@memo-code/memo 0.7.3 → 0.8.5

package/README.md

@@ -5,20 +5,24 @@
 </div>
 
 <p align="center">
-  <a href="README.zh.md">Chinese Documentation</a>
+  <a href="https://github.com/minorcell/memo-code/actions/workflows/ci.yml">
+    <img src="https://github.com/minorcell/memo-code/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI">
+  </a>
+  <a href="https://codecov.io/gh/minorcell/memo-code">
+    <img src="https://codecov.io/gh/minorcell/memo-code/graph/badge.svg" alt="Coverage">
+  </a>
+  <a href="https://www.npmjs.com/package/@memo-code/memo">
+    <img src="https://img.shields.io/npm/v/%40memo-code%2Fmemo" alt="npm version">
+  </a>
 </p>
 
 <p align="center">
-  <a href="public/memo-code-cli-show-01.mp4">
-    <img src="https://img.shields.io/badge/📹-Watch%20Demo%20Video-1a1a1a?style=for-the-badge" alt="Demo Video">
-  </a>
+  <a href="README.zh.md">Chinese Documentation</a>
 </p>
 
----
-
-<video src="public/memo-code-cli-show-01.mp4" width="100%"></video>
+<img src="public/demo.png" width="100%" alt="Memo Code demo">
 
-Built with Node.js + TypeScript. DeepSeek is the default provider, and OpenAI-compatible APIs are supported.
+Built with Node.js + TypeScript. Memo works with OpenAI-compatible APIs.
 
 Memo Code is an open-source coding agent that lives in your terminal, understands repository context, and helps you move faster with natural-language commands.
 
@@ -36,10 +40,13 @@ yarn global add @memo-code/memo
 bun add -g @memo-code/memo
 ```
 
+Note: npm distribution is prebuilt and already includes required runtime/web assets.  
+`pnpm run build` is only needed when you run Memo from source checkout in this repository.
+
 ### 2. Configure API Key
 
 ```bash
-export DEEPSEEK_API_KEY=your_key  # or OPENAI_API_KEY
+export OPENAI_API_KEY=your_key
 ```
 
 ### 3. Start
@@ -53,11 +60,30 @@ memo
 
 - Interactive mode: `memo` (default TUI; supports multi-turn chat, tool visualization, shortcuts).
 - Plain mode (non-TTY): `echo "your prompt" | memo` (plain text output; useful for scripts).
+- One-shot mode: `memo --once "your prompt"` or `memo -once "your prompt"` (run one turn and exit; recommended with `--dangerous` in trusted repos).
+- Continue latest local session: `memo --prev` or `memo -prev` (load latest session context for current directory).
 - Dangerous mode: `memo --dangerous` or `memo -d` (skip tool approvals; use carefully).
 - Version: `memo --version` or `memo -v`.
+- Start web server: `memo web --host 127.0.0.1 --port 5494 --open` (npm package already includes web assets; source checkout needs `pnpm run build`).
 - Startup project guidance: if `AGENTS.md` exists in the startup root, Memo appends it to the system prompt automatically.
+- Skills: Memo auto-discovers `SKILL.md` files and appends an available-skills section into the system prompt.
 - MCP activation selection: when MCP servers are configured, startup shows a multi-select to activate servers for this run.
-- Session titles: Memo generates a short title from the first user prompt and uses it in history/resume lists.
+- Web app supports multi-workspace project management and concurrent live sessions (up to 20 per server process).
+
+## Web Console
+
+```bash
+memo web --host 127.0.0.1 --port 5494 --open
+```
+
+- npm distribution already bundles web server + web UI assets.
+- if running from source checkout, build once with `pnpm run build`.
+- Web auth config is stored in `~/.memo/server.yaml` by default (`MEMO_SERVER_CONFIG` can override path).
+- On first startup, Memo creates this file with generated auth secrets and a random password.
+- Login page uses `server.yaml` credentials (`auth.username` / `auth.password`).
+- Sidebar includes dedicated entries for `MCP Servers` and `Skills`:
+    - MCP: create/edit/remove/login/logout and active toggles.
+    - Skills: create/delete, detail preview, and active toggles.
 
 ## Configuration
 
@@ -66,17 +92,42 @@ Location: `~/.memo/config.toml` (can be changed via `MEMO_HOME`).
 ### Provider Configuration
 
 ```toml
-current_provider = "deepseek"
-
-[[providers.deepseek]]
-name = "deepseek"
-env_api_key = "DEEPSEEK_API_KEY"
-model = "deepseek-chat"
-base_url = "https://api.deepseek.com"
+current_provider = "openai_compatible"
+auto_compact_threshold_percent = 80
+
+[[providers.openai_compatible]]
+name = "openai_compatible"
+env_api_key = "OPENAI_API_KEY"
+model = "gpt-4.1-mini"
+base_url = "https://api.openai.com/v1"
 ```
 
 You can configure multiple providers and switch with `current_provider`.
 
+Optional: override model capability profiles (local capability gating) without code changes:
+
+```toml
+[model_profiles.gpt-5]
+supports_parallel_tool_calls = true
+supports_reasoning_content = true
+context_window = 272000
+
+[model_profiles."openai:gpt-5"]
+supports_parallel_tool_calls = false # provider-specific override
+```
+
+Context window policy:
+
+- priority 1: `model_profiles."provider:model".context_window`
+- priority 2: `model_profiles."<model>".context_window`
+- fallback: `120000`
+
+Auto compaction policy:
+
+- threshold: `auto_compact_threshold_percent` (default `80`)
+- trigger: estimated prompt tokens at step start reaches threshold
+- frequency: at most once per turn
+
 ### MCP Tool Configuration
 
 Both local and remote MCP servers are supported:
@@ -96,6 +147,11 @@ url = "https://your-mcp-server.com/mcp"
 # Optional: default active MCP servers at startup
 active_mcp_servers = ["local_tools", "remote"]
 # Optional: use [] to start with no MCP servers active
+
+# Optional: MCP OAuth credential storage mode: auto | keyring | file
+mcp_oauth_credentials_store_mode = "auto"
+# Optional: fixed localhost callback port for `memo mcp login`
+# mcp_oauth_callback_port = 33333
 ```
 
 You can also manage MCP configs via CLI (aligned with Codex CLI style):
@@ -110,11 +166,54 @@ memo mcp add local_tools -- /path/to/mcp-server --flag
 # Add remote MCP server (streamable HTTP)
 memo mcp add remote --url https://your-mcp-server.com/mcp --bearer-token-env-var MCP_TOKEN
 
+# OAuth login/logout for streamable_http servers
+memo mcp login remote --scopes read,write
+memo mcp logout remote
+
 # Show/remove
 memo mcp get remote
 memo mcp remove remote
 ```
 
+`memo mcp list` includes `auth_status` for each server: `unsupported`, `not_logged_in`, `bearer_token`, or `oauth`.
+
+## Skills
+
+Memo supports Agent Skills and auto-discovers `SKILL.md` files at startup.
+
+### Discovery Paths
+
+- Project scope: project root `.<agent>/skills` directories (for example: `.agents/skills`, `.claude/skills`, `.codex/skills`)
+- User scope: `$MEMO_HOME/skills` (or `~/.memo/skills`)
+- Not scanned: `~/.xxx/skills` hidden directories outside Memo home
+
+### Minimal Skill File
+
+```md
+---
+name: doc-writing
+description: Generate and update technical documentation.
+---
+
+# Doc Writing
+```
+
+Memo reads `name` and `description` from frontmatter and injects each skill as metadata:
+
+- `- <name>: <description> (file: <absolute-path-to-SKILL.md>)`
+
+In prompts, users can explicitly mention a skill with `$skill-name` (for example, `$doc-writing`).
+
+Optional: persist default active skills in `config.toml`:
+
+```toml
+# Unset: all discovered skills are active by default.
+# []: disable all skills by default.
+active_skills = [
+  "/absolute/path/to/.codex/skills/doc-writing/SKILL.md"
+]
+```
+
 ## Built-in Tools
 
 - `exec_command` / `write_stdin`: execute shell commands (default shell family)
@@ -166,13 +265,14 @@ pnpm start
 ### Build
 
 ```bash
-pnpm run build  # generates dist/index.js
+pnpm run build  # builds web-server/web-ui artifacts and generates dist/index.js
 ```
 
 ### Test
 
 ```bash
 pnpm test            # all tests
+pnpm run test:coverage  # all tests with coverage (threshold: >=70%)
 pnpm run test:core   # core package
 pnpm run test:tools  # tools package
 pnpm run test:tui    # tui package
@@ -188,7 +288,7 @@ npm run format:check  # check format (CI)
 ## Project Structure
 
 ```text
-memo-cli/
+memo-code/
 ├── packages/
 │   ├── core/       # core logic: Session, tool routing, config
 │   ├── tools/      # built-in tool implementations
@@ -200,13 +300,15 @@ memo-cli/
 ## CLI Shortcuts and Commands
 
 - `/help`: show help and shortcut guide.
-- `/models`: list available Provider/Model entries and switch with Enter; also supports direct selection like `/models deepseek`.
-- `/context`: open 80k/120k/150k/200k options and apply immediately.
+- `/models`: list available Provider/Model entries and switch with Enter; also supports direct selection like `/models openai_compatible`.
+- `/review <prNumber>`: run GitHub PR review and publish review comments (uses active GitHub MCP server first, then falls back to `gh` CLI).
+- `/compact`: manually compact current session context.
 - `/mcp`: show configured MCP servers in current session.
 - `resume` history: type `resume` to list and load past sessions for current directory.
 - Exit and clear: `exit` / `/exit`, `Ctrl+L` for new session, `Esc Esc` to cancel current run or clear input.
 - **Tool approval**: risky operations open an approval dialog with `once`/`session`/`deny`.
 - **Approval reminder**: risky approval prompts ring a bell and attempt a desktop notification in interactive TUI.
+- **Context percent**: footer updates on each step (not just turn end), based on next-step prompt token estimate.
 
 > Session logs are written only when a session contains user messages, to avoid empty files.
 
@@ -220,7 +322,8 @@ memo-cli/
 
 ## Related Docs
 
-- [User Guide](./site/content/docs/README.md) - User-facing docs by module
+- [User Guide (EN)](./site/content/docs/en/README.mdx) - User-facing docs by module
+- [用户指南 (ZH)](./site/content/docs/zh/README.mdx) - 中文文档入口
 - [Core Architecture](./docs/core.md) - Core implementation details
 - [CLI Adaptation History](./docs/cli-update.md) - Historical migration notes (Tool Use API)
 - [Contributing](./CONTRIBUTING.md) - Contribution guide

package/README.zh.md

@@ -5,16 +5,20 @@
 </div>
 
 <p align="center">
-  <a href="public/memo-code-cli-show-01.mp4">
-    <img src="https://img.shields.io/badge/📹-观看演示视频-1a1a1a?style=for-the-badge" alt="Demo Video">
+  <a href="https://github.com/minorcell/memo-code/actions/workflows/ci.yml">
+    <img src="https://github.com/minorcell/memo-code/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI">
+  </a>
+  <a href="https://codecov.io/gh/minorcell/memo-code">
+    <img src="https://codecov.io/gh/minorcell/memo-code/graph/badge.svg" alt="Coverage">
+  </a>
+  <a href="https://www.npmjs.com/package/@memo-code/memo">
+    <img src="https://img.shields.io/npm/v/%40memo-code%2Fmemo" alt="npm version">
   </a>
 </p>
 
----
-
-<video src="public/memo-code-cli-show-01.mp4" width="100%"></video>
+<img src="public/demo.png" width="100%" alt="Memo Code 演示图">
 
-基于 Node.js + TypeScript，默认对接 DeepSeek，兼容 OpenAI API。
+基于 Node.js + TypeScript，兼容 OpenAI API 格式。
 
 Memo Code 是一个开源的终端编码代理，能够理解项目上下文，并通过自然语言协助你更快完成编码、排障和日常开发任务。
 
@@ -32,28 +36,50 @@ yarn global add @memo-code/memo
 bun add -g @memo-code/memo
 ```
 
+说明：npm 分发包是预构建产物，已包含运行所需的 CLI/Web 资源。  
+只有在直接运行本仓库源码时，才需要执行 `pnpm run build`。
+
 ### 2. 配置 API Key
 
 ```bash
-export DEEPSEEK_API_KEY=your_key  # 或 OPENAI_API_KEY
+export OPENAI_API_KEY=your_key
 ```
 
 ### 3. 启动使用
 
 ```bash
 memo
-# 首次运行会引导配置 provider/model，并（保存到 ~/.memo/config.toml）
+# 首次运行会引导配置 provider/model，并保存到 ~/.memo/config.toml
 ```
 
 ## 使用方式
 
 - 交互式：`memo`（默认 TUI，支持多轮、工具可视化、快捷键）。
 - 非交互纯文本模式（非 TTY）：`echo "你的问题" | memo`（适合脚本）。
+- 单次执行：`memo --once "你的问题"` 或 `memo -once "你的问题"`（执行一轮后退出；在可信仓库建议搭配 `--dangerous`）。
+- 继续最近会话：`memo --prev` 或 `memo -prev`（加载当前目录最近会话上下文）。
 - 危险模式：`memo --dangerous` 或 `memo -d`（跳过工具审批，谨慎使用）。
 - 查看版本：`memo --version` 或 `memo -v`。
+- 启动 Web 服务：`memo web --host 127.0.0.1 --port 5494 --open`（npm 分发包已包含 web 资源；源码运行需先 `pnpm run build`）。
 - 启动目录约定：若启动根目录存在 `AGENTS.md`，Memo 会自动将其拼接进系统提示词。
+- Skills：Memo 会自动发现 `SKILL.md` 并把可用 skills 列表拼接进系统提示词。
 - MCP 启动选择：当配置了 MCP server 时，启动会弹出多选以决定本次会话激活哪些 server。
-- 会话标题：Memo 会基于首条用户输入生成简短标题，并在历史/恢复列表中展示。
+- Web app 支持多 workspace 项目管理，并可并发运行多个会话（单个 server 进程上限 20）。
+
+## Web 控制台
+
+```bash
+memo web --host 127.0.0.1 --port 5494 --open
+```
+
+- npm 分发包已内置 web server 与 web UI 资源。
+- 若从源码仓库运行，请先执行一次 `pnpm run build`。
+- Web 认证配置默认存储在 `~/.memo/server.yaml`（可通过 `MEMO_SERVER_CONFIG` 覆盖路径）。
+- 首次启动会自动生成该文件，包含认证密钥和随机初始密码。
+- 登录页使用 `server.yaml` 中的 `auth.username` / `auth.password`。
+- 侧边栏包含独立的 `MCP Servers` 与 `Skills` 入口：
+    - MCP：创建/编辑/删除/登录/登出与激活开关。
+    - Skills：创建/删除、详情预览与激活开关。
 
 ## 配置文件
 
@@ -62,13 +88,13 @@ memo
 ### Provider 配置
 
 ```toml
-current_provider = "deepseek"
+current_provider = "openai_compatible"
 
-[[providers.deepseek]]
-name = "deepseek"
-env_api_key = "DEEPSEEK_API_KEY"
-model = "deepseek-chat"
-base_url = "https://api.deepseek.com"
+[[providers.openai_compatible]]
+name = "openai_compatible"
+env_api_key = "OPENAI_API_KEY"
+model = "gpt-4.1-mini"
+base_url = "https://api.openai.com/v1"
 ```
 
 支持配置多个 Provider，通过 `current_provider` 切换。
@@ -92,6 +118,11 @@ url = "https://your-mcp-server.com/mcp"
 # 可选：启动时默认激活的 MCP server
 active_mcp_servers = ["local_tools", "remote"]
 # 可选：设为 [] 表示启动时不激活任何 MCP server
+
+# 可选：MCP OAuth 凭据存储模式：auto | keyring | file
+mcp_oauth_credentials_store_mode = "auto"
+# 可选：为 `memo mcp login` 指定固定本地回调端口
+# mcp_oauth_callback_port = 33333
 ```
 
 也可以通过 CLI 管理 MCP 配置（对齐 Codex CLI 风格）：
@@ -106,11 +137,54 @@ memo mcp add local_tools -- /path/to/mcp-server --flag
 # 添加远程 MCP server（streamable HTTP）
 memo mcp add remote --url https://your-mcp-server.com/mcp --bearer-token-env-var MCP_TOKEN
 
+# streamable_http server 的 OAuth 登录/登出
+memo mcp login remote --scopes read,write
+memo mcp logout remote
+
 # 查看/删除
 memo mcp get remote
 memo mcp remove remote
 ```
 
+`memo mcp list` 会输出每个 server 的 `auth_status`：`unsupported`、`not_logged_in`、`bearer_token`、`oauth`。
+
+## Skills
+
+Memo 支持 Agent Skills，并会在启动时自动发现 `SKILL.md`。
+
+### 发现路径
+
+- 项目级：项目根目录下 `.<agent>/skills`（例如：`.agents/skills`、`.claude/skills`、`.codex/skills`）
+- 用户级：`$MEMO_HOME/skills`（或 `~/.memo/skills`）
+- 不扫描：Memo Home 之外的 `~/.xxx/skills` 隐藏目录
+
+### 最小 Skill 文件示例
+
+```md
+---
+name: doc-writing
+description: Generate and update technical documentation.
+---
+
+# Doc Writing
+```
+
+Memo 会读取 frontmatter 的 `name` 和 `description`，并以元数据形式注入：
+
+- `- <name>: <description> (file: <absolute-path-to-SKILL.md>)`
+
+在对话里可通过 `$skill-name` 显式提及某个 skill（例如 `$doc-writing`）。
+
+可选：在 `config.toml` 中持久化默认激活 skills：
+
+```toml
+# 未设置：默认激活所有已发现 skills
+# []：默认不激活任何 skill
+active_skills = [
+  "/absolute/path/to/.codex/skills/doc-writing/SKILL.md"
+]
+```
+
 ## 内置工具
 
 - `exec_command` / `write_stdin`：执行命令（默认执行工具族）
@@ -162,13 +236,14 @@ pnpm start
 ### 构建
 
 ```bash
-pnpm run build  # 生成 dist/index.js
+pnpm run build  # 构建 web-server/web-ui 产物并生成 dist/index.js
 ```
 
 ### 测试
 
 ```bash
 pnpm test            # 全量测试
+pnpm run test:coverage  # 带覆盖率测试（阈值：>=70%）
 pnpm run test:core   # 测试 core 包
 pnpm run test:tools  # 测试 tools 包
 pnpm run test:tui    # 测试 tui 包
@@ -184,7 +259,7 @@ npm run format:check  # 检查格式（CI）
 ## 项目结构
 
 ```
-memo-cli/
+memo-code/
 ├── packages/
 │   ├── core/       # 核心逻辑：Session、工具路由、配置
 │   ├── tools/      # 内置工具实现
@@ -196,8 +271,9 @@ memo-cli/
 ## CLI 快捷键与命令
 
 - `/help`：显示帮助与快捷键说明。
-- `/models`：列出现有 Provider/Model，回车切换；支持直接 `/models deepseek` 精确选择。
+- `/models`：列出现有 Provider/Model，回车切换；支持直接 `/models openai_compatible` 精确选择。
 - `/context`：弹出 80k/120k/150k/200k 选项并立即设置上限。
+- `/review <prNumber>`：执行 GitHub PR 审查并直接发布评论（优先使用已激活的 GitHub MCP，失败时回退 `gh` CLI）。
 - `/mcp`：查看当前会话加载的 MCP 服务器配置。
 - `resume` 历史：输入 `resume` 查看并加载本目录的历史会话。
 - 退出与清屏：`exit` / `/exit`，`Ctrl+L` 新会话，`Esc Esc` 取消运行或清空输入。
@@ -216,7 +292,8 @@ memo-cli/
 
 ## 相关文档
 
-- [用户指南](./site/content/docs/README.md) - 面向使用者的分模块说明
+- [User Guide (EN)](./site/content/docs/en/README.mdx) - 英文文档入口
+- [用户指南 (ZH)](./site/content/docs/zh/README.mdx) - 中文文档入口
 - [Core 架构](./docs/core.md) - 核心实现详解
 - [CLI 适配更新](./docs/cli-update.md) - Tool Use API 迁移说明
 - [开发指南](./CONTRIBUTING.md) - 贡献指南