dsh-project-based-learning 1.1.0 → 2.0.0

package/CHANGELOG.md

@@ -1,12 +1,31 @@
 # Changelog
 
-本文件记录 dsh-coach 的对外变更。版本号含义：
+本文件记录 `dsh-project-based-learning` 的对外变更。版本号含义：
 
 - **插件包版本**（`package.json#version`）：npm 包与组合包层的版本。
 - **引擎版本**（`SKILL.md` 与 `references/engine/` 的教学法逻辑）：`1.0.0` 起，交互协议变化才递增。
 - **状态 schema 版本**（`state.json#schemaVersion`）：`1.0`，字段语义变化才递增。
 - **领域包版本**（`references/domains/<id>/manifest.yml#version`）：各学科独立演进，与引擎版本解耦。
 
+## [2.0.0] - 2026-09-12
+
+**技能名与包名统一，学科机制文档化。** 技能注册名属于包的公开接口，改名会让引用旧名的提示词失效，故按破坏性变更升主版本。
+
+### 变更（破坏性）
+
+- **技能注册名由 `dsh-coach` 改为 `dsh-project-based-learning`**；技能目录由 `skills/dsh-coach/` 改为 `skills/dsh-project-based-learning/`。仓库名、npm 包名、技能名三者自此一致（社区先例：`dsh-ops-skill` 同样以一个同名技能分发）。引用旧技能名的提示词、脚本与文档需同步更新。
+- `cordis.patch.yml` 的插件行 `id` 与加载器模块名同步为包名——该 `name` 必须等于 npm 包名，否则层加载会因模块解析失败而报错。
+
+### 文档
+
+- README 新增「学科与领域包」章节：折叠式列出随包提供的 Unity / C# 包内容（16 题 / 6 原型 / 5 配方 / 7 文件）；给出「让 AI 生成领域包」的可复制提示词；说明切换学科的指令与三条边界（缺小节回退通用行为、校验器不解析题库、领域包须位于技能目录内）。
+- 指令手册补 `切换或替换学科：<id>`。
+- 历史审核记录 `docs/DESIGN-AUDIT.md` 保留 1.0.0 / 1.1.0 时期的旧路径原文，仅在文首加一条改名勘误，不篡改记录。
+
+### 说明
+
+- **引擎版本保持 `1.1.0`**：本次不涉及教学法逻辑，只涉及技能标识与外层文档；`state.json#schemaVersion` 仍为 `1.0`。
+
 ## [1.1.0] - 2026-09-12
 
 **教学缺陷修复**（来自真实使用反馈：只问不教、过度要求实测、该教的让人自己悟）。引擎交互协议变化 → 引擎版本与插件包版本同步递增。
@@ -50,7 +69,7 @@
 
 ### 新增
 
-- **教学引擎**（学科无关）：`skills/dsh-coach/SKILL.md` + `references/engine/`（intake、diagnosis、route、task-loop、review-acceptance、adapt、state、permissions、domain-contract）。
+- **教学引擎**（学科无关）：`skills/dsh-project-based-learning/SKILL.md` + `references/engine/`（intake、diagnosis、route、task-loop、review-acceptance、adapt、state、permissions、domain-contract）。
 - **状态契约**：`.coach/state.json` 为唯一事实源，`.coach/PROGRESS.md` 为生成视图；含机械不变量与多项补充检查（1.0.0 时为 13 条；**1.1.0 增至 21 条**，见 `references/engine/state.md`）。
 - **校验器**：`coach-validate.mjs`（状态层 + 领域包结构 + 引擎分层检查，零依赖）。
 - **回归自测**：`coach-selftest.mjs`（含反向夹具与分层负例自动化；不启动子进程，可在受限沙箱内运行）。

package/CONTRIBUTING.md

@@ -1,6 +1,6 @@
 # 贡献指南
 
-感谢你考虑为 dsh-coach 做贡献。本项目的核心设计是**教学法引擎与学科内容分离**，因此最有价值的贡献通常是**新增或改进一个领域包**。
+感谢你考虑为 dsh-project-based-learning 做贡献。本项目的核心设计是**教学法引擎与学科内容分离**，因此最有价值的贡献通常是**新增或改进一个领域包**。
 
 ## 环境要求
 
@@ -11,8 +11,8 @@
 
 ```bash
 node test/entry.smoke.mjs                                   # 组合包入口契约 + 技能资源齐全
-node skills/dsh-coach/scripts/coach-selftest.mjs            # 校验器回归自测（含反向夹具与分层负例）
-node skills/dsh-coach/scripts/coach-validate.mjs --state examples/state.demo.json
+node skills/dsh-project-based-learning/scripts/coach-selftest.mjs            # 校验器回归自测（含反向夹具与分层负例）
+node skills/dsh-project-based-learning/scripts/coach-validate.mjs --state examples/state.demo.json
 ```
 
 若你手上有 `dsh-plugin-guide` 提供的工具链，可再跑一次静态检查：
@@ -25,18 +25,18 @@ dsh-plugin-dev check --strict
 
 ### 1. 新增学科领域包（最欢迎）
 
-引擎是学科无关的；学科内容全部住在 `skills/dsh-coach/references/domains/<domain-id>/`。新增一个学科 = 新增一个目录，**不需要改引擎**。
+引擎是学科无关的；学科内容全部住在 `skills/dsh-project-based-learning/references/domains/<domain-id>/`。新增一个学科 = 新增一个目录，**不需要改引擎**。
 
 **步骤**
 
-1. 读契约：`skills/dsh-coach/references/engine/domain-contract.md`。
+1. 读契约：`skills/dsh-project-based-learning/references/engine/domain-contract.md`。
 2. 新建 `references/domains/<domain-id>/`，按契约补齐七个文件：
    `manifest.yml`、`archetypes.md`、`diagnosis-bank.md`、`verification.md`、`pitfalls.md`、`example.md`、`glossary.md`。
 3. `manifest.yml` 必须恰好包含契约规定的 8 个键（不增删），且 `id` 与目录名一致。
 4. 跑校验器确认契约通过：
 
    ```bash
-   node skills/dsh-coach/scripts/coach-validate.mjs --state examples/state.demo.json --domain-dir skills/dsh-coach/references/domains/<domain-id>
+   node skills/dsh-project-based-learning/scripts/coach-validate.mjs --state examples/state.demo.json --domain-dir skills/dsh-project-based-learning/references/domains/<domain-id>
    ```
 
 **内容要求（会被人工审）**
@@ -49,7 +49,7 @@ dsh-plugin-dev check --strict
 
 ### 2. 改引擎
 
-引擎文件是 `skills/dsh-coach/SKILL.md` 与 `references/engine/*.md`。
+引擎文件是 `skills/dsh-project-based-learning/SKILL.md` 与 `references/engine/*.md`。
 
 **红线**：`SKILL.md`、`references/engine/*.md`、`assets/*.md` 中**不得出现学科专有词条**。校验器的 `LY01` 检查会用硬令牌黑名单拦截（NFKC 归一化 + 整词匹配；`scripts/` 不参与扫描，因为校验器自身含词表）。这条红线是"可替换特化"能成立的唯一保证。
 
@@ -57,7 +57,7 @@ dsh-plugin-dev check --strict
 
 ### 3. 改校验器与工具
 
-`skills/dsh-coach/scripts/` 下三个脚本，同样零依赖：
+`skills/dsh-project-based-learning/scripts/` 下三个脚本，同样零依赖：
 
 - 新增不变量时，**必须同时**：在 `references/engine/state.md` 声明该不变量、在 `coach-selftest.mjs` 增加断言、并在反向夹具 `examples/state.selftest-invalid.json` 里覆盖它。
 - 禁止引入子进程与管道（受限沙箱下会失败）；自测必须在进程内调用校验函数。

package/README.en.md

@@ -0,0 +1,206 @@
+<p align="center">
+  <strong>A project-based learning coach: the learner does the work, the coach runs the loop</strong><br/>
+  A teaching skill and plugin bundle for <a href="https://github.com/deepseek-ai/deepseek-harness">DeepSeek Harness</a> (DSH) — teaching method and subject matter are separated, domain packs are swappable
+</p>
+
+<p align="center">
+  <a href="README.md">简体中文</a> · <a href="README.en.md"><strong>English</strong></a>
+</p>
+
+<p align="center">
+  <a href="https://www.npmjs.com/package/dsh-project-based-learning"><img src="https://img.shields.io/npm/v/dsh-project-based-learning?style=for-the-badge&logo=npm&label=npm&color=CB3837" alt="npm version" /></a>
+  <a href="https://github.com/Kirisame1969/dsh-project-based-learning/actions"><img src="https://img.shields.io/github/actions/workflow/status/Kirisame1969/dsh-project-based-learning/ci.yml?style=for-the-badge&logo=github&label=CI" alt="CI" /></a>
+  <a href="LICENSE"><img src="https://img.shields.io/github/license/Kirisame1969/dsh-project-based-learning?style=for-the-badge&color=blue" alt="License" /></a>
+  <a href="https://github.com/Kirisame1969/dsh-project-based-learning/stargazers"><img src="https://img.shields.io/github/stars/Kirisame1969/dsh-project-based-learning?style=for-the-badge&logo=github&color=yellow" alt="Stars" /></a>
+</p>
+
+<p align="center">
+  <a href="https://github.com/topics/dsh-plugin"><img src="https://img.shields.io/badge/DSH-plugin-4B6BFB?style=for-the-badge" alt="DSH plugin" /></a>
+  <img src="https://img.shields.io/badge/zero--dependency-no%20build%20step-16A34A?style=for-the-badge" alt="Zero dependency" />
+</p>
+
+<br/>
+
+This project constrains AI tutoring into an **executable teaching protocol** that advances along a real project, and ships both as a DSH skill and as an installable plugin bundle.
+
+- **Outputs**: a staged route, task cards, an evidence file (`.coach/state.json` as the single source of truth, `.coach/PROGRESS.md` rendered from it), and stage acceptance verdicts.
+- **Enforced rules**: ability conclusions their evidence does not support are rejected; a stage cannot pass while blockers are open; subject-specific tokens must not appear in engine files.
+- **How it is enforced**: a zero-dependency validator checks these rules mechanically, not merely in prose.
+- **Subject decoupling**: the method lives in the engine, the knowledge lives in **swappable domain packs**; changing subject never requires touching the engine. The first pack ships in-repo: Unity / C#.
+
+> **Naming**: the repository, the npm package and the registered skill all share one name, `dsh-project-based-learning` (the model invokes it as `skill("dsh-project-based-learning")`).
+
+## 📐 Design stance
+
+| Teaching problem | Treatment in this project |
+|---|---|
+| Factual knowledge and skills treated alike | Factual knowledge is **taught directly** (concept → why the current task needs it → minimal example → the learner applies it once → one confirmation question); scaffolding and graded hints are **for skills only** |
+| Self-reports either believed blindly or dismissed | Three classes: an **ability** claim is a lead only; a **gap** claim is believed immediately and switches the coach into teaching; an **action** report is accepted as partially verified unless contradicted — no re-running demanded |
+| One correct answer treated as mastery | Conclusions are layered by type: knowledge requires **explaining the mechanism unprompted and transferring it to a new situation**; behaviour requires reproducible material |
+| Stage acceptance without criteria | A **three-tier verdict** (pass / conditional / fail) with a *reproducible + explainable + modifiable* sufficiency test |
+| AI-written work counted as ability | AI-assisted work counts only after the learner can **explain, modify and verify** it |
+
+## 🚫 What it is not
+
+- **Not a question bank or a drill tool**: diagnostic questions locate and grade; they are not the teaching content.
+- **Not a ghost-writing tool**: a full reference answer is given only on explicit request and never counts as evidence.
+- **Not an official DeepSeek plugin**: this is a third-party implementation.
+- **Not tied to a subject**: Unity / C# is the first domain pack shipped in-repo; it is not part of the engine.
+
+## 🚀 Quick start
+
+```bash
+dsh plugin --profile web add dsh-project-based-learning
+```
+
+Then enter teaching mode explicitly in a session and name the subject:
+
+> 教学模式：我完全没学过 <subject>，请评估我的水平
+
+Replace `<subject>` with what you want to learn. When the missing piece is factual knowledge, the coach teaches first (concept → why the current task needs it → minimal example → you apply it once → one confirmation question) and only then returns to the route — instead of opening with a set of diagnostic questions.
+
+The Unity / C# pack shipped in-repo is demonstrated in `skills/dsh-project-based-learning/references/domains/unity-csharp/example.md`.
+
+## 📦 Install
+
+### Option 1 — plugin bundle (recommended)
+
+```bash
+dsh plugin --profile web add dsh-project-based-learning     # or --profile headless, or your own profile
+dsh --profile web --dump-config                             # the dsh-project-based-learning layer should appear
+```
+
+Without npm, install straight from this repository:
+
+```bash
+dsh plugin --profile web add github:Kirisame1969/dsh-project-based-learning
+```
+
+The bundle layer (`cordis.patch.yml`) registers the packaged skill through `ctx.skills.register()`. The plugin only consumes the `skills` service — it imports nothing from the harness and brings no second copy of Cordis.
+
+Uninstall:
+
+```bash
+dsh plugin --profile web remove dsh-project-based-learning
+```
+
+Learning data lives in `.coach/` inside your workspace; uninstalling does not delete it.
+
+### Option 2 — skill files only
+
+Into any DSH skill root (project-scoped `.dsh/skills/`, or `$DSH_HOME/skills/` for every workspace):
+
+```bash
+npx -y -p dsh-project-based-learning coach-install --dest-root "$DSH_HOME/skills"
+```
+
+From a checkout, the same installer runs locally and supports `--dry-run`, `--link` and `--force`:
+
+```bash
+node skills/dsh-project-based-learning/scripts/coach-install.mjs --dry-run
+node skills/dsh-project-based-learning/scripts/coach-install.mjs --dest-root "$DSH_HOME/skills"
+```
+
+### Option 3 — no install
+
+Point your agent at `skills/dsh-project-based-learning/SKILL.md` and ask it to follow that file. Engine protocol, domain pack and scripts are plain files.
+
+## 💬 Commands
+
+| Command | Effect |
+|---|---|
+| `开始诊断` | Goal and experience intake, then a three-class minimum-coverage diagnostic |
+| `制定路线` | Create or adjust the staged route |
+| `本次任务：…` | One task loop (deliverable → attempt → hints → evidence) |
+| `给提示，级别 N` | Only the requested hint level (1–5) |
+| `审阅成果：…` | Review by severity, mechanism, impact, minimal fix, verification |
+| `验收阶段` | Three-tier verdict, item-by-item checklist, retrieval recap |
+| `复盘` | Capability delta, error patterns, next step |
+| `调整节奏` | Re-plan by time or difficulty |
+| `直接答案` | Full reference implementation (recorded as **not** evidence) |
+| `查看学习档案` / `更新学习档案` | Read / write the state file and validate it |
+| `切换或替换学科：<id>` | Switch domain pack; updates `state.domain` and `domainVersion`, keeps existing evidence |
+
+The command vocabulary is Chinese today; the engine prose is subject-neutral, and translations are a welcome contribution.
+
+## 🧩 How it works
+
+```mermaid
+flowchart LR
+    A["Confirm the goal"] --> B["Diagnose with real tasks"] --> C["Capability baseline"] --> D["Staged route"]
+    D --> E["Learner attempts"] --> F["Graded hints and review"] --> G{"Stage acceptance"}
+    G -->|pass| D
+    G -->|conditional or fail| E
+```
+
+```
+skills/dsh-project-based-learning/
+├── SKILL.md                            # engine: the loop, standing rules, command → file map
+├── references/engine/                  # 9 protocol files, loaded on demand
+├── references/domains/unity-csharp/    # the domain pack shipped in-repo (7 files)
+├── assets/                             # state template + task / review / acceptance templates
+└── scripts/                            # zero-dependency validator, regression selftest, installer
+```
+
+- **State**: `.coach/state.json` is the single source of truth; `.coach/PROGRESS.md` is rendered from it (never hand-edited).
+- **The validator** checks the state file, the domain-pack contract, and one **layering rule**: engine files must contain no subject-specific tokens. Run it after every accepted action:
+
+  ```bash
+  node skills/dsh-project-based-learning/scripts/coach-validate.mjs --state .coach/state.json --render
+  ```
+
+## 🎛️ Subjects and domain packs
+
+The engine is bound to no subject: it references a domain pack only by **section name and id**, and hard-codes no subject id. The active subject is the `domain` field in the state file.
+
+<details>
+<summary><b>The pack shipped in-repo: Unity / C#</b> (click to expand)</summary>
+
+| File | Content |
+|---|---|
+| `manifest.yml` | `id: unity-csharp`, version, engine compatibility range, per-task time window `[30, 90]` minutes |
+| `archetypes.md` | 6 project archetypes (stage split, minimum verifiable output, acceptance points, failure modes) |
+| `diagnosis-bank.md` | 16 diagnostic questions across 7 dimensions; each labelled with a **type** (factual / reasoning / mixed) and its **prerequisites**; factual items never open first contact |
+| `verification.md` | 5 executable check recipes (preconditions, command, expected output, meaning of failure, required sandbox mode) |
+| `pitfalls.md` | Grouped pitfalls with symptom, mechanism, minimal fix and the capability dimension each maps to |
+| `example.md` | One fully worked example: intake → capability profile → stages → review → acceptance verdict |
+| `glossary.md` | Bilingual terminology table |
+
+Commands marked「（未验证）」inside that pack must be confirmed on a machine with Unity Editor before they can support an acceptance verdict.
+
+</details>
+
+**Adding a subject pack** — two routes:
+
+1. **Let the AI generate it** — say this in a session:
+
+   > Following the contract in `references/engine/domain-contract.md`, generate a domain pack for "<subject>": create manifest.yml, archetypes.md, diagnosis-bank.md, verification.md, pitfalls.md, example.md and glossary.md under `references/domains/<id>/`; label diagnostic questions with type and prerequisites; validate with `coach-validate.mjs --domain-dir` when done.
+
+2. **Write it by hand** — fill the same seven contract files; steps are in [`CONTRIBUTING.md`](CONTRIBUTING.md).
+
+**Switching subject**: put the pack under `references/domains/<id>/`, then say `切换或替换学科：<id>`. The engine updates `state.domain` and `domainVersion` and records the reason in the route changes; **existing evidence and conclusions are kept** — only the route ahead and the question source change.
+
+Three boundaries, stated plainly:
+
+- When a pack lacks a section, the engine falls back to its generic behaviour and says so ("this pack does not provide X; using the generic approach this time").
+- The validator only checks the keys of `manifest.yml` and the presence and size of section files; it does **not** parse the question bank. Field requirements on questions are a manual review item, not a mechanical gate.
+- A domain pack must currently live under the skill's `references/domains/`; keeping packs in your workspace, or distributing them across repositories, is not supported yet.
+
+## 📋 Requirements
+
+- Node.js `^22.19.0 || >=24.0.0` (the scripts themselves need only `fs.cpSync`, available since 16.7; the plugin matrix matches the harness).
+- No npm dependencies, no build step: `lib/index.js` is hand-written source, not a build artifact.
+
+## 📚 Documentation
+
+- [`docs/installing.zh.md`](docs/installing.zh.md) — installation details (native DSH / DSH Desktop / each skill root)
+- [`CHANGELOG.md`](CHANGELOG.md) — version history
+- [`CONTRIBUTING.md`](CONTRIBUTING.md) — how to add a domain pack
+
+## 🤝 Contributing
+
+The most valuable contribution is a **new domain pack**; start with [`CONTRIBUTING.md`](CONTRIBUTING.md).
+
+## 📄 License
+
+[MIT](LICENSE).

package/README.md

@@ -1,120 +1,206 @@
-[简体中文](README.zh.md)
+<p align="center">
+  <strong>项目制学习教练：学员动手，教练跑闭环</strong><br/>
+  面向 <a href="https://github.com/deepseek-ai/deepseek-harness">DeepSeek Harness</a>（DSH）的教学技能与组合包 —— 教学法与学科内容分离，领域包可整体替换
+</p>
 
-# dsh-coach
+<p align="center">
+  <a href="README.md"><strong>简体中文</strong></a> · <a href="README.en.md">English</a>
+</p>
 
-**A project-based learning coach for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (DSH): the learner does the work, the coach runs the loop.**
+<p align="center">
+  <a href="https://www.npmjs.com/package/dsh-project-based-learning"><img src="https://img.shields.io/npm/v/dsh-project-based-learning?style=for-the-badge&logo=npm&label=npm&color=CB3837" alt="npm 版本" /></a>
+  <a href="https://github.com/Kirisame1969/dsh-project-based-learning/actions"><img src="https://img.shields.io/github/actions/workflow/status/Kirisame1969/dsh-project-based-learning/ci.yml?style=for-the-badge&logo=github&label=CI" alt="CI" /></a>
+  <a href="LICENSE"><img src="https://img.shields.io/github/license/Kirisame1969/dsh-project-based-learning?style=for-the-badge&color=blue" alt="许可证" /></a>
+  <a href="https://github.com/Kirisame1969/dsh-project-based-learning/stargazers"><img src="https://img.shields.io/github/stars/Kirisame1969/dsh-project-based-learning?style=for-the-badge&logo=github&color=yellow" alt="Star" /></a>
+</p>
 
-One tree, two forms: a **skill**, and an installable **plugin bundle**. The teaching engine is subject-agnostic; subject knowledge lives in swappable **domain packs** (one ships today: Unity / C#).
+<p align="center">
+  <a href="https://github.com/topics/dsh-plugin"><img src="https://img.shields.io/badge/DSH-插件-4B6BFB?style=for-the-badge" alt="DSH 插件" /></a>
+  <img src="https://img.shields.io/badge/零依赖-无构建步骤-16A34A?style=for-the-badge" alt="零依赖" />
+</p>
 
-It does not tutor by lecturing. It runs a loop: confirm the goal → diagnose with real tasks → build an evidence-backed capability baseline → set staged deliverables → let the learner attempt first → review with graded problems → accept or reject a stage on evidence.
+<br/>
 
-## Why it exists
+本项目把「AI 辅导」约束成一套**可执行的教学协议**，以真实项目为主线推进教学，并以 DSH 技能与可安装组合包两种形态分发。
 
-Most AI "tutoring" collapses into one of two failures: lecturing from zero, or handing over finished work. Both destroy learning. This skill encodes the opposite rules as a protocol the agent must follow:
+- **产出物**：分阶段路线、任务卡、证据档案（`.coach/state.json` 为唯一事实源，`.coach/PROGRESS.md` 为生成视图）、阶段验收结论。
+- **强制规则**：证据不支撑的能力结论被拒绝；存在未解决阻塞项时阶段不能判通过；学科专有词不得出现在引擎文件中。
+- **执行方式**：上述规则由零依赖校验器机械检查，不只写在提示词里。
+- **学科解耦**：教学法在引擎中，学科知识在**可整体替换的领域包**中；换学科不需要改引擎。随包提供第一个领域包：Unity / C#。
 
-- **The learner attempts first.** A complete answer is a last resort, not a first move.
-- **Hints are graded** (1 remind the goal → 5 full reference), and jumping to level 5 unasked is forbidden.
-- **Self-reports are graded, never taken on faith and never waved away.** An *ability* claim ("I'm fluent") is only a lead and never counts as verified by itself; a *gap* claim ("I never learned X") is believed immediately and switches the coach into **teaching** instead of quizzing; an *action* report ("I ran it, the output was X") is accepted as partially verified unless contradicted — the coach will not keep asking you to re-run it.
-- **Facts get taught; skills get practised.** When the missing piece is factual knowledge (an API name or signature, a call order, a language rule, a documented default), the coach explains it — mechanism, why your task needs it, a minimal example, then one confirmation question. Scaffolding and graded hints are for *skills* (writing code, debugging, designing), not for facts.
-- **AI-assisted work counts only once the learner can explain, modify and verify it.**
-- **Acceptance is three-tiered** (pass / conditional / fail), with a *reproducible + explainable + modifiable* test for evidence sufficiency.
+> **命名**：仓库名、npm 包名、技能注册名三者一致，均为 `dsh-project-based-learning`（模型侧通过 `skill("dsh-project-based-learning")` 调用）。
 
-These rules are not just prose: a zero-dependency validator enforces them mechanically on the learner's state file.
+## 📐 设计取向
 
-## Install
+| 教学问题 | 本项目的处理 |
+|---|---|
+| 事实性知识与技能混为一谈 | 事实性知识**直接讲授**（概念 → 为什么当前任务需要它 → 最小示例 → 学员应用一次 → 一道确认题）；脚手架与分级提示**只用于技能** |
+| 学员自述被一律采信或不采信 | 三类分别处理：**能力自述**仅作线索；**缺口自述**直接采信并转入讲授；**操作自述**无反证时按部分验证接受，不要求重复实测 |
+| 单题答对即认定掌握 | 结论按类型分层：知识类要求**无提示解释机制并迁移到新情境**；行为类要求可复现材料 |
+| 阶段验收缺少判据 | **三档结论**（通过／有条件通过／未通过），证据充分性判据为「可复现 + 可解释 + 可修改」 |
+| AI 代写被计入能力 | AI 参与的成果，须由学员**解释、修改、验证**后才计为能力证据 |
+
+## 🚫 它不是什么
 
-### Option 1 — plugin bundle (recommended)
+- **不是题库或刷题工具**：诊断题只用于定位与分级，不作为教学内容。
+- **不是代写工具**：完整参考答案仅在明确请求时给出，且不计入能力证据。
+- **不是官方 DeepSeek 插件**：本项目为第三方实现。
+- **不限定学科**：Unity / C# 是随包提供的第一个领域包，不属于引擎。
 
-Into any DSH profile:
+## 🚀 快速开始
 
 ```bash
-dsh plugin --profile web add dsh-project-based-learning    # or --profile headless, or your own profile
-dsh --profile web --dump-config                 # the dsh-project-based-learning layer should appear
+dsh plugin --profile web add dsh-project-based-learning
 ```
 
-Without npm, install straight from this repository:
+在会话中显式进入教学模式，并给出要学的主题：
+
+> 教学模式：我完全没学过 <主题>，请评估我的水平
+
+把 `<主题>` 换成你要学的内容即可。遇到事实性知识缺口时，教练会先讲授（概念 → 为什么当前任务需要它 → 最小示例 → 你应用一次 → 一道确认题），再回到路线安排，而不是先出一组诊断题。
+
+随包提供的 Unity / C# 领域包示范见 `skills/dsh-project-based-learning/references/domains/unity-csharp/example.md`。
+
+## 📦 安装
+
+### 方式一：组合包（推荐）
 
 ```bash
-dsh plugin --profile web add github:Kirisame1969/dsh-project-based-learning
+dsh plugin --profile web add dsh-project-based-learning     # 也可用 --profile headless 或你自己的 profile
+dsh --profile web --dump-config                             # 应出现 dsh-project-based-learning 层
 ```
 
-The bundle layer (`cordis.patch.yml`) registers the packaged skill through `ctx.skills.register()`. The plugin only consumes the `skills` service — it imports nothing from the harness and brings no second copy of Cordis.
+不经 npm，直接从本仓库安装：
 
-### Option 2 — skill files only
+```bash
+dsh plugin --profile web add github:Kirisame1969/dsh-project-based-learning
+```
 
-Into a DSH skill root (project-scoped `.dsh/skills/`, or `$DSH_HOME/skills/` for every workspace):
+组合包层（`cordis.patch.yml`）通过 `ctx.skills.register()` 注册随包分发的技能。插件只消费 `skills` 服务，不 import 任何 harness 包，也不会带进第二份 Cordis。
+
+卸载：
 
 ```bash
-npx -y -p dsh-project-based-learning coach-install --dest-root "$DSH_HOME/skills"
+dsh plugin --profile web remove dsh-project-based-learning
 ```
 
-From a checkout, the same installer runs locally and supports `--dry-run`, `--link` and `--force`:
+学习数据位于工作区的 `.coach/`，卸载技能不会删除它。
+
+### 方式二：只装技能文件
+
+装进任意 DSH 技能根（项目级 `.dsh/skills/`，或用户级 `$DSH_HOME/skills/`）：
 
 ```bash
-node skills/dsh-coach/scripts/coach-install.mjs --dry-run
-node skills/dsh-coach/scripts/coach-install.mjs --dest-root "$DSH_HOME/skills"
+npx -y -p dsh-project-based-learning coach-install --dest-root "$DSH_HOME/skills"
 ```
 
-### Option 3 — no install at all
-
-Point your agent at `skills/dsh-coach/SKILL.md` and ask it to follow that file. Everything (engine protocol, domain pack, scripts) is plain files.
+若已有本仓库检出，可直接运行同一个安装器，支持 `--dry-run`、`--link` 与 `--force`：
 
-## Using it
+```powershell
+node skills\dsh-project-based-learning\scripts\coach-install.mjs --dry-run
+node skills\dsh-project-based-learning\scripts\coach-install.mjs --dest-root "$env:DSH_HOME\skills"
+```
 
-Once loaded, drive it with natural-language commands:
+### 方式三：不安装
 
-| Command | What happens |
-|---|---|
-| `开始诊断` | Goal + experience intake, then a 3-task minimum diagnostic |
-| `制定路线` | Stage plan with deliverables, non-goals and acceptance criteria |
-| `本次任务：…` | One task loop (deliverable → attempt → hints → evidence) |
-| `给提示，级别 N` | Only the requested hint level (1–5) |
-| `审阅成果：…` | Review with severity + mechanism + impact + minimal fix + verification |
-| `验收阶段` | Three-tier verdict, `userOnly` checklist, retrieval recap |
-| `复盘` | Capability delta, error patterns, next step |
-| `调整节奏` | Re-plan for time/difficulty |
-| `直接答案` | Full reference answer (recorded as *not* evidence) |
-| `查看学习档案` / `更新学习档案` | Read / write the state file |
+将 agent 指向 `skills/dsh-project-based-learning/SKILL.md`，令其按该文件执行。引擎协议、领域包与脚本均为普通文件。
 
-The command vocabulary is Chinese today. The engine prose is language-neutral — translating it is a welcome contribution.
+## 💬 指令手册
 
-## How it works
+| 指令 | 作用 |
+|---|---|
+| `开始诊断` | 收集目标与经验，做三类最小覆盖诊断 |
+| `制定路线` | 生成 / 调整阶段路线 |
+| `本次任务：…` | 进入单次任务循环（交付物 → 尝试 → 提示 → 证据） |
+| `给提示，级别 N` | 只给指定级别的提示（1–5） |
+| `审阅成果：…` | 按严重度、机制、影响、最小修法、验证方式审阅 |
+| `验收阶段` | 三档结论、逐项核对、检索式复述 |
+| `复盘` | 能力变化 / 错误模式 / 下一步 |
+| `调整节奏` | 按时间或难度调整路线 |
+| `直接答案` | 给出完整参考实现（记录为**不计**能力证据） |
+| `查看学习档案` / `更新学习档案` | 读 / 写状态并校验 |
+| `切换或替换学科：<id>` | 更换领域包，更新 `state.domain` 与 `domainVersion`，保留既有证据 |
+
+指令词目前为中文；引擎正文与学科无关，翻译属于受欢迎的贡献。
+
+## 🧩 工作原理
+
+```mermaid
+flowchart LR
+    A["确认目标"] --> B["真实任务诊断"] --> C["能力基线"] --> D["阶段路线"]
+    D --> E["学员先动手"] --> F["分级提示与审阅"] --> G{"阶段验收"}
+    G -->|通过| D
+    G -->|有条件或未通过| E
+```
 
 ```
-skills/dsh-coach/
-├── SKILL.md                  # engine: the loop, the standing rules, command → file map
-├── references/engine/        # 9 protocol files, loaded on demand
-├── references/domains/unity-csharp/   # the swappable subject pack (7 files)
-├── assets/                   # state template + task/review/acceptance templates
-└── scripts/                  # zero-dependency validator, selftest, installer
+skills/dsh-project-based-learning/
+├── SKILL.md                            # 引擎：闭环、常驻规则、指令 → 必读文件映射
+├── references/engine/                  # 9 个协议文件，按需加载
+├── references/domains/unity-csharp/    # 随包提供的领域包（7 个文件）
+├── assets/                             # 状态模板 + 任务卡 / 审阅 / 验收模板
+└── scripts/                            # 零依赖校验器、回归自测、安装器
 ```
 
-- **State** lives in `.coach/state.json` — a single source of truth — with `.coach/PROGRESS.md` rendered from it (never hand-edited).
-- **The validator** checks the state file, the domain-pack contract, and a **layering rule**: no subject-specific tokens may appear in engine files. Run it after every accepted action:
+- **状态**：`.coach/state.json` 为唯一事实源，`.coach/PROGRESS.md` 由其渲染（禁止手工编辑）。
+- **校验器**检查状态文件、领域包契约，以及一条**分层规则**：引擎文件中不得出现学科专有词条。每次可验收动作后运行：
 
   ```bash
-  node skills/dsh-coach/scripts/coach-validate.mjs --state .coach/state.json --render
+  node skills/dsh-project-based-learning/scripts/coach-validate.mjs --state .coach/state.json --render
   ```
 
-## Adding another subject
+## 🎛️ 学科与领域包
+
+引擎不绑定任何学科：它只用**小节名与 id** 引用领域包，不硬编码学科 id。当前学科由状态文件的 `domain` 字段决定。
+
+<details>
+<summary><b>随包提供的领域包：Unity / C#</b>（点击展开）</summary>
+
+| 文件 | 内容 |
+|---|---|
+| `manifest.yml` | `id: unity-csharp`、版本、引擎兼容范围、单次任务时长区间 `[30, 90]` 分钟 |
+| `archetypes.md` | 6 个项目原型（阶段切分、最小可验证成果、验收要点、失败模式） |
+| `diagnosis-bank.md` | 16 道诊断题，覆盖 7 个维度；每题标注**类型**（事实性／推理性／综合）与**前置知识**，事实性题不作首次接触题 |
+| `verification.md` | 5 条可执行核对配方（前置条件、命令、期望输出、失败含义、所需沙箱模式） |
+| `pitfalls.md` | 陷阱分组的症状、机制、最小修复与对应能力维度 |
+| `example.md` | 一份填好的完整示例：intake → 能力画像 → 阶段 → 审阅 → 验收结论 |
+| `glossary.md` | 中英对照术语表 |
+
+该包内标注「（未验证）」的命令，需在装有 Unity Editor 的环境实测确认后才可用于验收判定。
+
+</details>
+
+**新增一个学科包**，两条路径：
+
+1. **让 AI 生成**——在会话中直接说：
+
+   > 按 `references/engine/domain-contract.md` 的契约，为「<学科>」生成领域包：在 `references/domains/<id>/` 下建立 manifest.yml、archetypes.md、diagnosis-bank.md、verification.md、pitfalls.md、example.md、glossary.md 七个文件；诊断题标注类型与前置知识；生成后用 `coach-validate.mjs --domain-dir` 校验。
+
+2. **手工编写**——按同一份契约补齐七个文件，步骤见 [`CONTRIBUTING.md`](CONTRIBUTING.md)。
+
+**切换学科**：把包放入 `references/domains/<id>/`，然后说 `切换或替换学科：<id>`。引擎会更新 `state.domain` 与 `domainVersion` 并在路线变更中记录原因；**已完成的能力证据与结论保留**，只影响后续路线与题目来源。
+
+三点边界（如实声明）：
 
-The engine is subject-agnostic; adding a subject never requires touching it. Create `references/domains/<new-id>/` with the seven contract files and fill in `manifest.yml` — see [`CONTRIBUTING.md`](CONTRIBUTING.md) and `skills/dsh-coach/references/engine/domain-contract.md`.
+- 领域包缺某个小节时，引擎回退到通用行为，并在回复中说明「该学科包未提供 X，本次按通用做法处理」。
+- 校验器只检查 `manifest.yml` 的键、小节文件是否存在与大小，**不解析题库内容**；题目的字段要求属于人工核对项，不是机械门禁。
+- 领域包目前必须位于技能目录的 `references/domains/` 下；把领域包放在工作区、或跨仓库分发领域包，尚无支持。
 
-## Requirements
+## 📋 环境要求
 
-- Node.js `^22.19.0 || >=24.0.0` (the scripts themselves only need `fs.cpSync`, available since 16.7; the plugin matrix matches the harness).
-- No npm dependencies, no build step. `lib/index.js` is hand-written source, not a build artifact.
+- Node.js `^22.19.0 || >=24.0.0`（脚本本身仅需 ≥ 16.7 的 `fs.cpSync`；插件矩阵与 harness 对齐）。
+- 无 npm 依赖、无构建步骤：`lib/index.js` 为手写来源，不是构建产物。
 
-## Documentation
+## 📚 文档
 
-- [`docs/installing.zh.md`](docs/installing.zh.md) — installation details (native DSH / DSH Desktop / each skill root)
-- [`CHANGELOG.md`](CHANGELOG.md) — version history
-- [`CONTRIBUTING.md`](CONTRIBUTING.md) — how to add a domain pack
+- [`docs/installing.zh.md`](docs/installing.zh.md) —— 安装细则（原生 DSH / DSH Desktop / 各技能根）
+- [`CHANGELOG.md`](CHANGELOG.md) —— 版本变更
+- [`CONTRIBUTING.md`](CONTRIBUTING.md) —— 新增领域包的方法
 
-## Contributing
+## 🤝 贡献
 
-The most valuable contribution is a **new domain pack**. Start with [`CONTRIBUTING.md`](CONTRIBUTING.md).
+最有价值的贡献是**新增一个学科领域包**，请先阅读 [`CONTRIBUTING.md`](CONTRIBUTING.md)。
 
-## License
+## 📄 许可
 
-[MIT](LICENSE).
+[MIT](LICENSE)。

package/cordis.patch.yml

@@ -1,4 +1,4 @@
-# dsh-coach 组合包层：把随包分发的教练技能注册为一个 DSH 运行时技能。
+# dsh-project-based-learning 组合包层：把随包分发的教练技能注册为一个 DSH 运行时技能。
 #
 # 本文件是 `package.json#dsh.bundle.patch` 指向的层。层语义：一个 YAML 数组的
 # 行动词（row verbs）；`insert` 向组合后的配置加入插件行，后续层可按 `id`
@@ -11,5 +11,5 @@
 # 前提：目标 profile 已挂载 `@deepseek-ai/dsh-skill`（`@deepseek-ai/dsh-base` 已包含）。
 # 本包不贡献任何工具，也不改动既有行。
 - insert:
-    - id: dsh-coach
+    - id: dsh-project-based-learning
       name: dsh-project-based-learning

package/docs/DESIGN-AUDIT.md

@@ -1,5 +1,7 @@
 # 教学教练插件：弱点分析与改动审核（第 1–2 轮）
 
+> **勘误（2026-09-12）**：本文记录的是 1.0.0 / 1.1.0 时期的实况。技能目录已于 **1.2.0** 由 `skills/dsh-coach/` 改名为 `skills/dsh-project-based-learning/`，技能注册名同步变更；下文旧路径请按此对应，其余内容未改写。
+
 对象：`通用AI项目制教学工作流.md`（379 行，14 节，下称"原文"）
 产物：可安装的 DSH 技能包 `coach/`（引擎 + Unity/C# 领域包 + 状态契约 + 校验器）
 本文性质：**改动前置审核记录**。任何对原文的改动都必须先通过下方六项检验，且必须经过第二轮反方复审；未通过者驳回或延后，不以"改进"名义直接下手。

package/docs/installing.zh.md

@@ -1,6 +1,6 @@
 # 安装细则
 
-本文覆盖 dsh-coach 的全部安装路径（技能目录方式与组合包方式）。
+本文覆盖 `dsh-project-based-learning`（注册技能名 `dsh-project-based-learning`）的全部安装路径：技能目录方式与组合包方式。
 
 ## 先选路径
 
@@ -29,14 +29,14 @@
 ### A1 项目级
 
 ```powershell
-# 在本仓库根执行；默认目标是 <cwd>/.dsh/skills/dsh-coach
-node skills/dsh-coach/scripts/coach-install.mjs
+# 在本仓库根执行；默认目标是 <cwd>/.dsh/skills/dsh-project-based-learning
+node skills/dsh-project-based-learning/scripts/coach-install.mjs
 
 # 先看会做什么（不写盘）
-node skills/dsh-coach/scripts/coach-install.mjs --dry-run
+node skills/dsh-project-based-learning/scripts/coach-install.mjs --dry-run
 
 # 用目录联接安装：源文件改动即时生效，适合边改边用
-node skills/dsh-coach/scripts/coach-install.mjs --link
+node skills/dsh-project-based-learning/scripts/coach-install.mjs --link
 ```
 
 `--link` 在 Windows 上建立**目录联接（Junction）**，不需要管理员权限；其它平台建立目录符号链接。**分发时不要用 `--link`**——对方删掉你的源目录，技能就失效了。
@@ -44,7 +44,7 @@ node skills/dsh-coach/scripts/coach-install.mjs --link
 ### A2 用户级（跨工作区）
 
 ```powershell
-node skills/dsh-coach/scripts/coach-install.mjs --dest-root "$env:DSH_HOME\skills"
+node skills/dsh-project-based-learning/scripts/coach-install.mjs --dest-root "$env:DSH_HOME\skills"
 # 原生 DSH 若未设置 DSH_HOME，其默认 home 为 ~/.dsh
 ```
 
@@ -53,13 +53,13 @@ node skills/dsh-coach/scripts/coach-install.mjs --dest-root "$env:DSH_HOME\skill
 任何被扫描的根都可以，只要目录名是技能名：
 
 ```powershell
-node skills/dsh-coach/scripts/coach-install.mjs --dest-root "<项目根>\.agents\skills"
+node skills/dsh-project-based-learning/scripts/coach-install.mjs --dest-root "<项目根>\.agents\skills"
 ```
 
 ## B. 组合包（插件）安装
 
 ```bash
-dsh plugin --profile <profile> add /path/to/dsh-coach                     # 本地检出
+dsh plugin --profile <profile> add /path/to/dsh-project-based-learning                     # 本地检出
 dsh plugin --profile <profile> add dsh-project-based-learning             # 从 npm
 dsh plugin --profile <profile> add github:Kirisame1969/dsh-project-based-learning   # 不经 npm，直接从仓库
 dsh --profile <profile> --dump-config                     # 应出现 dsh-project-based-learning 层
@@ -73,23 +73,23 @@ dsh plugin --profile <profile> remove dsh-project-based-learning          # 卸
 
 ## C. DSH Desktop
 
-- 桌面端有自己的插件界面（市场插件 `dshmarket`，站点 <https://dshmarket.com>，仓库 [dsh-market/dsh-market](https://github.com/dsh-market/dsh-market)）。若 dsh-coach 已被收录，可直接在界面里一键安装。
+- 桌面端有自己的插件界面（市场插件 `dshmarket`，站点 <https://dshmarket.com>，仓库 [dsh-market/dsh-market](https://github.com/dsh-market/dsh-market)）。若 dsh-project-based-learning 已被收录，可直接在界面里一键安装。
 - 尚未收录时，用 A1/A2 的技能目录方式即可——桌面端同样扫描这些技能根。
 - 桌面端把插件装在 `$DSH_HOME/profiles/<profile>/` 下的"代际"目录中，并通过 profile 的 `package.json` 依赖与 pnpm override 指向它。
 
 ## D. 不安装
 
-让 agent 直接读本仓库的 `skills/dsh-coach/SKILL.md` 并按它执行。适合评估与调试。
+让 agent 直接读本仓库的 `skills/dsh-project-based-learning/SKILL.md` 并按它执行。适合评估与调试。
 
 ## 怎么确认装好了
 
-1. **会话技能目录**里出现 `dsh-coach`（模型会在下一步看到它）；
-2. 让模型调用一次 `skill("dsh-coach")`，应返回技能正文，并在资源提示里给出技能目录的绝对路径；
-3. 目录检查：`<技能根>/dsh-coach/SKILL.md` 存在，且 `references/domains/unity-csharp/` 下有 7 个文件。
+1. **会话技能目录**里出现 `dsh-project-based-learning`（模型会在下一步看到它）；
+2. 让模型调用一次 `skill("dsh-project-based-learning")`，应返回技能正文，并在资源提示里给出技能目录的绝对路径；
+3. 目录检查：`<技能根>/dsh-project-based-learning/SKILL.md` 存在，且 `references/domains/unity-csharp/` 下有 7 个文件。
 
 ## 卸载
 
-- 纯技能：删除 `<技能根>/dsh-coach` 目录（联接方式则删除联接本身，源目录不受影响）。
+- 纯技能：删除 `<技能根>/dsh-project-based-learning` 目录（联接方式则删除联接本身，源目录不受影响）。
 - 组合包：`dsh plugin --profile <profile> remove dsh-project-based-learning`。
 - 学习数据在**你自己的工作区** `.coach/` 下，与安装无关；卸载技能不会删除它。
 

package/docs/releasing.zh.md

@@ -8,8 +8,8 @@
 
 ```bash
 node test/entry.smoke.mjs                                   # 入口契约 + 资源齐全 → PASS
-node skills/dsh-coach/scripts/coach-selftest.mjs            # 校验器回归 → 9/9 PASS
-node skills/dsh-coach/scripts/coach-validate.mjs --state examples/state.demo.json
+node skills/dsh-project-based-learning/scripts/coach-selftest.mjs            # 校验器回归 → 9/9 PASS
+node skills/dsh-project-based-learning/scripts/coach-validate.mjs --state examples/state.demo.json
 dsh-plugin-dev check                                        # 社区静态检查 → ok=true（9 通过 / 0 失败 / 2 警告）
 ```
 
@@ -23,9 +23,10 @@ dsh-plugin-dev check                                        # 社区静态检查
 **可用通道：HTTPS + gh token**。`git push` 的 dry-run 已到达 GitHub 并完成鉴权（仅因当时尚无共同历史而被拒 non-fast-forward）：
 
 ```powershell
+# 在仓库根目录执行（下称 <仓库根>）
 $t = gh auth token
 $b64 = [Convert]::ToBase64String([Text.Encoding]::ASCII.GetBytes("x-access-token:$t"))
-git -C coach -c http.extraheader="AUTHORIZATION: basic $b64" `
+git -C "<仓库根>" -c http.extraheader="AUTHORIZATION: basic $b64" `
     push https://github.com/Kirisame1969/dsh-project-based-learning.git main
 ```
 
@@ -49,7 +50,7 @@ dsh plugin --profile web add dsh-project-based-learning
 
 ## 3. 提交到社区列表（"被搜到"的关键）
 
-**已确认的提交形态**：向 [awesome-dsh-plugin/awesome-dsh-plugin](https://github.com/awesome-dsh-plugin/awesome-dsh-plugin) 提 PR，新增一个目录条目文件 `data/plugins/<owner>__<repo>.yml`（本地草稿在 `.publish/awesome-submission/`）。标题遵循仓库既有先例（已合并的 [PR #3405](https://github.com/awesome-dsh-plugin/awesome-dsh-plugin/pull/3405)）：
+**已确认的提交形态**：向 [awesome-dsh-plugin/awesome-dsh-plugin](https://github.com/awesome-dsh-plugin/awesome-dsh-plugin) 提 PR，新增一个目录条目文件 `data/plugins/<owner>__<repo>.yml`。标题遵循仓库既有先例（已合并的 [PR #3405](https://github.com/awesome-dsh-plugin/awesome-dsh-plugin/pull/3405)）：
 
 ```
 feat(catalog): add dsh-project-based-learning by Kirisame1969

package/docs/zero-knowledge-path.zh.md

@@ -42,7 +42,7 @@
 5. **Z6／Z7 判据**：跑校验器，看是否有 `ST-W7` 提醒：
 
    ```bash
-   node skills/dsh-coach/scripts/coach-validate.mjs --state .coach/state.json
+   node skills/dsh-project-based-learning/scripts/coach-validate.mjs --state .coach/state.json
    ```
 
    > `ST-W7` 是 **warn，不是门禁**（CI 退出码只看 error）。这是 2.1 修订的明确取舍：漏报的代价是"可能漂移"（有本清单兜底），误报的代价是每次正常教学都被阻塞。

package/lib/index.js

@@ -1,7 +1,7 @@
-// dsh-coach bundle entry point.
+// dsh-project-based-learning bundle entry point.
 //
 // 把随包分发的教练技能注册为一个 DSH 运行时技能。技能正文是本仓库的
-// `skills/dsh-coach/SKILL.md`；它的相对引用（`references/`、`assets/`、`scripts/`）
+// `skills/dsh-project-based-learning/SKILL.md`；它的相对引用（`references/`、`assets/`、`scripts/`）
 // 通过目录型 resourceBase 解析到技能目录，因此 agent 只在需要时才读取引擎协议、
 // 领域包与模板（progressive disclosure）。
 //
@@ -12,13 +12,13 @@ import { readFileSync } from 'node:fs'
 import { dirname, join } from 'node:path'
 import { fileURLToPath } from 'node:url'
 
-export const name = 'dsh-coach'
+export const name = 'dsh-project-based-learning'
 
 /** 依赖 `skills` 服务：缺失该服务的组合不会加载本插件。 */
 export const inject = ['skills']
 
 const packageRoot = dirname(dirname(fileURLToPath(import.meta.url)))
-const skillRoot = join(packageRoot, 'skills', 'dsh-coach')
+const skillRoot = join(packageRoot, 'skills', 'dsh-project-based-learning')
 
 const FALLBACK_DESCRIPTION =
   '项目制学习教练：以真实项目为主线，通过诊断、分级提示、证据审阅与阶段验收，引导学员独立完成成果。仅在用户明确要求教学模式时启用。'
@@ -54,7 +54,7 @@ export function apply(ctx) {
 
   ctx.effect(() =>
     ctx.skills.register({
-      name: 'dsh-coach',
+      name: 'dsh-project-based-learning',
       source: 'bundled',
       description: summary ?? FALLBACK_DESCRIPTION,
       content: body,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-project-based-learning",
-  "version": "1.1.0",
+  "version": "2.0.0",
   "description": "项目制学习教练：学科无关的教学引擎 + 可替换领域包（随包提供 Unity/C# 特化）。以 DSH 技能与组合包双形态分发。",
   "type": "module",
   "main": "lib/index.js",
@@ -10,9 +10,9 @@
     "./package.json": "./package.json"
   },
   "bin": {
-    "coach-validate": "./skills/dsh-coach/scripts/coach-validate.mjs",
-    "coach-selftest": "./skills/dsh-coach/scripts/coach-selftest.mjs",
-    "coach-install": "./skills/dsh-coach/scripts/coach-install.mjs"
+    "coach-validate": "./skills/dsh-project-based-learning/scripts/coach-validate.mjs",
+    "coach-selftest": "./skills/dsh-project-based-learning/scripts/coach-selftest.mjs",
+    "coach-install": "./skills/dsh-project-based-learning/scripts/coach-install.mjs"
   },
   "files": [
     "lib",
@@ -22,7 +22,7 @@
     "examples",
     "LICENSE",
     "README.md",
-    "README.zh.md",
+    "README.en.md",
     "CHANGELOG.md",
     "CONTRIBUTING.md"
   ],
@@ -71,7 +71,7 @@
     }
   },
   "scripts": {
-    "validate": "node skills/dsh-coach/scripts/coach-validate.mjs --state examples/state.demo.json",
-    "selftest": "node skills/dsh-coach/scripts/coach-selftest.mjs"
+    "validate": "node skills/dsh-project-based-learning/scripts/coach-validate.mjs --state examples/state.demo.json",
+    "selftest": "node skills/dsh-project-based-learning/scripts/coach-selftest.mjs"
   }
 }

package/skills/{dsh-coach → dsh-project-based-learning}/SKILL.md

@@ -1,5 +1,5 @@
 ---
-name: dsh-coach
+name: dsh-project-based-learning
 description: 项目制学习教练。以真实项目为主线，通过诊断、分级提示、证据审阅与阶段验收，引导学员独立完成成果。仅在用户明确要求教学模式时启用。
 whenToUse: 用户明确要求"教学模式/带我学/教我/辅导我做项目/验收我的成果/复盘学习进度"，或提到学习路线、能力诊断、阶段验收。普通答疑、直接要代码、与学习无关的任务不要启用。
 user-invocable: true

package/skills/{dsh-coach → dsh-project-based-learning}/scripts/coach-install.mjs

@@ -5,7 +5,7 @@
  * 用法：
  *   node scripts/coach-install.mjs [--dest-root <dir>] [--link] [--force] [--dry-run]
  *
- * 默认目标： <cwd>/.dsh/skills/dsh-coach
+ * 默认目标： <cwd>/.dsh/skills/dsh-project-based-learning
  *   `.dsh/skills` 是 DSH 的 project-dsh 技能根（rank 100），按官方说明新增 skill
  *   会在下一次模型步骤进入会话目录，无需重启宿主。
  *

package/README.zh.md

@@ -1,118 +0,0 @@
-[English](README.md)
-
-# dsh-coach
-
-**面向 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness)（DSH）的项目制学习教练：学员动手，教练跑闭环。**
-
-同一棵树同时提供两种形态：**技能（skill）**，以及可安装的**组合包（plugin bundle）**。教学引擎与学科内容分离——学科知识住在可整体替换的**领域包**里（当前提供 Unity / C#）。
-
-它不靠讲课来"教"。它跑一条闭环：确认目标 → 用真实任务做诊断 → 建立有证据支撑的能力基线 → 设定分阶段可交付成果 → 让学员先动手 → 用分级问题审阅 → 依据证据判阶段通过与否。
-
-## 为什么做这个
-
-AI"辅导"通常掉进两个坑：从零开始照本宣科，或者直接替人把活干完。两者都毁掉学习。本项目把相反的规则写成了 agent 必须遵守的协议：
-
-- **学员先尝试。** 完整答案是最后手段，不是第一步。
-- **提示分级**（1 提醒目标 → 5 完整参考），未经请求跳到第 5 级是禁止的。
-- **自述按三类区别对待**，既不一律采信、也不一律不采信：**能力自述**（"我熟练"）只是线索，不能单独作为已验证；**缺口自述**（"我没学过 X"）**直接采信并转入讲授**，而不是先考一遍；**操作自述**（"我跑了一次，输出是 X"）无反证时按部分验证接受——教练不会再要求你重跑一遍，也不会再要你先补截图。
-- **知识直接教，技能才靠练。** 缺的是事实性知识（API 名称与签名、调用顺序、语言规则、文档已明的默认值）时，教练会**讲清楚**：概念 → 为什么你的任务需要它 → 最小示例 → 一道确认题。脚手架与分级提示用于**技能**（写代码、排错、设计），不用于事实。
-- **AI 参与的成果，只有在学员能解释、修改、验证之后**才算作他的能力证据。
-- **验收三档**（通过 / 有条件通过 / 未通过），证据充分性的判据是"可复现 + 可解释 + 可修改"。
-
-这些规则不只是文字：一个**零依赖校验器**会在学员的状态文件上机械地执行它们。
-
-## 安装
-
-### 方式一：组合包（推荐）
-
-装进任意 DSH profile：
-
-```bash
-dsh plugin --profile web add dsh-project-based-learning    # 也可用 --profile headless 或你自己的 profile
-dsh --profile web --dump-config                 # 应出现 dsh-project-based-learning 层
-```
-
-不经 npm，直接从本仓库安装：
-
-```bash
-dsh plugin --profile web add github:Kirisame1969/dsh-project-based-learning
-```
-
-组合包层（`cordis.patch.yml`）通过 `ctx.skills.register()` 注册随包分发的技能。插件只消费 `skills` 服务，不 import 任何 harness 包，也不会带进第二份 Cordis。
-
-### 方式二：只装技能文件
-
-装进任意 DSH 技能根（项目级 `.dsh/skills/`，或用户级 `$DSH_HOME/skills/`）：
-
-```bash
-npx -y -p dsh-project-based-learning coach-install --dest-root "$DSH_HOME/skills"
-```
-
-若已有本仓库检出，可直接运行同一个安装器，支持 `--dry-run`、`--link` 与 `--force`：
-
-```powershell
-node skills\dsh-coach\scripts\coach-install.mjs --dry-run
-node skills\dsh-coach\scripts\coach-install.mjs --dest-root "$env:DSH_HOME\skills"
-```
-
-### 方式三：完全不安装
-
-把 agent 指向 `skills/dsh-coach/SKILL.md`，让它按该文件执行。所有内容（引擎协议、领域包、脚本）都是普通文件。
-
-## 用法
-
-加载后用自然语言指令驱动：
-
-| 指令 | 作用 |
-|---|---|
-| `开始诊断` | 收集目标与经验，做三类最小覆盖诊断 |
-| `制定路线` | 生成/调整阶段路线 |
-| `本次任务：…` | 进入单次任务循环 |
-| `给提示，级别 N` | 只给指定级别的提示 |
-| `审阅成果：…` | 按五要素 + 依据 + 核对状态审阅 |
-| `验收阶段` | 三档结论 + `userOnly` 逐项核对 + 检索式复述 |
-| `复盘` | 能力变化 / 错误模式 / 下一步 |
-| `调整节奏` | 按时间或难度调整路线 |
-| `直接答案` | 给完整参考实现（记录为**不计**能力证据） |
-| `查看学习档案` / `更新学习档案` | 读/写状态并校验 |
-
-## 工作原理
-
-```
-skills/dsh-coach/
-├── SKILL.md                  # 引擎：闭环、常驻规则、指令→必读文件映射
-├── references/engine/        # 9 个协议文件，按需加载
-├── references/domains/unity-csharp/   # 可替换的学科包（7 个文件）
-├── assets/                   # 状态模板 + 任务卡/审阅/验收模板
-└── scripts/                  # 零依赖校验器、回归自测、安装器
-```
-
-- **状态**在 `.coach/state.json`——唯一事实源；`.coach/PROGRESS.md` 由它渲染而来（**禁止手工编辑**）。
-- **校验器**检查状态文件、领域包契约，以及一条**分层规则**：引擎文件中不得出现学科专有词条。每次可验收动作后运行：
-
-  ```bash
-  node skills/dsh-coach/scripts/coach-validate.mjs --state .coach/state.json --render
-  ```
-
-## 替换学科特化
-
-引擎学科无关；换学科**不需要改引擎**。新增 `references/domains/<new-id>/` 七个文件并按契约填 `manifest.yml` 即可，步骤见 [`CONTRIBUTING.md`](CONTRIBUTING.md) 与 `skills/dsh-coach/references/engine/domain-contract.md`。
-
-## 环境要求
-
-- Node.js `^22.19.0 || >=24.0.0`（脚本本身只需 ≥ 16.7 的 `fs.cpSync`；插件矩阵与 harness 对齐）。
-- 无 npm 依赖、无构建步骤。`lib/index.js` 是手写来源，不是构建产物。
-
-## 文档
-
-- [`docs/installing.zh.md`](docs/installing.zh.md) —— 安装细则（原生 DSH / DSH Desktop / 各技能根）
-- [`CHANGELOG.md`](CHANGELOG.md) —— 版本变更
-- [`CONTRIBUTING.md`](CONTRIBUTING.md) —— 如何新增领域包
-
-## 贡献
-
-最受欢迎的贡献是**新增一个学科领域包**。请先读 [`CONTRIBUTING.md`](CONTRIBUTING.md)。
-
-## 许可
-
-[MIT](LICENSE)。