |
+| HUD | [`omd-hud`](skills/omd-hud/SKILL.md) | inline `dsh-ui` dashboard (genui) |
+
+**Prompt-trigger + utility skills** (natural-language only, no slash command):
+
+| OMC trigger | oh-my-dsh skill | DSH primitive |
 |----------|-----------------|--------------|
-| tdd（测试驱动） | [`omd-tdd`](skills/omd-tdd/SKILL.md) | `pwsh`/`bash` 测试 + red-green-refactor |
-| deepsearch（代码检索） | [`omd-deepsearch`](skills/omd-deepsearch/SKILL.md) | `grep`/`glob`/`read` 本地定位 |
-| deep-analyze（深度分析） | [`omd-deep-analyze`](skills/omd-deep-analyze/SKILL.md) | 并行 `subagent` 多视角 |
-| ultrathink（深度推理） | [`omd-ultrathink`](skills/omd-ultrathink/SKILL.md) | 纯推理（先不动工具） |
-| ai-slop-cleaner（去废话） | [`omd-ai-slop-cleaner`](skills/omd-ai-slop-cleaner/SKILL.md) | `review`/`edit` 精简 |
-| omc-doctor（自检） | [`omd-doctor`](skills/omd-doctor/SKILL.md) | 安装/挂载诊断 |
-| debug（调试） | [`omd-debug`](skills/omd-debug/SKILL.md) | 复现→定位→最小修复→回归 |
-| release（发布） | [`omd-release`](skills/omd-release/SKILL.md) | git + 测试 + changelog + tag |
-| remember（跨会话记忆） | [`omd-remember`](skills/omd-remember/SKILL.md) | 工作区记忆文件 |
-| cancel（取消模式） | [`omd-cancel`](skills/omd-cancel/SKILL.md) | goal pause + interrupt_agent + job_kill |
-| minimal-code-discipline（最小改动） | [`omd-minimal-code-discipline`](skills/omd-minimal-code-discipline/SKILL.md) | YAGNI 阶梯 |
-| external-context（外部上下文） | [`omd-external-context`](skills/omd-external-context/SKILL.md) | `web_fetch`/`read` |
-| visual-verdict（视觉 QA） | [`omd-visual-verdict`](skills/omd-visual-verdict/SKILL.md) | `read_image` 对比 |
-| skillify（提取 skill） | [`omd-skillify`](skills/omd-skillify/SKILL.md) | 写 `SKILL.md` |
-| wiki（知识库） | [`omd-wiki`](skills/omd-wiki/SKILL.md) | 工作区 markdown |
-| omc-help（使用指南） | [`omd-help`](skills/omd-help/SKILL.md) | 实时列 skill/命令索引 |
-
-> **`ultrawork` 已退役**：OMC 在 5.0.0 删除了 `ultrawork`（以及 `ultraqa`/`swarm`/`pipeline` 等），oh-my-dsh 同步移除；其"最大并行扇出"能力已并入 `execute`/`team`/`ralph` 的内部并行纪律。
-
-关键差异：OMC 的 HUD 是 CLI 状态栏，而 oh-my-dsh 的 HUD 是**内联在回答正文里的 `dsh-ui` 仪表盘**（stat/progress/timeline/table），因为 DSH 的 Web GUI 本身就在渲染结构化 UI。
-
-## 目录结构
+| tdd | [`omd-tdd`](skills/omd-tdd/SKILL.md) | `pwsh`/`bash` tests + red-green-refactor |
+| deepsearch | [`omd-deepsearch`](skills/omd-deepsearch/SKILL.md) | `grep`/`glob`/`read` local lookup |
+| deep-analyze | [`omd-deep-analyze`](skills/omd-deep-analyze/SKILL.md) | parallel `subagent` multi-angle |
+| ultrathink | [`omd-ultrathink`](skills/omd-ultrathink/SKILL.md) | pure reasoning (no tools first) |
+| ai-slop-cleaner | [`omd-ai-slop-cleaner`](skills/omd-ai-slop-cleaner/SKILL.md) | `review`/`edit` tightening |
+| omc-doctor | [`omd-doctor`](skills/omd-doctor/SKILL.md) | install/mount diagnostics |
+| debug | [`omd-debug`](skills/omd-debug/SKILL.md) | reproduce→isolate→fix→regress |
+| release | [`omd-release`](skills/omd-release/SKILL.md) | git + tests + changelog + tag |
+| remember | [`omd-remember`](skills/omd-remember/SKILL.md) | workspace memory file |
+| cancel | [`omd-cancel`](skills/omd-cancel/SKILL.md) | goal pause + `interrupt_agent` + `job_kill` |
+| minimal-code-discipline | [`omd-minimal-code-discipline`](skills/omd-minimal-code-discipline/SKILL.md) | YAGNI ladder |
+| external-context | [`omd-external-context`](skills/omd-external-context/SKILL.md) | `web_fetch`/`read` |
+| visual-verdict | [`omd-visual-verdict`](skills/omd-visual-verdict/SKILL.md) | `read_image` comparison |
+| skillify | [`omd-skillify`](skills/omd-skillify/SKILL.md) | write `SKILL.md` |
+| wiki | [`omd-wiki`](skills/omd-wiki/SKILL.md) | workspace markdown |
+| omc-help | [`omd-help`](skills/omd-help/SKILL.md) | live skill/command index |
+
+> **`ultrawork` is retired** — OMC removed it in 5.0.0 (along with `ultraqa`/`swarm`/`pipeline`); oh-my-dsh removed it too, folding its "maximum parallel fan-out" into `execute`/`team`/`ralph`.
+
+Key difference: OMC's HUD is a CLI status line, while oh-my-dsh's HUD is an **inline `dsh-ui` dashboard in the reply** (stat/progress/timeline/table), because DSH's Web GUI already renders structured UI.
+
+## Directory structure
 
 ```
 oh-my-dsh/
-├── package.json              # dsh.bundle manifest（组合包安装入口）
-├── cordis.patch.yml          # 组合包 patch 层（全局命令 + 全局 skill 挂载）
-├── index.js                  # 组合包模块入口
-├── agent.cordis.yml          # agent preset 组合文件（= standard + persona + skill + 命令）
-├── preset.yml                # preset 显示元数据
-├── install.ps1 / install.sh  # preset 手动安装脚本
+├── package.json              # dsh.bundle manifest (bundle install entry)
+├── cordis.patch.yml          # bundle patch layer (global commands + skills)
+├── index.js                  # bundle module entry
+├── agent.cordis.yml          # agent preset composition (= standard + persona + skills + commands)
+├── preset.yml                # preset display metadata
+├── install.ps1 / install.sh  # preset manual install scripts
 ├── LICENSE
 ├── commands/
-│   └── omd-commands.mjs    # 斜杠命令生产者（ESM 插件）
+│   └── omd-commands.mjs      # slash-command producer (ESM plugin)
 ├── skills/
 │   ├── omd-plan/SKILL.md
 │   ├── omd-execute/SKILL.md
@@ -107,23 +136,23 @@ oh-my-dsh/
     └── RELEASE.md
 ```
 
-## 安装 / 使用
+## Install / usage
 
-两种安装方式，**任选其一即可**（bundle 装全局、preset 按会话选）。
+Two forms — pick one (bundle for global, preset for per-session).
 
-**方式 A：组合包 bundle（主要，等价 OMC 的插件市场安装）** —— 全局装好 `/omd-*` 命令与 29 个 skill，profile 里所有会话可见：
+**Form A: bundle (primary, the OMC plugin-marketplace equivalent)** — installs the `/omd-*` commands and 29 skills globally, visible to every session in the profile:
 
 ```sh
-# git 源码安装（纯 JS/Markdown，零构建，无需 allowBuilds 授权；用 @<tag> 可锁定版本）
+# git source (pure JS/Markdown, zero build, no allowBuilds needed; pin with @<tag>)
 dsh plugin --profile web add github:hawk2048/oh-my-dsh
 
-# 或从 npm 安装（等价 OMC 的 npm i -g）
+# or npm (the npm i -g equivalent)
 dsh plugin --profile web add @hawk2048/oh-my-dsh
 ```
 
-装完重启 profile 即可。用户全程只敲一条命令，由 pnpm 拉取，**不手动下载压缩包**。
+Restart the profile. One command, pulled by pnpm — no manual tarball download.
 
-**方式 B：agent preset（可选，按会话选择）** —— 含 persona 与编排工具集的全量会话：
+**Form B: agent preset (optional, per-session)** — full session with persona and the orchestration toolkit:
 
 ```sh
 # Windows
@@ -132,75 +161,75 @@ dsh plugin --profile web add @hawk2048/oh-my-dsh
 ./install.sh
 ```
 
-然后在 DSH 里**新建会话时选择 "oh-my-dsh 编排"** preset（或把它设为 `agent-presets.default`）。详见 [docs/RELEASE.md](docs/RELEASE.md)。
-
-## 怎么用（会话内）
-
-两种方式，效果一样：
-
-**方式一：自然语言**（代理自动加载对应 skill）
-
-- `先规划一下怎么实现 X` → `omd-plan`
-- `按计划实现` → `omd-execute`
-- `评审一下刚才的改动` → `omd-review`
-- `验证一下，测试要全绿` → `omd-verify`
-- `用 team 做一个任务管理应用` → `omd-team`
-- `autopilot：搭一个 REST API` → `omd-autopilot`
-- `ralph 重构认证，必须测试全绿` → `omd-ralph`
-- `先深度访谈，我想做个 xx` → `omd-deep-interview`
-- `ralplan：多角度论证一下这个架构选型` → `omd-ralplan`
-- `调研一下 X 的最佳实践，要带来源` → `omd-research`
-- `autoresearch：把这份报告打磨到达标` → `omd-autoresearch`
-- `ultragoal：这个长期目标拆成子目标推进` → `omd-ultragoal`
-- `tdd：先写失败测试再实现` → `omd-tdd`
-- `deepsearch：定位谁调用了这个函数` → `omd-deepsearch`
-- `deep-analyze：深度分析一下这块架构` → `omd-deep-analyze`
-- `ultrathink：想清楚再动手` → `omd-ultrathink`
-- `去 AI 废话，精简这段` → `omd-ai-slop-cleaner`
-- `自检一下 oh-my-dsh 装好没` → `omd-doctor`
-- `debug：排查这个报错` → `omd-debug`
-- `发布 v1.2，写 changelog` → `omd-release`
-- `记住这个决定，下个会话还要用` → `omd-remember`
-- `停掉正在跑的 autopilot` → `omd-cancel`
-- `最小改动，别过度设计` → `omd-minimal-code-discipline`
-- `看下这个 issue/链接` → `omd-external-context`
-- `对比这两张截图` → `omd-visual-verdict`
-- `把这段流程沉淀成 skill` → `omd-skillify`
-- `记到知识库/更新文档` → `omd-wiki`
-- `omd 能做什么 / 有哪些命令` → `omd-help`
-
-**方式二：斜杠命令**（在输入框敲 `/omd` 弹出候选）
-
-| 命令 | 等价 skill |
+Then create a session and pick the "oh-my-dsh 编排" preset (or set it as `agent-presets.default`). See [docs/RELEASE.md](docs/RELEASE.md).
+
+## Usage (in-session)
+
+Two equivalent ways:
+
+**Way 1: natural language** (the agent auto-loads the matching skill)
+
+- `plan how to implement X` → `omd-plan`
+- `execute the plan` → `omd-execute`
+- `review my changes` → `omd-review`
+- `verify — tests must pass` → `omd-verify`
+- `team, build a task-management app` → `omd-team`
+- `autopilot: build a REST API` → `omd-autopilot`
+- `ralph the auth refactor until green` → `omd-ralph`
+- `deep interview me — I want to build X` → `omd-deep-interview`
+- `ralplan this architecture decision` → `omd-ralplan`
+- `research X best practices, with sources` → `omd-research`
+- `autoresearch this report to passing` → `omd-autoresearch`
+- `ultragoal: break this long goal into subgoals` → `omd-ultragoal`
+- `tdd: failing test first` → `omd-tdd`
+- `deepsearch: who calls this function` → `omd-deepsearch`
+- `deep-analyze this architecture` → `omd-deep-analyze`
+- `ultrathink before acting` → `omd-ultrathink`
+- `strip the AI slop` → `omd-ai-slop-cleaner`
+- `self-check the install` → `omd-doctor`
+- `debug this error` → `omd-debug`
+- `release v1.2 with changelog` → `omd-release`
+- `remember this decision` → `omd-remember`
+- `stop the running autopilot` → `omd-cancel`
+- `minimal change, no over-engineering` → `omd-minimal-code-discipline`
+- `read this issue/link` → `omd-external-context`
+- `compare these screenshots` → `omd-visual-verdict`
+- `turn this flow into a skill` → `omd-skillify`
+- `write this to the wiki` → `omd-wiki`
+- `what can omd do` → `omd-help`
+
+**Way 2: slash commands** (type `/omd` for candidates)
+
+| Command | Equivalent skill |
 |------|-----------|
-| `/omd-plan <任务>` | `omd-plan` |
-| `/omd-execute <任务>` | `omd-execute` |
-| `/omd-review <任务>` | `omd-review` |
-| `/omd-verify <任务>` | `omd-verify` |
-| `/omd-team <任务>` | `omd-team` |
-| `/omd-autopilot <目标>` | `omd-autopilot` |
-| `/omd-ralph <目标>` | `omd-ralph` |
-| `/omd-deep-interview <主题>` | `omd-deep-interview` |
-| `/omd-ralplan <描述>` | `omd-ralplan` |
-| `/omd-research <问题>` | `omd-research` |
-| `/omd-autoresearch <任务>` | `omd-autoresearch` |
-| `/omd-ultragoal <目标>` | `omd-ultragoal` |
+| `/omd-plan <task>` | `omd-plan` |
+| `/omd-execute <task>` | `omd-execute` |
+| `/omd-review <task>` | `omd-review` |
+| `/omd-verify <task>` | `omd-verify` |
+| `/omd-team <task>` | `omd-team` |
+| `/omd-autopilot <goal>` | `omd-autopilot` |
+| `/omd-ralph <goal>` | `omd-ralph` |
+| `/omd-deep-interview <topic>` | `omd-deep-interview` |
+| `/omd-ralplan <desc>` | `omd-ralplan` |
+| `/omd-research <question>` | `omd-research` |
+| `/omd-autoresearch <task>` | `omd-autoresearch` |
+| `/omd-ultragoal <goal>` | `omd-ultragoal` |
 
-代理会加载对应的 skill，按该模式编排，并用 HUD 实时汇报进度。
+The agent loads the matching skill, runs that mode, and reports progress via the HUD.
 
-## 为什么是 skill + bundle/preset，而不是新工具
+## Why skills + bundle/preset, not new tools
 
-DSH 的原生多智能体原语（`workflow` 的流水线/并行、`subagent` 的扇出、`ralph` 的 fresh-agent 循环、`goal` 的跨轮目标）**已经覆盖了 OMC 的编排语义**。oh-my-dsh 因此不引入新的宿主工具或运行时，而是把「什么时候用哪个原语、怎么编排、怎么收尾」沉淀成可复用的 skill。
+DSH's native multi-agent primitives (`workflow` pipelines/parallel, `subagent` fan-out, `ralph` fresh-agent loops, `goal` cross-turn objectives) **already cover OMC's orchestration semantics**. oh-my-dsh therefore adds no new host tools or runtime — it distills "which primitive to use, when, how to orchestrate, and how to close out" into reusable skills.
 
-分发也复用 DSH 原生机制：`cordis.patch.yml` + `package.json` 的 `dsh.bundle` manifest 让 `dsh plugin add` 装成**组合包**（全局命令 + 全局 skill），`agent.cordis.yml` 让它作为**agent preset** 按会话挂载（persona + 命令 + skill）。斜杠命令的 `commands/omd-commands.mjs` 是一个 ESM 插件，注入 DSH 自带的 `commands` 服务注册命令——零运行时风险，升级 DSH 不受影响。
+Distribution reuses DSH's native mechanisms: `cordis.patch.yml` + the `dsh.bundle` manifest make `dsh plugin add` install a **bundle** (global commands + skills), and `agent.cordis.yml` makes it an **agent preset** (per-session persona + commands + skills). The slash-command `commands/omd-commands.mjs` is an ESM plugin that injects DSH's built-in `commands` registry — zero runtime risk, survives DSH upgrades.
 
-## 扩展
+## Extending
 
-- **加模式**：在 `skills/` 下新增 `omd-xxx/SKILL.md`（YAML frontmatter 写 `name` + `description`）。
-- **加命令**：在 `commands/omd-commands.mjs` 的 `MODES` 数组里追加一条。
-- **改 persona**：编辑 `agent.cordis.yml` 顶部的 `persona` 行。
-- **把 skill 做成全局**：走 bundle（`dsh plugin add`），见上文方式 A；或把 `skills/*` 复制到 `${DSH_HOME}/skills/`。
+- **Add a mode**: add `skills/omd-xxx/SKILL.md` (YAML frontmatter with `name` + `description`).
+- **Add a command**: append an entry to the `MODES` array in `commands/omd-commands.mjs`.
+- **Change the persona**: edit the `persona` row at the top of `agent.cordis.yml`.
+- **Make a skill global**: use the bundle (`dsh plugin add`) or copy `skills/*` to `${DSH_HOME}/skills/`.
 
-## 许可
+## License
 
-MIT。灵感来自 [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode)。
+MIT. Inspired by [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode).

package/README.zh.md

@@ -0,0 +1,235 @@
+# oh-my-dsh
+
+[![npm version](https://img.shields.io/npm/v/@hawk2048/oh-my-dsh?color=cb3837)](https://www.npmjs.com/package/@hawk2048/oh-my-dsh)
+[![GitHub stars](https://img.shields.io/github/stars/hawk2048/oh-my-dsh?style=flat&color=yellow)](https://github.com/hawk2048/oh-my-dsh/stargazers)
+[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)
+[![dsh-plugin](https://img.shields.io/badge/dsh-plugin-orange)](https://github.com/topics/dsh-plugin)
+
+**面向 DeepSeek Harness (DSH) 的多智能体编排层** —— 把 [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode) 的编排模式，映射到 DSH 原生原语（`workflow` / `subagent` / `ralph` / `goal`）之上。
+
+> 两种触发方式：**自然语言**（29 个随 preset 打包的编排 skill，代理自动加载）+ **斜杠命令**（12 条，敲 `/omd` 即弹出候选）。启动一次会话，编排能力就位。
+
+[English](README.md) | 中文
+
+## 快速开始
+
+一条命令装好全部 29 个 skill + 12 条 `/omd-*` 命令（等价 OMC 的插件市场安装）：
+
+```sh
+dsh plugin --profile web add github:hawk2048/oh-my-dsh
+```
+
+或从 npm 安装（等价 OMC 的 npm i -g）：
+
+```sh
+dsh plugin --profile web add @hawk2048/oh-my-dsh
+```
+
+装完重启 profile，敲 `/omd` 弹出命令候选；或直接用自然语言（「先规划一下」「autopilot 搭个 API」）触发对应 skill。
+
+## 要求
+
+- [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness)（`dsh` CLI）
+- [pnpm](https://pnpm.io)（`dsh plugin` 底层转发给 pnpm）
+
+## 为什么用 oh-my-dsh
+
+- **零配置**——一条 `dsh plugin add` 装好全部；bundle 自注册，无需安装向导。
+- **Team 优先编排**——Team 是标准分阶段流水线（`team-plan → team-prd → team-exec → team-verify → team-fix`），构建在 DSH 原生 `workflow` 上。
+- **自然语言交互**——说清需求即可，29 个 `omd-*` skill 自动加载，无需记命令。
+- **自动并行**——复杂任务扇出到并行 `subagent`。
+- **持久执行**——Ralph / Autopilot / Ultragoal 不达验证完成不罢休（原生 `ralph` 循环 + `goal` 跨轮目标）。
+- **零运行时开销**——不引入新宿主工具或运行时，只是 DSH 原生原语之上的 skill，保留 DSH 自己的模型路由与沙箱。
+- **从经验中学习**——`omd-skillify` 提取可复用 skill；`omd-remember` 跨会话持久记忆。
+- **实时可见**——内联 HUD 渲染实时 `dsh-ui` 进度仪表盘。
+
+## 与 oh-my-claudecode 的对应关系
+
+对齐 OMC **v5.3.0** 的命令面（核心流水线 `plan → execute → review → verify` + 保留模式）：
+
+| OMC 工作流 | oh-my-dsh skill | DSH 原生载体 |
+|-----------|-----------------|--------------|
+| Plan（规划） | [`omd-plan`](skills/omd-plan/SKILL.md) | 只读探索 + `subagent` 并行调研 |
+| Execute（执行） | [`omd-execute`](skills/omd-execute/SKILL.md) | `subagent` 并行 + `todo_write` |
+| Review（评审） | [`omd-review`](skills/omd-review/SKILL.md) | `subagent_fork` 独立评审 |
+| Verify（验证） | [`omd-verify`](skills/omd-verify/SKILL.md) | `pwsh`/`bash` 测试 + `subagent` 复核 |
+| Team（分阶段流水线） | [`omd-team`](skills/omd-team/SKILL.md) | `workflow` + `subagent` |
+| Autopilot（自动驾驶） | [`omd-autopilot`](skills/omd-autopilot/SKILL.md) | `goal` + `todo` + `subagent` |
+| Ralph（持久验证循环） | [`omd-ralph`](skills/omd-ralph/SKILL.md) | `ralph` + `goal` 验证循环 |
+| Deep Interview（需求澄清） | [`omd-deep-interview`](skills/omd-deep-interview/SKILL.md) | `ask_user_question` |
+| Ralplan（共识规划） | [`omd-ralplan`](skills/omd-ralplan/SKILL.md) | 并行 `subagent` 草案 + 仲裁收敛 |
+| Research（有据调研） | [`omd-research`](skills/omd-research/SKILL.md) | `web_search`/`web_fetch` + 并行 `subagent` |
+| Autoresearch（评估驱动改进） | [`omd-autoresearch`](skills/omd-autoresearch/SKILL.md) | `goal` + 评估子代理循环 |
+| Ultragoal（持久多目标） | [`omd-ultragoal`](skills/omd-ultragoal/SKILL.md) | `goal` + `todo_write` + checkpoint 工件 |
+| HUD（状态栏） | [`omd-hud`](skills/omd-hud/SKILL.md) | `dsh-ui` 内联仪表盘（genui） |
+
+**提示词触发 + 工具 skill**（自然语言触发，无独立斜杠命令）：
+
+| OMC 触发 | oh-my-dsh skill | DSH 原生载体 |
+|----------|-----------------|--------------|
+| tdd（测试驱动） | [`omd-tdd`](skills/omd-tdd/SKILL.md) | `pwsh`/`bash` 测试 + red-green-refactor |
+| deepsearch（代码检索） | [`omd-deepsearch`](skills/omd-deepsearch/SKILL.md) | `grep`/`glob`/`read` 本地定位 |
+| deep-analyze（深度分析） | [`omd-deep-analyze`](skills/omd-deep-analyze/SKILL.md) | 并行 `subagent` 多视角 |
+| ultrathink（深度推理） | [`omd-ultrathink`](skills/omd-ultrathink/SKILL.md) | 纯推理（先不动工具） |
+| ai-slop-cleaner（去废话） | [`omd-ai-slop-cleaner`](skills/omd-ai-slop-cleaner/SKILL.md) | `review`/`edit` 精简 |
+| omc-doctor（自检） | [`omd-doctor`](skills/omd-doctor/SKILL.md) | 安装/挂载诊断 |
+| debug（调试） | [`omd-debug`](skills/omd-debug/SKILL.md) | 复现→定位→最小修复→回归 |
+| release（发布） | [`omd-release`](skills/omd-release/SKILL.md) | git + 测试 + changelog + tag |
+| remember（跨会话记忆） | [`omd-remember`](skills/omd-remember/SKILL.md) | 工作区记忆文件 |
+| cancel（取消模式） | [`omd-cancel`](skills/omd-cancel/SKILL.md) | goal pause + interrupt_agent + job_kill |
+| minimal-code-discipline（最小改动） | [`omd-minimal-code-discipline`](skills/omd-minimal-code-discipline/SKILL.md) | YAGNI 阶梯 |
+| external-context（外部上下文） | [`omd-external-context`](skills/omd-external-context/SKILL.md) | `web_fetch`/`read` |
+| visual-verdict（视觉 QA） | [`omd-visual-verdict`](skills/omd-visual-verdict/SKILL.md) | `read_image` 对比 |
+| skillify（提取 skill） | [`omd-skillify`](skills/omd-skillify/SKILL.md) | 写 `SKILL.md` |
+| wiki（知识库） | [`omd-wiki`](skills/omd-wiki/SKILL.md) | 工作区 markdown |
+| omc-help（使用指南） | [`omd-help`](skills/omd-help/SKILL.md) | 实时列 skill/命令索引 |
+
+> **`ultrawork` 已退役**：OMC 在 5.0.0 删除了 `ultrawork`（以及 `ultraqa`/`swarm`/`pipeline` 等），oh-my-dsh 同步移除；其"最大并行扇出"能力已并入 `execute`/`team`/`ralph` 的内部并行纪律。
+
+关键差异：OMC 的 HUD 是 CLI 状态栏，而 oh-my-dsh 的 HUD 是**内联在回答正文里的 `dsh-ui` 仪表盘**（stat/progress/timeline/table），因为 DSH 的 Web GUI 本身就在渲染结构化 UI。
+
+## 目录结构
+
+```
+oh-my-dsh/
+├── package.json              # dsh.bundle manifest（组合包安装入口）
+├── cordis.patch.yml          # 组合包 patch 层（全局命令 + 全局 skill 挂载）
+├── index.js                  # 组合包模块入口
+├── agent.cordis.yml          # agent preset 组合文件（= standard + persona + skill + 命令）
+├── preset.yml                # preset 显示元数据
+├── install.ps1 / install.sh  # preset 手动安装脚本
+├── LICENSE
+├── commands/
+│   └── omd-commands.mjs    # 斜杠命令生产者（ESM 插件）
+├── skills/
+│   ├── omd-plan/SKILL.md
+│   ├── omd-execute/SKILL.md
+│   ├── omd-review/SKILL.md
+│   ├── omd-verify/SKILL.md
+│   ├── omd-team/SKILL.md
+│   ├── omd-autopilot/SKILL.md
+│   ├── omd-ralph/SKILL.md
+│   ├── omd-deep-interview/SKILL.md
+│   ├── omd-ralplan/SKILL.md
+│   ├── omd-research/SKILL.md
+│   ├── omd-autoresearch/SKILL.md
+│   ├── omd-ultragoal/SKILL.md
+│   ├── omd-tdd/SKILL.md
+│   ├── omd-deepsearch/SKILL.md
+│   ├── omd-deep-analyze/SKILL.md
+│   ├── omd-ultrathink/SKILL.md
+│   ├── omd-ai-slop-cleaner/SKILL.md
+│   ├── omd-doctor/SKILL.md
+│   ├── omd-debug/SKILL.md
+│   ├── omd-release/SKILL.md
+│   ├── omd-remember/SKILL.md
+│   ├── omd-cancel/SKILL.md
+│   ├── omd-minimal-code-discipline/SKILL.md
+│   ├── omd-external-context/SKILL.md
+│   ├── omd-visual-verdict/SKILL.md
+│   ├── omd-skillify/SKILL.md
+│   ├── omd-wiki/SKILL.md
+│   ├── omd-help/SKILL.md
+│   └── omd-hud/SKILL.md
+└── docs/
+    ├── ARCHITECTURE.md
+    └── RELEASE.md
+```
+
+## 安装 / 使用
+
+两种安装方式，**任选其一即可**（bundle 装全局、preset 按会话选）。
+
+**方式 A：组合包 bundle（主要，等价 OMC 的插件市场安装）** —— 全局装好 `/omd-*` 命令与 29 个 skill，profile 里所有会话可见：
+
+```sh
+# git 源码安装（纯 JS/Markdown，零构建，无需 allowBuilds 授权；用 @<tag> 可锁定版本）
+dsh plugin --profile web add github:hawk2048/oh-my-dsh
+
+# 或从 npm 安装（等价 OMC 的 npm i -g）
+dsh plugin --profile web add @hawk2048/oh-my-dsh
+```
+
+装完重启 profile 即可。用户全程只敲一条命令，由 pnpm 拉取，**不手动下载压缩包**。
+
+**方式 B：agent preset（可选，按会话选择）** —— 含 persona 与编排工具集的全量会话：
+
+```sh
+# Windows
+.\install.ps1
+# macOS / Linux
+./install.sh
+```
+
+然后在 DSH 里**新建会话时选择 "oh-my-dsh 编排"** preset（或把它设为 `agent-presets.default`）。详见 [docs/RELEASE.md](docs/RELEASE.md)。
+
+## 怎么用（会话内）
+
+两种方式，效果一样：
+
+**方式一：自然语言**（代理自动加载对应 skill）
+
+- `先规划一下怎么实现 X` → `omd-plan`
+- `按计划实现` → `omd-execute`
+- `评审一下刚才的改动` → `omd-review`
+- `验证一下，测试要全绿` → `omd-verify`
+- `用 team 做一个任务管理应用` → `omd-team`
+- `autopilot：搭一个 REST API` → `omd-autopilot`
+- `ralph 重构认证，必须测试全绿` → `omd-ralph`
+- `先深度访谈，我想做个 xx` → `omd-deep-interview`
+- `ralplan：多角度论证一下这个架构选型` → `omd-ralplan`
+- `调研一下 X 的最佳实践，要带来源` → `omd-research`
+- `autoresearch：把这份报告打磨到达标` → `omd-autoresearch`
+- `ultragoal：这个长期目标拆成子目标推进` → `omd-ultragoal`
+- `tdd：先写失败测试再实现` → `omd-tdd`
+- `deepsearch：定位谁调用了这个函数` → `omd-deepsearch`
+- `deep-analyze：深度分析一下这块架构` → `omd-deep-analyze`
+- `ultrathink：想清楚再动手` → `omd-ultrathink`
+- `去 AI 废话，精简这段` → `omd-ai-slop-cleaner`
+- `自检一下 oh-my-dsh 装好没` → `omd-doctor`
+- `debug：排查这个报错` → `omd-debug`
+- `发布 v1.2，写 changelog` → `omd-release`
+- `记住这个决定，下个会话还要用` → `omd-remember`
+- `停掉正在跑的 autopilot` → `omd-cancel`
+- `最小改动，别过度设计` → `omd-minimal-code-discipline`
+- `看下这个 issue/链接` → `omd-external-context`
+- `对比这两张截图` → `omd-visual-verdict`
+- `把这段流程沉淀成 skill` → `omd-skillify`
+- `记到知识库/更新文档` → `omd-wiki`
+- `omd 能做什么 / 有哪些命令` → `omd-help`
+
+**方式二：斜杠命令**（在输入框敲 `/omd` 弹出候选）
+
+| 命令 | 等价 skill |
+|------|-----------|
+| `/omd-plan <任务>` | `omd-plan` |
+| `/omd-execute <任务>` | `omd-execute` |
+| `/omd-review <任务>` | `omd-review` |
+| `/omd-verify <任务>` | `omd-verify` |
+| `/omd-team <任务>` | `omd-team` |
+| `/omd-autopilot <目标>` | `omd-autopilot` |
+| `/omd-ralph <目标>` | `omd-ralph` |
+| `/omd-deep-interview <主题>` | `omd-deep-interview` |
+| `/omd-ralplan <描述>` | `omd-ralplan` |
+| `/omd-research <问题>` | `omd-research` |
+| `/omd-autoresearch <任务>` | `omd-autoresearch` |
+| `/omd-ultragoal <目标>` | `omd-ultragoal` |
+
+代理会加载对应的 skill，按该模式编排，并用 HUD 实时汇报进度。
+
+## 为什么是 skill + bundle/preset，而不是新工具
+
+DSH 的原生多智能体原语（`workflow` 的流水线/并行、`subagent` 的扇出、`ralph` 的 fresh-agent 循环、`goal` 的跨轮目标）**已经覆盖了 OMC 的编排语义**。oh-my-dsh 因此不引入新的宿主工具或运行时，而是把「什么时候用哪个原语、怎么编排、怎么收尾」沉淀成可复用的 skill。
+
+分发也复用 DSH 原生机制：`cordis.patch.yml` + `package.json` 的 `dsh.bundle` manifest 让 `dsh plugin add` 装成**组合包**（全局命令 + 全局 skill），`agent.cordis.yml` 让它作为**agent preset** 按会话挂载（persona + 命令 + skill）。斜杠命令的 `commands/omd-commands.mjs` 是一个 ESM 插件，注入 DSH 自带的 `commands` 服务注册命令——零运行时风险，升级 DSH 不受影响。
+
+## 扩展
+
+- **加模式**：在 `skills/` 下新增 `omd-xxx/SKILL.md`（YAML frontmatter 写 `name` + `description`）。
+- **加命令**：在 `commands/omd-commands.mjs` 的 `MODES` 数组里追加一条。
+- **改 persona**：编辑 `agent.cordis.yml` 顶部的 `persona` 行。
+- **把 skill 做成全局**：走 bundle（`dsh plugin add`），见上文方式 A；或把 `skills/*` 复制到 `${DSH_HOME}/skills/`。
+
+## 许可
+
+MIT。灵感来自 [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode)。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hawk2048/oh-my-dsh",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "description": "oh-my-dsh: DeepSeek Harness multi-agent orchestration layer — OMC-style workflow skills, slash commands, and an agent preset.",
   "type": "module",
   "main": "index.js",
@@ -12,7 +12,8 @@
     "index.js",
     "commands/",
     "skills/",
-    "cordis.patch.yml"
+    "cordis.patch.yml",
+    "README.zh.md"
   ],
   "dsh": {
     "bundle": {
@@ -30,6 +31,6 @@
   ],
   "repository": {
     "type": "git",
-    "url": "https://github.com/hawk2048/oh-my-dsh"
+    "url": "git+https://github.com/hawk2048/oh-my-dsh.git"
   }
 }

package/skills/omd-ai-slop-cleaner/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-ai-slop-cleaner
-description: "AI-slop-cleaner: strip AI-generated filler, boilerplate, redundant comments, and vague prose from code or text. Use when the user says ai-slop / 去废话 / 精简 / 去水 / 太啰嗦 / 去AI味, or wants output tightened."
+description: "去AI味：清除代码或文本里 AI 生成的废话、套话、冗余注释与含糊表述。触发词：ai-slop / 去废话 / 精简 / 去水 / 太啰嗦 / 去AI味；或想让输出更紧凑时。"
 ---
 
 # oh-my-dsh · AI-slop-cleaner 去废话

package/skills/omd-autopilot/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-autopilot
-description: "Autopilot: one lead agent (you) drives a single objective to verified completion across goal rounds, delegating via subagents and tracking with todo_write. Use for end-to-end autonomous feature work with minimal ceremony, or when the user says autopilot/自动驾驶/自主执行/自动驾驶式."
+description: "自动驾驶：一个主控代理（你）跨目标轮次把单一目标持续做到已验证完成，用子代理委派并以 todo_write 跟踪。适合低仪式感的端到端自主功能开发；触发词：autopilot / 自动驾驶 / 自主执行 / 自动驾驶式。"
 ---
 
 # oh-my-dsh · Autopilot 自动驾驶

package/skills/omd-autoresearch/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-autoresearch
-description: "Autoresearch: a bounded evaluator-driven improvement loop — produce a version, an evaluator subagent scores it against the target criteria, then improve, repeat until it meets the bar or the round cap. Use when the user asks to autoresearch/自动调研改进/边做边评/持续打磨到达标."
+description: "自动调研改进：有界的评估器驱动改进循环——产出一版，评估子代理对照目标标准打分，再改进、再评，直到达标或到达轮次上限。触发词：autoresearch / 自动调研改进 / 边做边评 / 持续打磨到达标。"
 ---
 
 # oh-my-dsh · Autoresearch 评估器驱动的改进循环

package/skills/omd-cancel/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-cancel
-description: "Cancel: stop an in-flight orchestration mode — halt goal rounds, interrupt background subagents, kill jobs — and leave a clean checkpoint. Use when the user says cancel / 停止 / 停下 / 取消 / 别跑了 / stop / abort."
+description: "取消：干净地停止进行中的编排模式——暂停目标轮次、中断后台子代理、终止任务——并留下干净的检查点。触发词：cancel / 停止 / 停下 / 取消 / 别跑了 / stop / abort。"
 ---
 
 # oh-my-dsh · Cancel 取消进行中模式

package/skills/omd-debug/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-debug
-description: "Debug: reproduce the failure, isolate the cause with evidence, fix minimally, and lock it with a regression test. Use when the user says debug / 调试 / 报错 / 崩溃 / 为什么失败 / 修 bug / 排查."
+description: "调试：复现失败，用证据定位根因，最小化修复，并用回归测试锁定。触发词：debug / 调试 / 报错 / 崩溃 / 为什么失败 / 修 bug / 排查。"
 ---
 
 # oh-my-dsh · Debug 结构化调试

package/skills/omd-deep-analyze/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-deep-analyze
-description: "Deep-analyze: multi-angle, evidence-backed analysis of code, architecture, or a problem. Use when the user says deep-analyze / 深度分析 / 分析一下 / 为什么这样 / 值得吗, and wants thorough reasoning, not a quick fix."
+description: "深度分析：对代码、架构或问题做多角度、有证据支撑的分析。需要透彻推理而不是快速修补时使用；触发词：deep-analyze / 深度分析 / 分析一下 / 为什么这样 / 值得吗。"
 ---
 
 # oh-my-dsh · Deep-analyze 深度分析

package/skills/omd-deep-interview/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-deep-interview
-description: "Deep Interview: Socratic requirement clarification before any code — converge a vague idea across weighted dimensions using ask_user_question. Use when requirements are vague/ambiguous, or the user says deep-interview/深度访谈/先澄清需求/需求访谈."
+description: "深度访谈：写代码前用苏格拉底式追问澄清需求——用 ask_user_question 在加权维度上收敛模糊想法。需求模糊或有歧义时使用；触发词：deep-interview / 深度访谈 / 先澄清需求 / 需求访谈。"
 ---
 
 # oh-my-dsh · Deep Interview 需求澄清

package/skills/omd-deepsearch/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-deepsearch
-description: "Deepsearch: locate implementations, call sites, definitions, and data flow inside the local codebase. Use when the user says deepsearch / 找一下 / 代码在哪 / 定位 / 谁调用了 / 哪里实现的, or asks where something lives rather than how to do it."
+description: "代码库聚焦检索：定位本地代码库中的实现、调用点、定义与数据流。问「东西在哪」而非「怎么做」时使用；触发词：deepsearch / 找一下 / 代码在哪 / 定位 / 谁调用了 / 哪里实现的。"
 ---
 
 # oh-my-dsh · Deepsearch 代码库聚焦检索

package/skills/omd-doctor/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-doctor
-description: "Doctor: self-check the oh-my-dsh install — verify the skills directory, the /omd-* slash commands, and the preset/bundle mount — and diagnose why a mode isn't working. Use when the user says doctor / 自检 / 诊断 / 为什么没生效 / 检查一下安装 / omd 有问题."
+description: "自检诊断：检查 oh-my-dsh 装得对不对——核对技能目录、/omd-* 斜杠命令与 preset/bundle 挂载——并诊断某个模式为什么没生效。触发词：doctor / 自检 / 诊断 / 为什么没生效 / 检查一下安装 / omd 有问题。"
 ---
 
 # oh-my-dsh · Doctor 自检诊断

package/skills/omd-execute/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-execute
-description: "Execute: implement an agreed plan — route multi-file changes/refactors/debugging to executor subagents (background parallel), work directly on trivial edits, track with todo_write. Use as the implementation stage of plan→execute→review→verify, or when the user asks to execute/实现/开发/动手/开始写."
+description: "执行：按已确认的计划实现——多文件改动/重构/调试交给执行子代理（后台并行），琐碎改动直接做，用 todo_write 跟踪。作为 plan→execute→review→verify 流水线的实现阶段；触发词：execute / 实现 / 开发 / 动手 / 开始写。"
 ---
 
 # oh-my-dsh · Execute 执行

package/skills/omd-external-context/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-external-context
-description: "External-context: pull context from outside the repo — a URL, a doc, an issue, a package README — and summarize it before reasoning or editing. Use when the user gives a URL/issue/doc reference, or says load context / 看下这个链接 / 拉一下上下文 / 参考这个文档."
+description: "外部上下文：从仓库之外拉取上下文——URL、文档、issue、包 README——在推理或编辑前先做摘要。用户给出 URL/issue/文档引用时使用；触发词：load context / 看下这个链接 / 拉一下上下文 / 参考这个文档。"
 ---
 
 # oh-my-dsh · External-context 加载外部上下文

package/skills/omd-help/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-help
-description: "Help: show the oh-my-dsh menu — every omd-* skill and /omd-* command, grouped by category, and how to trigger each. Use when the user asks help / 帮助 / omd 能做什么 / 有哪些命令 / 有什么模式 / usage."
+description: "帮助：展示 oh-my-dsh 菜单——全部 omd-* 技能与 /omd-* 命令，按分类分组并说明各自触发方式。触发词：help / 帮助 / omd 能做什么 / 有哪些命令 / 有什么模式 / usage。"
 ---
 
 # oh-my-dsh · Help 使用索引

package/skills/omd-hud/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-hud
-description: "HUD: render live orchestration progress as an inline dsh-ui dashboard — stat cards, progress bars, timeline, and a subagent-status table. Use to show real-time progress of any oh-my-dsh mode (plan/execute/review/verify/team/autopilot/ralph/deep-interview/ralplan/research/autoresearch/ultragoal/tdd/deepsearch/deep-analyze/ultrathink/ai-slop-cleaner/debug/release/remember/cancel/minimal-code-discipline/external-context/visual-verdict/skillify/wiki/help/doctor)."
+description: "HUD：把编排进度渲染成内联 dsh-ui 仪表盘——统计卡片、进度条、时间线与子代理状态表。用于实时展示任意 oh-my-dsh 模式的进度（plan/execute/review/verify/team/autopilot/ralph/deep-interview/ralplan/research/autoresearch/ultragoal/tdd/deepsearch/deep-analyze/ultrathink/ai-slop-cleaner/debug/release/remember/cancel/minimal-code-discipline/external-context/visual-verdict/skillify/wiki/help/doctor）。"
 ---
 
 # oh-my-dsh · HUD 实时进度仪表盘

package/skills/omd-minimal-code-discipline/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-minimal-code-discipline
-description: "Minimal-code-discipline: make the smallest correct diff — existence-first, reuse before writing, shortest correct change. Use when the user says minimal / 最小改动 / YAGNI / 别过度设计 / 只改必要的 / 能复用就复用."
+description: "最小改动纪律：做最小而正确的改动——先确认存在、能复用就不新写、最短的正确修改。触发词：minimal / 最小改动 / YAGNI / 别过度设计 / 只改必要的 / 能复用就复用。"
 ---
 
 # oh-my-dsh · Minimal-code-discipline 最小改动纪律

package/skills/omd-plan/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-plan
-description: "Plan: explore first with non-mutating reads/search/static analysis, then produce a concrete implementation plan (task breakdown with dependencies + acceptance criteria + file-level changes). No implementation. Use as the first stage of the plan→execute→review→verify pipeline, or when the user asks to plan/规划/方案/计划/怎么实现."
+description: "规划：先用只读方式探索（读/搜索/静态分析），再产出具体的实现计划（带依赖的任务拆解 + 验收标准 + 文件级改动）。不写代码。作为 plan→execute→review→verify 流水线的第一阶段；触发词：plan / 规划 / 方案 / 计划 / 怎么实现。"
 ---
 
 # oh-my-dsh · Plan 规划

package/skills/omd-ralph/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-ralph
-description: "Ralph: persistent verify/fix loops with no silent partials. Use the native ralph tool for explicit fresh-agent iteration, or the manual goal+verify loop otherwise. Use when the task must complete fully with verification, or the user says ralph/拉夫/verify-fix/必须做完."
+description: "拉夫：持久验证/修复循环，不留静默半成品。显式全新代理迭代用原生 ralph 工具，否则用手动 goal+verify 循环。任务必须完整完成并验证时使用；触发词：ralph / 拉夫 / verify-fix / 必须做完。"
 ---
 
 # oh-my-dsh · Ralph 持久验证循环

package/skills/omd-ralplan/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-ralplan
-description: "Ralplan: consensus planning — fan out several independent planner subagents from different angles, then run structured deliberation that adjudicates disagreements into one consensus plan. Use for high-risk or ambiguous features that need multi-perspective planning, or when the user says ralplan/共识规划/多方论证/审慎规划."
+description: "共识规划：多视角并行规划——派出多个独立规划子代理从不同角度切入，再运行结构化审议裁决分歧，收敛为一份共识计划。适合高风险或需求模糊、需要多视角规划的特性；触发词：ralplan / 共识规划 / 多方论证 / 审慎规划。"
 ---
 
 # oh-my-dsh · Ralplan 共识规划

package/skills/omd-release/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-release
-description: "Release: walk the repo's release process — version bump, changelog, tests, tag, build — and produce a release checklist. Use when the user says release / 发布 / 发版 / 打 tag / 打版本 / changelog."
+description: "发布：走完仓库的发布流程——版本号、changelog、测试、打 tag、构建——并产出发布检查清单。触发词：release / 发布 / 发版 / 打 tag / 打版本 / changelog。"
 ---
 
 # oh-my-dsh · Release 发布

package/skills/omd-remember/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-remember
-description: "Remember: persist durable notes, decisions, and facts to a workspace memory file so they survive compaction and later sessions. Use when the user says remember / 记住 / 记下来 / 别忘 / 存个备忘 / 记录这个决定."
+description: "记住：把持久化的笔记、决策与事实写入工作区记忆文件，让它们在压缩和后续会话中留存。触发词：remember / 记住 / 记下来 / 别忘 / 存个备忘 / 记录这个决定。"
 ---
 
 # oh-my-dsh · Remember 跨会话记忆

package/skills/omd-research/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-research
-description: "Research: investigate an open question and return grounded findings — every claim backed by a source. Fan out parallel research subagents (web_search/web_fetch + repo docs), then synthesize a sourced report. Use when the user asks to research/调研/查证/调查/求证."
+description: "调研：查证开放问题并返回有据结论——每条结论都有来源背书。并行派出研究子代理（web_search/web_fetch + 仓库文档），再综合成有来源的报告。触发词：research / 调研 / 查证 / 调查 / 求证。"
 ---
 
 # oh-my-dsh · Research 有据调研

package/skills/omd-review/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-review
-description: "Review: a separate review pass that evaluates the implementation — never self-approve in the same context. Route to a reviewer subagent (subagent_fork or subagent), read the diffs, produce an issues list. Use as the review stage of plan→execute→review→verify, or when the user asks to review/评审/审查/code review."
+description: "评审：独立的评审环节，评估实现——绝不在同一上下文里自我背书。交给评审子代理（subagent_fork 或 subagent）、读 diff、产出缺陷清单。作为 plan→execute→review→verify 的评审阶段；触发词：review / 评审 / 审查 / code review。"
 ---
 
 # oh-my-dsh · Review 评审

package/skills/omd-skillify/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-skillify
-description: "Skillify: extract a repeatable workflow from the current session into a new reusable SKILL.md. Use when the user says skillify / 把这段流程存成 skill / 沉淀成技能 / 做成可复用 / 记住这套做法."
+description: "技能化：把当前会话里可复用的流程提炼成新的可复用 SKILL.md。触发词：skillify / 把这段流程存成 skill / 沉淀成技能 / 做成可复用 / 记住这套做法。"
 ---
 
 # oh-my-dsh · Skillify 从会话提取 skill

package/skills/omd-tdd/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-tdd
-description: "TDD (test-driven development): write the failing test first, then the minimal implementation, then refactor (red-green-refactor). Use when the user says tdd / 测试驱动 / 先写测试 / red-green / 红绿重构, or wants tests to drive development."
+description: "测试驱动开发：先写会失败的测试，再做最小实现，然后重构（红-绿-重构）。触发词：tdd / 测试驱动 / 先写测试 / red-green / 红绿重构；或想用测试驱动开发时。"
 ---
 
 # oh-my-dsh · TDD 测试驱动开发

package/skills/omd-team/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-team
-description: "Team multi-agent orchestration: run a staged pipeline (team-plan → team-prd → team-exec → team-verify → team-fix loop) as a DSH workflow script with subagent fan-out. Use when the user asks for team orchestration, a multi-agent pipeline, or a feature spanning several subsystems (team/团队/多智能体流水线)."
+description: "团队多智能体编排：运行分阶段流水线（team-plan → team-prd → team-exec → team-verify → team-fix 循环），以 DSH workflow 脚本 + 子代理扇出实现。需要团队编排、多智能体流水线或跨子系统特性时使用；触发词：team / 团队 / 多智能体流水线。"
 ---
 
 # oh-my-dsh · Team 分阶段流水线

package/skills/omd-ultragoal/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-ultragoal
-description: "Ultragoal: a durable multi-goal workflow — split one big outcome into multiple sub-goals, checkpoint each to workspace artifacts, and keep a ledger so progress survives turns and sessions. Use when the user asks to ultragoal/持久目标/多目标/长期目标/带台账推进."
+description: "持久目标：持久的多目标工作流——把一个大成果拆成多个子目标，各自 checkpoint 到工作区工件，并维护台账让进度跨回合、跨会话存活。触发词：ultragoal / 持久目标 / 多目标 / 长期目标 / 带台账推进。"
 ---
 
 # oh-my-dsh · Ultragoal 持久多目标

package/skills/omd-ultrathink/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-ultrathink
-description: "Ultrathink: before acting, exhaustively enumerate edge cases, weigh alternatives, and rehearse failure modes. Use when the user says ultrathink / 深度思考 / 想清楚 / 多想想 / 别急着动手 / 三思, and wants deep reasoning before any tool use."
+description: "深度思考：行动前穷举边界情况、权衡备选方案、预演失败模式。触发词：ultrathink / 深度思考 / 想清楚 / 多想想 / 别急着动手 / 三思；任何工具使用前需要深度推理时。"
 ---
 
 # oh-my-dsh · Ultrathink 深度推理

package/skills/omd-verify/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-verify
-description: "Verify: verify before claiming completion — run tests/build, check against acceptance criteria, collect evidence, size the verifier (small inline, large/security via subagent), and iterate on failure. Use as the final stage of plan→execute→review→verify, or when the user asks to verify/验证/测试/检查/跑通."
+description: "验证：宣称完成前先验证——跑测试/构建、对照验收标准、收集证据（小规模内联，大规模/安全走子代理），失败则迭代。作为 plan→execute→review→verify 的最终阶段；触发词：verify / 验证 / 测试 / 检查 / 跑通。"
 ---
 
 # oh-my-dsh · Verify 验证

package/skills/omd-visual-verdict/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-visual-verdict
-description: "Visual-verdict: compare a screenshot or rendered output against a reference and give a structured pass/fail verdict with specific deltas. Use when the user says visual-verdict / 对比截图 / 看下 UI 对不对 / 视觉检查 / 对比这两张图."
+description: "视觉裁决：把截图或渲染输出与参考对照，给出结构化的通过/失败裁决与具体差异。触发词：visual-verdict / 对比截图 / 看下 UI 对不对 / 视觉检查 / 对比这两张图。"
 ---
 
 # oh-my-dsh · Visual-verdict 视觉 QA

package/skills/omd-wiki/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: omd-wiki
-description: "Wiki: maintain a lightweight markdown knowledge base in the workspace — create, update, and cross-link pages and keep them consistent with the code. Use when the user says wiki / 记到知识库 / 更新文档 / 维护文档 / 写进 wiki."
+description: "知识库：维护工作区里的轻量 markdown 知识库——创建、更新、交叉链接页面，并让它们与代码保持一致。触发词：wiki / 记到知识库 / 更新文档 / 维护文档 / 写进 wiki。"
 ---
 
 # oh-my-dsh · Wiki 知识库维护