nano-duoshe 0.1.0-alpha.2

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 guangdino
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,282 @@
+# nano-duoshe
+
+> Local-first project memory for AI coding agents (Claude Code / Codex / Cursor).
+>
+> Keep project conventions, decisions, gotchas, and module boundaries next to your code — so the agent doesn't start from zero every session.
+
+**English** | [简体中文](./README.zh-CN.md)
+
+## Why
+
+AI coding agents are powerful, but every new session hits the same cold-start problem:
+
+- the agent doesn't remember your project structure;
+- you re-explain architecture, conventions, and "don't touch this" rules every time;
+- it re-suggests fixes you already rejected;
+- it re-opens bugs you already solved;
+- `CLAUDE.md` / `AGENTS.md` help, but they're flat hand-edited files — no review, no provenance, no search.
+
+DuoShe drops a small local memory layer into each repo: long-term memory as Markdown, candidate memories that go through a review queue before becoming permanent, and SQLite FTS5 for search. Local, fast, rebuildable.
+
+---
+
+## 30-second start
+
+```bash
+npm install -g nano-duoshe        # install
+cd your-project
+duoshe init --guided              # init + 3 core questions + smart skill suggestion
+```
+
+That's it. After it finishes you'll have three daily commands:
+
+```bash
+duoshe remember "..."   # capture something you don't want to forget
+duoshe review           # see pending candidates, save or drop
+duoshe search "..."     # find what you wrote before
+```
+
+---
+
+## What it creates
+
+```text
+your-project/
+├─ .duoshe/                 # project memory vault (some public, some private)
+│  ├─ PROJECT.md            # ✅ overview, stack, conventions (commit)
+│  ├─ CODEMAP.md            # ✅ code map, entry points, routing notes (commit)
+│  ├─ DECISIONS.md          # ✅ architecture decisions and rationale (commit)
+│  ├─ TROUBLESHOOTING.md    # ✅ known issues and fixes (commit)
+│  ├─ MODULES.md            # ✅ module boundaries — especially "does NOT own" (commit)
+│  ├─ TODO.md               # ✅ current and upcoming work (commit)
+│  ├─ SETUP.md              # ✅ team onboarding guide (commit)
+│  ├─ SKILLS/               # ✅ enabled skill manifests (commit)
+│  ├─ config.json           # ❌ local config (personal preference, don't commit)
+│  ├─ CANDIDATES/           # ❌ pending drafts the user hasn't decided on (don't commit)
+│  └─ index.db              # ❌ SQLite FTS5 index, rebuildable (don't commit)
+├─ AGENTS.md                # auto-appended DuoShe block pointing Codex at .duoshe/
+├─ CLAUDE.md                # auto-appended DuoShe block pointing Claude Code at .duoshe/
+└─ .gitignore               # auto-maintained block for local state
+```
+
+When run inside a git repo, `duoshe init` **auto-maintains `.gitignore`** so private state stays local. Public memory (`PROJECT.md` etc.) you just `git add` / `git commit` normally.
+
+The core rule: **agents can propose memory; humans decide what becomes long-term knowledge.**
+
+---
+
+## Two ways to use it
+
+### 1. First time (use --guided)
+
+```bash
+duoshe init --guided
+```
+
+This:
+
+1. scans the project and generates draft templates;
+2. detects the project profile (embedded / algo / site-maintenance / AI-app / general / kid);
+3. installs the bundled skills (disabled by default);
+4. syncs `CLAUDE.md` / `AGENTS.md`;
+5. asks 3 core questions (what's the project, rules the AI must follow, zones not to touch);
+6. if it detected an embedded / algo / WordPress project, offers to enable the matching skill.
+
+Takes about a minute end-to-end.
+
+### 2. Daily use (3 commands)
+
+```bash
+duoshe remember "DB access must go through the service layer, not direct from routes"
+duoshe review              # list pending, then  save <id>  or  drop <id>
+duoshe search "service"    # find prior notes
+```
+
+---
+
+## Skills (opt-in domain support)
+
+DuoShe ships several optional domain skills, **disabled by default**:
+
+| Skill | When it helps |
+|---|---|
+| `embedded` | C firmware / FPGA / PLC (STM32, ESP32, Vivado, Codesys, TwinCAT) |
+| `matlab` | Algorithms / control engineering (MATLAB / Simulink / heavy math) |
+| `devops` | Infrastructure / IaC (Terraform, Ansible, Kubernetes) |
+| `wordpress` | WordPress site maintenance (also fine for non-developers) |
+| `graph` | Import-graph analysis, cycle detection, hot modules |
+
+```bash
+duoshe skill list                # list everything that's installed
+duoshe skill enable embedded     # turn one on (prints a short readme intro)
+duoshe rescan                    # re-scan to apply the new detectors / dir labels
+duoshe skill disable embedded    # turn it off
+```
+
+Once enabled, `rescan` uses the skill's extra detectors and directory labels to enrich project memory.
+
+---
+
+## Profile (project type)
+
+`duoshe init` auto-detects a project profile, which controls the first-step hints and which skills get suggested.
+
+```bash
+duoshe profile show              # current profile + what a fresh scan would guess now
+duoshe profile list              # all available profiles
+duoshe profile set embedded      # set manually — won't get overwritten by auto-detect
+```
+
+Profiles: `kid` (learning / tutorial), `non_dev_site` (site maintenance), `algo` (algorithms / research), `embedded` (firmware / FPGA / PLC), `ai_app` (AI app / agent), `general` (default).
+
+---
+
+## All commands
+
+| Command | Description |
+|---|---|
+| `duoshe init` | Initialize memory vault (scan + generate drafts) in the current directory |
+| `duoshe init --guided` | Recommended: init + 3 questions + smart skill suggestion |
+| `duoshe init --force` | Force-rewrite drafts (sections marked `<!-- USER-CONFIRMED -->` are preserved) |
+| `duoshe init --quick` | Skip git history scan (faster on large repos) |
+| `duoshe guide` | Just the 3 questions (re-runnable any time after init) |
+| `duoshe rescan` | Re-scan and refresh the code skeleton (preserves confirmed sections) |
+| `duoshe remember "..."` | Add a pending memory candidate |
+| `duoshe review` | List pending candidates |
+| `duoshe save <id>` | Promote to long-term memory (alias: `publish`) |
+| `duoshe drop <id>` | Discard (alias: `reject`) |
+| `duoshe search "..."` | Search long-term memory via SQLite FTS5 |
+| `duoshe reindex` | Rebuild the local search index |
+| `duoshe skill list` | List available / enabled skills |
+| `duoshe skill enable <name>` | Enable a skill |
+| `duoshe skill disable <name>` | Disable a skill |
+| `duoshe profile show / list / set` | View or change the project profile |
+| `duoshe sync` | Sync DuoShe blocks in `CLAUDE.md` / `AGENTS.md` |
+| `duoshe graph` | Analyze import dependencies (requires `graph` skill enabled) |
+| `duoshe upgrade` | Check for a new nano-duoshe release |
+| `duoshe uninstall` | Remove DuoShe blocks from `CLAUDE.md` / `AGENTS.md` (does not delete `.duoshe/`) |
+| `duoshe mcp` | _Planned_: MCP stdio server |
+| `duoshe session ...` | _Planned_: import conversation transcripts |
+
+Candidate types: `decision`, `troubleshooting`, `module_boundary`, `project_fact`, `user_preference`.
+
+---
+
+## Team workflow
+
+- After you `git push`, **teammates don't need to run `duoshe init`** — the repo already has the shared `.duoshe/` memory.
+- For them to use `duoshe` commands locally: `npm i -g nano-duoshe`, then run anywhere in the project.
+- First run rebuilds the local search index automatically (doesn't affect others).
+- To make tools that don't read `CLAUDE.md` (e.g. Cursor) use the memory, see `.duoshe/SETUP.md`.
+
+---
+
+## Current status
+
+`0.1.0-alpha.0`. Implemented:
+
+- repo scan + PROJECT.md / CODEMAP.md / DECISIONS.md draft generation
+- auto profile detection (6 profiles)
+- 5 bundled optional skills (embedded / matlab / devops / wordpress / graph)
+- CLAUDE.md / AGENTS.md auto-sync (preserves existing content)
+- candidate → review → save / drop flow
+- SQLite FTS5 search with CJK + Latin tokenization
+- `.gitignore` auto-maintenance (private state isolation)
+- simple guided mode (`init --guided`)
+- soft update notifier + `duoshe upgrade`
+- CI: Linux / macOS / Windows × Node 20/22 lint / typecheck / test / build
+
+Planned:
+
+- **M4**: session transcript import and deduplication
+- **M5**: MCP stdio server (`memory.search`, `memory.get_project_context`, candidate tools)
+- **M6**: Claude Code hook templates
+- **v0.2**: optional LLM-assisted candidate extraction
+
+---
+
+## Design principles
+
+- **Local-first.** Memory lives in your repo as Markdown and SQLite. No cloud.
+- **Review before permanence.** Agent suggestions go through a candidate queue.
+- **Traceable.** Saved memory keeps source metadata.
+- **Tool-agnostic.** The CLI works everywhere; MCP is the next adapter.
+- **Non-invasive.** DuoShe only edits its own delimited block in `CLAUDE.md` / `AGENTS.md`.
+- **Rebuildable index.** `index.db` is derived data; regenerable any time.
+- **Occam's razor.** The core is 5 essential commands; domain features grow as opt-in skills.
+
+---
+
+## Install
+
+Once published:
+
+```bash
+npm install -g nano-duoshe
+```
+
+Local development:
+
+```bash
+git clone https://github.com/guangdino/nano-duoshe.git
+cd nano-duoshe
+npm install
+npm run build
+npm link
+```
+
+Requirements: Node.js 20+, npm, native build support for `better-sqlite3`.
+
+---
+
+## Privacy
+
+DuoShe is local-first. Public memory files are normal Markdown — you decide what enters git. `index.db`, `CANDIDATES/`, and future `SESSIONS/` are rebuildable / noisy / private and should generally stay out of version control (`duoshe init` adds them to `.gitignore` automatically).
+
+---
+
+## Development
+
+```bash
+npm install
+npm run lint
+npm run typecheck
+npm test
+npm run build
+```
+
+| Script | Purpose |
+|---|---|
+| `npm run dev` | Run the TypeScript CLI through `tsx` |
+| `npm run build` | Compile to `dist/` (with skill assets) |
+| `npm test` | Vitest |
+| `npm run lint` | Biome |
+| `npm run prepublishOnly` | Release gate: lint + typecheck + test + build |
+
+See [DESIGN.md](./DESIGN.md) for the engineering plan.
+
+---
+
+## Contributing
+
+DuoShe is early. Issues, discussions, and small focused PRs are welcome.
+
+Before opening a PR:
+
+```bash
+npm run prepublishOnly
+```
+
+Please keep changes aligned with these constraints:
+
+- don't auto-promote agent output to long-term memory;
+- preserve existing user content in `CLAUDE.md` / `AGENTS.md`;
+- keep core logic independent from CLI and future MCP adapters;
+- prefer simple local formats over remote services;
+- new domain capabilities go in as skills rather than into the core.
+
+---
+
+## License
+
+MIT. See [LICENSE](./LICENSE).

package/README.zh-CN.md

@@ -0,0 +1,285 @@
+# nano-duoshe（夺舍）
+
+> 本地优先的项目记忆层，给 AI 编程工具（Claude Code / Codex / Cursor）用。
+>
+> 把项目约定、决策、踩坑、模块边界放进代码仓库里 —— AI 下次进项目时不再像第一次来。
+
+[English](./README.md) | **简体中文**
+
+## 为什么做 DuoShe
+
+AI 编程工具已经很强，但有一个很烦的冷启动问题：
+
+- 每次新会话都不记得项目结构；
+- 你反复解释架构、约定和"千万别动"的地方；
+- 它会重复推荐你已经否掉的方案；
+- 它会重新踩你已经修过的坑；
+- `CLAUDE.md` / `AGENTS.md` 有用，但本质是扁平手写文件，没有 review、来源追踪、也搜不动。
+
+DuoShe 在每个项目里放一个轻量的本地记忆库：长期记忆是 Markdown，候选记忆先进 review 队列、确认后再保存，搜索用 SQLite FTS5。本地、快、可重建。
+
+---
+
+## 30 秒上手
+
+```bash
+npm install -g nano-duoshe        # 安装
+cd your-project
+duoshe init --guided              # 初始化 + 3 个核心问题 + 智能推荐 skill
+```
+
+就这样。完成后会告诉你日常用 3 个命令：
+
+```bash
+duoshe remember "..."   # 想到什么先记下来
+duoshe review           # 看待确认的记录，保存或丢弃
+duoshe search "..."     # 在项目记忆里搜
+```
+
+---
+
+## 它会生成什么
+
+```text
+your-project/
+├─ .duoshe/                 # 项目记忆库（部分文件公共、部分私人）
+│  ├─ PROJECT.md            # ✅ 项目概览、技术栈、约定（commit）
+│  ├─ CODEMAP.md            # ✅ 代码地图、入口、路由提示（commit）
+│  ├─ DECISIONS.md          # ✅ 架构决策和原因（commit）
+│  ├─ TROUBLESHOOTING.md    # ✅ 已知问题和修法（commit）
+│  ├─ MODULES.md            # ✅ 模块边界，特别是"不该做什么"（commit）
+│  ├─ TODO.md               # ✅ 当前 / 接下来的工作（commit）
+│  ├─ SETUP.md              # ✅ 给团队看的接入说明（commit）
+│  ├─ SKILLS/               # ✅ 已启用的技能 manifest（commit）
+│  ├─ config.json           # ❌ 本地配置（个人偏好，不 commit）
+│  ├─ CANDIDATES/           # ❌ 草稿状态的待确认记录（不 commit）
+│  └─ index.db              # ❌ SQLite FTS5 索引，可重建（不 commit）
+├─ AGENTS.md                # 自动追加 DuoShe 块，引导 Codex 读 .duoshe/
+├─ CLAUDE.md                # 自动追加 DuoShe 块，引导 Claude Code 读 .duoshe/
+└─ .gitignore               # 自动加入需要忽略的本地状态
+```
+
+`duoshe init` 在 git 仓库里跑时会**自动维护 `.gitignore`**，把私人状态加进去。
+公共记忆（PROJECT.md 等）正常 `git add` / `git commit` 即可。
+
+核心规则：**AI 可以提出记忆，人来决定什么进入长期记忆。**
+
+---
+
+## 用 DuoShe 的两种姿势
+
+### 1. 第一次跑（推荐 --guided）
+
+```bash
+duoshe init --guided
+```
+
+这会：
+
+1. 扫描项目，生成草稿模板；
+2. 检测项目类型（嵌入式 / 算法 / 网站 / AI 应用 / 通用 / 学习项目）；
+3. 安装内置技能（默认不启用）；
+4. 同步 `CLAUDE.md` / `AGENTS.md`；
+5. 问 3 个核心问题（项目是干嘛的、AI 必须记住的规矩、不要乱动的地方）；
+6. 如果检测到嵌入式 / 算法 / WordPress，问你要不要启用对应 skill。
+
+完整体验 1 分钟左右。
+
+### 2. 日常用（3 个命令）
+
+```bash
+duoshe remember "数据库操作必须走 service 层，不能在 route 里直查"
+duoshe review            # 看待确认，输入 save <id> / drop <id>
+duoshe search "service"  # 找过去记下的内容
+```
+
+---
+
+## Skills（按需启用的扩展能力）
+
+DuoShe 自带一些专项识别能力，**默认不启用**，按需打开：
+
+| Skill | 适合什么场景 |
+|---|---|
+| `embedded` | 嵌入式开发（C 固件 / FPGA / PLC / STM32 / ESP32 / Vivado / Codesys） |
+| `matlab` | 算法 / 控制工程（MATLAB / Simulink / 数学推导密集型项目） |
+| `devops` | 基础设施 / IaC（Terraform / Ansible / Kubernetes） |
+| `wordpress` | WordPress 站点维护（非开发者也能用） |
+| `graph` | 依赖图分析、循环依赖检测、热点模块 |
+
+操作：
+
+```bash
+duoshe skill list                # 看所有已安装的
+duoshe skill enable embedded     # 启用一个（会打印一段引导）
+duoshe rescan                    # 用新激活的 skill 重新扫描
+duoshe skill disable embedded    # 关掉
+```
+
+启用后 `rescan` 会用新增的探测器和目录标签丰富项目记忆。
+
+---
+
+## Profile（项目类型）
+
+DuoShe 在 init 时会自动猜你的项目类型 —— 这影响 PROJECT.md 的引导提示和默认建议的 skill。
+
+```bash
+duoshe profile show              # 看当前类型 + 现在重新扫描会猜啥
+duoshe profile list              # 看所有可选类型
+duoshe profile set embedded      # 手动设置，下次不会被自动检测覆盖
+```
+
+类型：`kid`（学习 / 练习）、`non_dev_site`（网站维护）、`algo`（算法 / 研究）、`embedded`（嵌入式 / FPGA / PLC）、`ai_app`（AI 应用 / Agent）、`general`（通用）。
+
+---
+
+## 全部命令
+
+| 命令 | 说明 |
+|---|---|
+| `duoshe init` | 在当前目录初始化记忆库（扫描 + 生成 PROJECT.md 等草稿） |
+| `duoshe init --guided` | 推荐：初始化 + 3 个问题 + 智能推荐 skill |
+| `duoshe init --force` | 强制重写草稿（带 `<!-- USER-CONFIRMED -->` 的内容会保留） |
+| `duoshe init --quick` | 跳过 git 历史扫描（大仓库更快） |
+| `duoshe guide` | 单独跑 3 个问题（init 后随时可再跑） |
+| `duoshe rescan` | 重扫项目，刷新代码骨架（保留你确认过的部分） |
+| `duoshe remember "..."` | 添加一条候选记忆（pending 状态） |
+| `duoshe review` | 看待确认的记录 |
+| `duoshe save <id>` | 保存到长期记忆（别名：`publish`） |
+| `duoshe drop <id>` | 丢弃（别名：`reject`） |
+| `duoshe search "..."` | 用 SQLite FTS5 搜长期记忆 |
+| `duoshe reindex` | 重建本地搜索索引 |
+| `duoshe skill list` | 列出所有可用 / 已启用技能 |
+| `duoshe skill enable <name>` | 启用技能 |
+| `duoshe skill disable <name>` | 禁用技能 |
+| `duoshe profile show / list / set` | 看 / 改项目类型 |
+| `duoshe sync` | 同步 CLAUDE.md / AGENTS.md 里的 DuoShe 块 |
+| `duoshe graph` | 分析代码 import 依赖（需启用 graph skill） |
+| `duoshe upgrade` | 检查 nano-duoshe 有没有新版本 |
+| `duoshe uninstall` | 从 CLAUDE.md / AGENTS.md 移除 DuoShe 块（不删除 .duoshe/） |
+| `duoshe mcp` | _规划中_：启动 MCP stdio server |
+| `duoshe session ...` | _规划中_：导入对话 transcript |
+
+候选类型：`decision`、`troubleshooting`、`module_boundary`、`project_fact`、`user_preference`。
+
+---
+
+## 团队协作
+
+- `git push` 之后，**其他人 pull 下来不用再跑 `duoshe init`** —— 仓库里已经有 `.duoshe/` 的公共记忆了。
+- 队友想用 `duoshe` 命令搜索 / 记录，让他们 `npm i -g nano-duoshe`，然后在项目目录跑就行。
+- 第一次跑命令时 duoshe 会自动重建本地搜索索引（不影响别人）。
+- 想让 Cursor 之类不读 `CLAUDE.md` 的工具也用上记忆，看 `.duoshe/SETUP.md` 里的说明。
+
+---
+
+## 当前状态
+
+`0.1.0-alpha.0`。已实现：
+
+- 仓库扫描 + 生成 PROJECT.md / CODEMAP.md / DECISIONS.md 等草稿
+- 项目类型自动检测（6 种 profile）
+- 内置 5 个可选 skill（embedded / matlab / devops / wordpress / graph）
+- CLAUDE.md / AGENTS.md 自动同步（不覆盖你的原内容）
+- 候选记忆 → review → save / drop 流程
+- SQLite FTS5 中英文搜索（CJK bigram + Latin 分词）
+- `.gitignore` 自动维护（私人状态隔离）
+- 简易引导模式（init --guided）
+- 软更新通知 + `duoshe upgrade`
+- CI workflow：Linux / macOS / Windows × Node 20/22 跑 lint / typecheck / test / build
+
+规划中：
+
+- M4: Session transcript 导入和去重
+- M5: MCP stdio server（`memory.search`、`memory.get_project_context` 等工具）
+- M6: Claude Code hooks 模板
+- v0.2: 可选的 LLM 辅助候选提取
+
+---
+
+## 设计原则
+
+- **本地优先。** 记忆存仓库里，用 Markdown 和 SQLite，不依赖云服务。
+- **先 review，再长期化。** AI 提的建议先进候选队列。
+- **可追溯。** 保存后的记忆保留来源元信息。
+- **工具中立。** CLI 永远可用，MCP 是下一步接入层。
+- **不侵入。** DuoShe 只动 `CLAUDE.md` / `AGENTS.md` 里自己那段带标记的块。
+- **索引可重建。** `index.db` 是派生数据，随时可以重新生成。
+- **奥卡姆剃刀。** 核心只保留最不可少的 5 个命令；专项功能用 skill 按需启用。
+
+---
+
+## 安装
+
+正常用：
+
+```bash
+npm install -g nano-duoshe
+```
+
+本地开发：
+
+```bash
+git clone https://github.com/guangdino/nano-duoshe.git
+cd nano-duoshe
+npm install
+npm run build
+npm link
+```
+
+要求：Node.js 20+、npm、能编译 `better-sqlite3` 的环境。
+
+---
+
+## 隐私
+
+DuoShe 默认本地优先。公共记忆是 Markdown，是否进 git 由你决定。`index.db`、`CANDIDATES/`、`SESSIONS/` 这些通常不建议提交（`duoshe init` 会自动加入 `.gitignore`）。
+
+---
+
+## 开发
+
+```bash
+npm install
+npm run lint
+npm run typecheck
+npm test
+npm run build
+```
+
+| 脚本 | 用途 |
+|---|---|
+| `npm run dev` | 通过 `tsx` 直接跑 TypeScript |
+| `npm run build` | 编译到 `dist/`（含 skill 资源） |
+| `npm test` | Vitest |
+| `npm run lint` | Biome |
+| `npm run prepublishOnly` | 发布前门禁：lint + typecheck + test + build |
+
+工程方案见 [DESIGN.md](./DESIGN.md)。
+
+---
+
+## 贡献
+
+DuoShe 还很早期，欢迎 issue、讨论和小而清晰的 PR。
+
+提交 PR 前请先跑：
+
+```bash
+npm run prepublishOnly
+```
+
+请尽量遵守这些约束：
+
+- 不要让 AI 自动把内容晋升为长期记忆；
+- 不要覆盖用户已有的 `CLAUDE.md` / `AGENTS.md` 内容；
+- 保持 core 逻辑独立于 CLI 和未来 MCP adapter；
+- 优先使用简单、本地、可检查的格式；
+- 新增域专项能力优先考虑做成 skill，而不是塞进核心。
+
+---
+
+## License
+
+MIT。见 [LICENSE](./LICENSE)。

package/dist/adapters/claude-md.d.ts

@@ -0,0 +1,40 @@
+export declare const BEGIN_MARK = "<!-- BEGIN DUOSHE -->";
+export declare const END_MARK = "<!-- END DUOSHE -->";
+export declare const SHELL_FILES: readonly ["CLAUDE.md", "AGENTS.md"];
+export type ShellFile = (typeof SHELL_FILES)[number];
+export type ShellAction = {
+    file: ShellFile;
+    status: "created";
+    path: string;
+} | {
+    file: ShellFile;
+    status: "appended";
+    path: string;
+} | {
+    file: ShellFile;
+    status: "updated";
+    path: string;
+} | {
+    file: ShellFile;
+    status: "unchanged";
+    path: string;
+} | {
+    file: ShellFile;
+    status: "skipped-no-existing";
+    path: string;
+} | {
+    file: ShellFile;
+    status: "removed";
+    path: string;
+};
+export declare function buildShellBlock(): string;
+export type SyncOptions = {
+    createIfMissing: boolean;
+};
+export declare function syncShells(projectRoot: string, opts: SyncOptions): ShellAction[];
+export declare function uninstallShells(projectRoot: string): ShellAction[];
+export declare function detectExistingShells(projectRoot: string): {
+    file: ShellFile;
+    exists: boolean;
+}[];
+//# sourceMappingURL=claude-md.d.ts.map

package/dist/adapters/claude-md.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-md.d.ts","sourceRoot":"","sources":["../../src/adapters/claude-md.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,UAAU,0BAA0B,CAAC;AAClD,eAAO,MAAM,QAAQ,wBAAwB,CAAC;AAE9C,eAAO,MAAM,WAAW,qCAAsC,CAAC;AAC/D,MAAM,MAAM,SAAS,GAAG,CAAC,OAAO,WAAW,CAAC,CAAC,MAAM,CAAC,CAAC;AAErD,MAAM,MAAM,WAAW,GACnB;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,SAAS,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GACpD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,UAAU,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GACrD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,SAAS,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GACpD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,WAAW,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GACtD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,qBAAqB,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,GAChE;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,SAAS,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAEzD,wBAAgB,eAAe,IAAI,MAAM,CAqCxC;AAUD,MAAM,MAAM,WAAW,GAAG;IACxB,eAAe,EAAE,OAAO,CAAC;CAC1B,CAAC;AAkCF,wBAAgB,UAAU,CAAC,WAAW,EAAE,MAAM,EAAE,IAAI,EAAE,WAAW,GAAG,WAAW,EAAE,CAEhF;AAED,wBAAgB,eAAe,CAAC,WAAW,EAAE,MAAM,GAAG,WAAW,EAAE,CAsBlE;AAED,wBAAgB,oBAAoB,CAAC,WAAW,EAAE,MAAM,GAAG;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,OAAO,CAAA;CAAE,EAAE,CAEhG"}