@researai/meos 0.1.0

package/README_ZH.md

@@ -0,0 +1,423 @@
+<h1 align="center">
+  <img src="assets/branding/logo.svg" alt="MeOS logo" width="92" />
+  MeOS
+</h1>
+
+<p align="center">
+  MeOS 是一个本地优先的操作层，用来把你的工作流、技术标准、偏好和纠偏规则整理成可复用的 agent 资产。
+</p>
+
+<p align="center">
+  它让 Codex、Claude Code、OpenCode 和 OpenClaw 更像“学会了你怎么工作”，同时默认把私有原始历史保留在本地。
+</p>
+
+<p align="center">
+  <a href="https://github.com/ResearAI/MeOS">https://github.com/ResearAI/MeOS</a>
+</p>
+
+<p align="center">
+  <a href="https://github.com/ResearAI/MeOS">GitHub</a> |
+  <a href="README.md">English README</a> |
+  <a href="#quick-start">快速开始</a> |
+  <a href="#runtime-setup">运行时配置</a> |
+  <a href="#repository-layout">仓库结构</a>
+</p>
+
+<p align="center">
+  <a href="https://github.com/ResearAI/MeOS"><img alt="GitHub stars" src="https://img.shields.io/github/stars/ResearAI/MeOS?style=for-the-badge&logo=github"></a>
+  <a href="LICENSE"><img alt="License MIT" src="https://img.shields.io/badge/License-MIT-yellow.svg?style=for-the-badge"></a>
+  <img alt="Format Agent Skill" src="https://img.shields.io/badge/Format-Agent%20Skill-green?style=for-the-badge" />
+  <img alt="Scope local-first" src="https://img.shields.io/badge/Scope-local--first-blue?style=for-the-badge" />
+</p>
+
+<p align="center">
+  <strong>不只是存储，而是实际应用</strong> ·
+  <strong>公开仓库，私有本地层</strong> ·
+  <strong>可编辑资产，不是黑箱记忆</strong> ·
+  <strong>一套安装器覆盖多个运行时</strong>
+</p>
+
+<p align="center">
+  <a href="#what-you-actually-get">能得到什么</a> •
+  <a href="#how-meos-works">工作方式</a> •
+  <a href="#promotion-and-privacy-rules">晋升与隐私规则</a> •
+  <a href="#key-skill-references">关键 Skill 参考</a>
+</p>
+
+![MeOS overview](assets/readme/00-overview.svg)
+
+MeOS 不是 memory dump。  
+不是角色扮演 prompt 包。  
+也不是一堆零散提示词。
+
+它是一个文件优先的系统，用来把“一个人的稳定工作方式”沉淀成未来 agent 真能用起来的资产。
+
+支持 Codex、Claude Code、OpenCode 和 OpenClaw。
+
+如果你已经厌倦了在每个新 agent 会话里重复解释同一套标准、同一种审美、同一类纠偏规则，MeOS 就是把这些重复劳动变成长期本地资产的那一层。
+
+## ✨ 为什么是 MeOS
+
+这一类系统通常各自只擅长一个点：
+
+- 把一个人蒸馏成一段 prompt
+- 从历史交互里抽技能
+- 保留一批可复用的说明文档
+
+MeOS 关注的是中间那层更长期、更稳定的“操作层”。
+
+| 常见做法 | 常见问题 | MeOS 的做法 |
+|---|---|---|
+| Memory dump | 噪声太多，复用性弱 | 只把稳定模式晋升成资产 |
+| Persona prompt | 听起来像你，但工作层很浅 | 直接存工作流、标准、原则和纠偏 |
+| 一次性画像总结 | 很快过时 | 把 `init`、`refresh`、`apply` 做成长期生命周期 |
+| 本地私人笔记 | 难以跨工具复用 | 用通用 `SKILL.md` 结构让多个运行时可加载 |
+
+核心思想很简单：
+
+> 不只是记住这个人，而是让这个人变得可复用
+
+<a id="what-you-actually-get"></a>
+## 🧩 你实际会得到什么
+
+| 资产类型 | 例子 | 改善什么 |
+|---|---|---|
+| `🛠` 工作标准 | 编码规则、review 标准、验收线 | 技术质量与一致性 |
+| `🧭` 工作流资产 | debug 顺序、架构评审顺序、交付清单 | agent 处理任务的方式 |
+| `🧠` 思维风格 | 推理模式、权衡习惯、决策风格 | 规划与判断质量 |
+| `🎨` 审美与偏好 | 输出结构、UI 审美、表达风格 | 最终结果的样子 |
+| `✍️` 纠偏规则 | 明确 override、历史否定过的东西 | 减少重复偏航 |
+| `📚` 知识资产 | 稳定事实、领域理解、可复用经验 | 超出单次聊天的上下文积累 |
+
+<a id="how-meos-works"></a>
+## ⚙️ MeOS 如何工作
+
+| 模式 | 目的 | 先读什么 | 会写回什么 |
+|---|---|---|---|
+| `🧱 init` | 从批准过的本地材料建立第一版资产 | source policy、extraction SOP、promotion policy、privacy policy | 初始资产和 evidence |
+| `🔁 refresh` | 用新增材料更新已有资产 | extraction SOP、promotion policy、correction policy | merge 后的更新、冲突、纠偏 |
+| `🎯 apply` | 在真实任务里使用已有资产 | 只读取最相关的资产 | 只有稳定新信息才写回 |
+
+`apply` 是最关键的模式。  
+这也是 MeOS 从“档案”变成“真正有用”的地方。
+
+<a id="quick-start"></a>
+## 🚀 Quick Start
+
+### 1. 克隆仓库
+
+```bash
+git clone https://github.com/ResearAI/MeOS.git
+cd MeOS
+```
+
+### 2. 安装到一个运行时
+
+安装器不会把整个仓库塞进运行时。  
+它真正安装的是 `./SKILL/` 的内容，目标目录是运行时里的 `meos/`。
+
+默认模式是 `--mode auto`。  
+它会用更安全的运行时策略：OpenClaw 用 `copy`，其他运行时默认 `symlink`。
+
+| 运行时 | 推荐命令 | 说明 |
+|---|---|---|
+| Codex | `bash install.sh --runtime codex` | 最简单的本地 skill 使用路径 |
+| Claude Code | `bash install.sh --runtime claude` | 注意 skill 目录名必须是小写 `meos` |
+| OpenClaw | `bash install.sh --runtime openclaw --force` | 更推荐复制目录 |
+| OpenCode | `bash install.sh --runtime opencode` | 只安装到一个兼容路径即可 |
+
+也可以使用 npm 包装器：
+
+```bash
+npm install -g @researai/meos
+meos install --runtime codex
+```
+
+如果你是在本地开发当前仓库，也可以这样装：
+
+```bash
+npm install -g .
+```
+
+### 3. 直接开始用
+
+```text
+Use meos in apply mode for this task. Read only the minimum relevant assets and use them to shape reasoning, workflow, and output.
+```
+
+常用 prompt：
+
+```text
+Use meos in init mode. Build the first sanitized operating-layer assets from the available local source material.
+```
+
+```text
+Use meos in refresh mode. Refresh the existing MeOS assets from new local material and only promote stable rules.
+```
+
+```text
+Use meos in apply mode for this task. Read only the minimum relevant assets and write back only stable new information.
+```
+
+<a id="runtime-setup"></a>
+## 🖥 运行时配置
+
+### Codex
+
+Codex 支持 `.agents/skills/` 和 `~/.agents/skills/` 这类 skill 目录。
+
+手工安装：
+
+```bash
+mkdir -p ~/.agents/skills
+ln -s /path/to/MeOS/SKILL ~/.agents/skills/meos
+```
+
+Codex 可以显式通过名字触发 MeOS，也可以通过 skill 的 `description` 隐式选择它。
+
+### Claude Code
+
+Claude Code 支持 `~/.claude/skills/<skill-name>/SKILL.md` 和 `.claude/skills/<skill-name>/SKILL.md`。
+
+手工安装：
+
+```bash
+mkdir -p ~/.claude/skills
+ln -s /path/to/MeOS/SKILL ~/.claude/skills/meos
+```
+
+常见用法：
+
+```text
+/meos
+Apply MeOS for this task. Read only the minimum relevant assets and use them to shape reasoning, workflow, and output.
+```
+
+### Claude Code + MiniMax
+
+如果你要让 Claude Code 走 MiniMax 的 Anthropic 兼容端点，本地 `~/.claude/settings.json` 可以这样写：
+
+```json
+{
+  "env": {
+    "ANTHROPIC_BASE_URL": "https://api.minimaxi.com/anthropic",
+    "ANTHROPIC_AUTH_TOKEN": "${MINIMAX_API_KEY}",
+    "API_TIMEOUT_MS": "3000000",
+    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
+  }
+}
+```
+
+验证：
+
+```bash
+claude -p --model MiniMax-M2.7 'Respond with exactly CLAUDE_MINIMAX_OK.'
+```
+
+### OpenClaw
+
+OpenClaw 支持 `~/.openclaw/skills`、`~/.agents/skills`、`<workspace>/.agents/skills` 和 `<workspace>/skills`。
+
+手动测试确认过的关键行为：
+
+- OpenClaw 会跳过真实路径逃离配置根目录的 symlink skill
+- 外部 symlink 安装并不稳
+- 最稳妥的是把 `meos` 作为真实目录复制到 `<workspace>/skills/meos` 或 `~/.openclaw/skills/meos`
+
+推荐手工安装：
+
+```bash
+mkdir -p <workspace>/skills
+cp -a /path/to/MeOS/SKILL <workspace>/skills/meos
+```
+
+验证：
+
+```bash
+openclaw skills info meos
+openclaw skills list | rg meos
+```
+
+### OpenCode
+
+OpenCode 会搜索多个兼容 skill 目录：
+
+- `.opencode/skills/<name>/SKILL.md`
+- `~/.config/opencode/skills/<name>/SKILL.md`
+- `.claude/skills/<name>/SKILL.md`
+- `~/.claude/skills/<name>/SKILL.md`
+- `.agents/skills/<name>/SKILL.md`
+- `~/.agents/skills/<name>/SKILL.md`
+
+只选一个路径安装即可：
+
+```bash
+mkdir -p ~/.config/opencode/skills
+ln -s /path/to/MeOS/SKILL ~/.config/opencode/skills/meos
+```
+
+如果你的 provider 或代理不支持 OpenCode 默认用来生成标题的副模型，请显式设置 `small_model`：
+
+```json
+{
+  "$schema": "https://opencode.ai/config.json",
+  "model": "openai/gpt-5.4",
+  "small_model": "openai/gpt-5.4"
+}
+```
+
+验证：
+
+```bash
+opencode run --model openai/gpt-5.4 --format json \
+  'Use meos in apply mode for this task. Reply with exactly OPENCODE_SKILL_OK.'
+```
+
+## 🗂 `apply` 模式下 agent 会读什么
+
+下面这些路径都相对于安装后的 skill 根目录。  
+在当前仓库里，它们实际位于 `SKILL/` 下。
+
+| 任务类型 | 优先读什么 | 会带来什么 |
+|---|---|---|
+| `🛠` 技术实现 | `assets/live/work/`、`assets/live/thought-style/`、`assets/live/workflow/`、`assets/live/principles/` | 更贴近你的技术标准和执行顺序 |
+| `🎨` UI / 产品 | `assets/live/taste/`、`assets/live/work/`、`assets/live/workflow/`、`assets/live/corrections/` | 更贴近你的审美和呈现要求 |
+| `🔬` 研究 / 写作 | `assets/live/work/`、`assets/live/thought-style/`、`assets/live/principles/`、`assets/live/knowledge/`、`assets/live/workflow/` | 更贴近你的结构、推理和知识组织方式 |
+| `💬` 风格敏感回复 | `assets/live/preferences/`、`assets/live/corrections/` | 更贴近你的表达形态和措辞习惯 |
+
+如果 `assets/live/corrections/` 和其他层冲突，以 correction 为准。
+
+<a id="promotion-and-privacy-rules"></a>
+## 🔒 晋升与隐私规则
+
+### 晋升流程
+
+1. 收集本地材料
+2. 判断来源类型
+3. 提取高信号候选事实
+4. 不确定内容先放到 `evidence/`
+5. 只有稳定规则才晋升到 `assets/live/`
+
+适合晋升的情况：
+
+- 用户明确说过
+- 在多个上下文里重复出现
+- 被用户明确纠正或强化过
+
+只该留在 `evidence/` 的情况：
+
+- 一次性行为
+- 噪声较大
+- 过度依赖上下文
+- 太敏感
+
+### 维护生命周期
+
+MeOS 的维护动作应该是：
+
+- add
+- merge
+- downgrade
+- discard
+
+重点不是堆越来越多 prompt，而是让资产层保持干净、可维护。
+
+### 隐私边界
+
+| 可公开内容 | 默认保留在本地 |
+|---|---|
+| `README.md` | `SKILL/private/` |
+| `README_ZH.md` | `SKILL/evidence/` |
+| `assets/branding/` | `SKILL/runtime/` |
+| `assets/readme/` | `SKILL/assets/live/` |
+| `SKILL/references/` | 原始导入材料 |
+| `SKILL/schemas/` | secrets 和 tokens |
+| `SKILL/assets/templates/` | 工作站本地备注 |
+| `SKILL/assets/examples/` | 私有原始对话 |
+
+不要提交：
+
+- token
+- API key
+- 个人身份标识
+- 原始 connector id
+- 不必要的私有路径
+- 私有原始对话
+
+![MeOS privacy boundary](assets/readme/03-privacy-boundary.svg)
+
+<a id="repository-layout"></a>
+## 🏗 仓库结构
+
+```text
+MeOS/
+├── README.md
+├── README_ZH.md
+├── LICENSE
+├── assets/
+│   ├── branding/
+│   └── readme/
+├── SKILL/
+│   ├── SKILL.md
+│   ├── references/
+│   ├── schemas/
+│   ├── assets/
+│   │   ├── templates/
+│   │   ├── examples/
+│   │   └── live/
+│   ├── evidence/
+│   ├── runtime/
+│   └── private/
+├── install.sh
+├── installer.py
+├── package.json
+└── bin/
+```
+
+关键结构就是：
+
+- 仓库根目录放公开项目材料
+- 运行时真正需要的 skill 内容都放在 `SKILL/`
+- 安装器负责把 `SKILL/` 安装进运行时 skill 目录
+- 本地 owner 校准层保留在 `SKILL/assets/live/`、`SKILL/evidence/`、`SKILL/private/` 和 `SKILL/runtime/`
+
+<a id="key-skill-references"></a>
+## 📌 关键 Skill 参考
+
+`SKILL/` 里最值得先看的文件是：
+
+- [SKILL/SKILL.md](SKILL/SKILL.md)
+- [SKILL/references/source-locations.md](SKILL/references/source-locations.md)
+- [SKILL/references/extraction-sop.md](SKILL/references/extraction-sop.md)
+- [SKILL/references/promotion-policy.md](SKILL/references/promotion-policy.md)
+- [SKILL/references/privacy-policy.md](SKILL/references/privacy-policy.md)
+- [SKILL/references/writeback-policy.md](SKILL/references/writeback-policy.md)
+- [SKILL/references/apply-task-map.md](SKILL/references/apply-task-map.md)
+
+## 🛤 当前方向
+
+MeOS 现在已经有：
+
+- 一套跨运行时的 `SKILL/` 包布局
+- extraction、promotion、privacy、writeback 参考文档
+- durable entry 的 JSON schema
+- example 和 template 资产树
+- 面向 Codex、Claude Code、OpenCode、OpenClaw 的安装路径
+
+下一步不应该是“继续堆 prompt 文本”。  
+而是继续完善资产、示例和公开展示，同时不泄露任何私有历史。
+
+## 📚 引用
+
+如果 MeOS 被用于人格对齐、风格蒸馏或长期操作层资产维护相关工作，也可以考虑引用：
+
+```bibtex
+@inproceedings{
+zhu2025personality,
+title={Personality Alignment of Large Language Models},
+author={Minjun Zhu and Yixuan Weng and Linyi Yang and Yue Zhang},
+booktitle={The Thirteenth International Conference on Learning Representations},
+year={2025},
+url={https://openreview.net/forum?id=0DZEs8NpUH}
+}
+```

package/SKILL/SKILL.md

@@ -0,0 +1,130 @@
+---
+name: meos
+description: Read the owner's local history and profile assets, extract stable workflows and preferences, and keep those assets updated for future tasks. Use when explicitly asked to initialize, refresh, or apply MeOS.
+license: MIT
+---
+
+# MeOS
+
+MeOS is a local-first operating-layer skill.
+
+It is not a chat memory dump.
+It is not literal roleplay.
+It is not a one-shot profile summary.
+
+It exists to keep the owner's long-lived working assets readable, editable, and reusable.
+
+## Modes
+
+- `init`
+- `refresh`
+- `apply`
+
+### `init`
+
+Use when MeOS has not been built yet and the agent must create the first durable asset set from local history sources.
+
+### `refresh`
+
+Use when new history or new corrections exist and the agent must update existing assets without duplicating or overwriting them blindly.
+
+### `apply`
+
+Use when the current task should be executed with MeOS calibration and only stable new information should be written back.
+
+`apply` is the most important mode in real usage.
+It is what makes MeOS useful in later tasks instead of becoming an archive-only system.
+
+## Hard rules
+
+1. Do not treat one-off behavior as a stable rule.
+2. Prefer explicit user statements over inferred patterns.
+3. Prefer correction signals over older assumptions.
+4. Do not promote a fact into `assets/` unless its evidence is strong enough.
+5. Keep private raw material out of git by default.
+6. Do not store secrets, tokens, or unnecessary personal identifiers in reusable assets.
+7. If evidence is weak or conflicting, store it in `evidence/` instead of pretending certainty.
+
+## Repository layout
+
+- `references/`
+  - operating manuals for extraction, promotion, privacy, and writeback
+- `schemas/`
+  - JSON schemas for durable entries
+- `assets/`
+  - stable, reusable owner assets
+- `evidence/`
+  - raw promoted facts, conflict records, and source maps
+- `runtime/`
+  - local run state
+- `private/`
+  - imported raw history and local-only material
+
+## Reading order
+
+### For `init`
+
+1. Read `references/source-locations.md`
+2. Read `references/extraction-sop.md`
+3. Read `references/promotion-policy.md`
+4. Read `references/privacy-policy.md`
+5. Read `references/writeback-policy.md`
+
+### For `refresh`
+
+1. Read `references/source-locations.md`
+2. Read `references/extraction-sop.md`
+3. Read `references/promotion-policy.md`
+4. Read `references/correction-policy.md`
+5. Read `references/writeback-policy.md`
+
+### For `apply`
+
+1. Read only the minimum relevant files under `assets/`
+2. Read `references/writeback-policy.md` before writing back any new durable update
+3. Read `references/apply-task-map.md` when you need the task-to-asset mapping
+
+## Promotion rule
+
+Use this promotion ladder:
+
+- weak or one-off evidence -> `evidence/`
+- repeated or explicit stable rule -> `assets/`
+- direct user correction -> `assets/corrections/`
+
+Never skip `evidence/` when uncertainty is real.
+
+## Write-back rule
+
+When stable new information appears:
+
+- update the appropriate file under `assets/`
+- append a matching source record under `evidence/`
+- if a conflict exists, append it to `evidence/conflicts.jsonl`
+- if the new information is a correction, prefer updating `assets/corrections/overrides.md`
+
+## Apply behavior
+
+When MeOS is used during a real task, do not load the whole asset tree by default.
+
+Instead:
+
+1. identify the task type
+2. read only the relevant live assets
+3. let those assets shape:
+   - reasoning
+   - workflow
+   - output
+   - correction handling
+4. only write back stable new information
+
+## Expected outputs
+
+Good MeOS work usually leaves behind:
+
+- one or more updated asset files
+- one or more evidence records
+- conflict records when needed
+- no private-secret leakage into reusable assets
+
+MeOS should make the owner more reusable, not more exposed.

package/SKILL/assets/INDEX.md

@@ -0,0 +1,21 @@
+# MeOS Assets Index
+
+This file is the human-readable index for promoted long-term assets.
+
+## Categories
+
+- `work/`
+- `thought-style/`
+- `workflow/`
+- `principles/`
+- `taste/`
+- `knowledge/`
+- `preferences/`
+- `corrections/`
+
+## Intended usage
+
+- read narrowly
+- load only the categories relevant to the current task
+- update assets conservatively
+- keep `corrections/` authoritative when conflicts appear

package/SKILL/assets/corrections/overrides.md

@@ -0,0 +1,3 @@
+# Corrections Overrides
+
+Use this file for explicit high-priority corrections that should override older inferred rules.

package/SKILL/assets/examples/README.md

@@ -0,0 +1,10 @@
+# Examples
+
+This directory is for sanitized example asset sets.
+
+Suggested families:
+
+- `minimal/`
+- `engineering/`
+- `research/`
+- `product/`

package/SKILL/assets/examples/engineering/README.md

@@ -0,0 +1,7 @@
+# Engineering Example
+
+This example should demonstrate a technical execution profile with:
+
+- architecture standards
+- verification rules
+- implementation and debug workflow

package/SKILL/assets/examples/engineering/corrections/overrides.md

@@ -0,0 +1,4 @@
+# Overrides
+
+- Do not report broad "healthy" progress after a concrete bug report without verifying the exact claim.
+- Do not treat rewritten control files as equivalent to a new measured result.

package/SKILL/assets/examples/engineering/principles/tradeoffs.md

@@ -0,0 +1,11 @@
+# Trade-offs
+
+## Preferred trade-offs
+
+- prefer a single convergent contract over parallel convenience hacks
+- prefer smaller surfaces with stronger semantics
+- prefer direct integration when the user explicitly rejects simplification
+
+## Boundaries
+
+- do not hide complexity by silently replacing the requested architecture with a weaker surrogate

package/SKILL/assets/examples/engineering/taste/ui-product.md

@@ -0,0 +1,12 @@
+# UI And Product Taste
+
+## Product direction
+
+- coherent navigation and object models
+- visible state transitions
+- enough detail to inspect without clutter
+
+## Component rule
+
+- use cards, dialogs, tabs, and steppers only when they support a real task contract
+- avoid random component dumping for visual novelty alone

package/SKILL/assets/examples/engineering/thought-style/reasoning-patterns.md

@@ -0,0 +1,11 @@
+# Reasoning Patterns
+
+## Preferred analytical frame
+
+- distinguish symptom, mechanism, contract, and route
+- use invariants and state transitions to reason about failures
+- reject generic reassurance in place of direct verification
+
+## Debugging rule
+
+- If a user says "the page still has not changed", check the live route, build output, variant, and served bundle before claiming success.

package/SKILL/assets/examples/engineering/work/technical-standards.md

@@ -0,0 +1,17 @@
+# Technical Standards
+
+## System Design
+
+- Every important runtime object should have a stable identifier or query path.
+- Prompt, skill, tool surface, and UI should describe the same workflow.
+- UI must not invent semantics that the backend cannot justify.
+
+## Engineering Quality
+
+- Installation and startup scripts are part of the product, not afterthoughts.
+- Frontend and backend changes must be checked together when the behavior spans both layers.
+
+## Verification
+
+- Verify the actual rendered component or running bundle, not only the edited file.
+- Distinguish source correctness from deployment correctness.

package/SKILL/assets/examples/engineering/workflow/default-operating-sequence.md

@@ -0,0 +1,16 @@
+# Default Operating Sequence
+
+## Large audit workflow
+
+1. read the governing plan or specification
+2. inspect only the named directories or modules
+3. map each planned feature to real implementation evidence
+4. return a structured status matrix
+
+## Multi-layer implementation workflow
+
+1. identify frontend files
+2. identify backend files
+3. identify data or API contract
+4. identify bootstrap or migration implications
+5. define verification steps

package/SKILL/assets/examples/minimal/README.md

@@ -0,0 +1,8 @@
+# Minimal Example
+
+This example should demonstrate the smallest useful MeOS profile:
+
+- one work rule
+- one thought-style rule
+- one workflow rule
+- one correction

package/SKILL/assets/examples/minimal/corrections/overrides.md

@@ -0,0 +1,4 @@
+# Overrides
+
+- If the user asks to continue extending the work, do not keep defending an older closeout state.
+- If the user says a concrete runtime detail looks wrong, verify that detail before giving a broad progress summary.