chore: release v1.116.3

docs: replace Claude Code references with Agent in READMEs
chore: release v1.116.2
2026-07-12 22:09:48 +08:00 · 2026-05-13 22:33:29 -05:00 · 2026-05-13 22:33:16 -05:00 · 2026-05-13 22:30:54 -05:00 · 2026-05-13 22:30:38 -05:00 · 2026-05-13 22:28:34 -05:00
33 changed files with 1811 additions and 114 deletions
@@ -6,7 +6,7 @@
  },
  "metadata": {
    "description": "Skills shared by Baoyu for improving daily work efficiency",
-    "version": "1.114.0"
+    "version": "1.116.3"
  },
  "plugins": [
    {
@@ -33,7 +33,8 @@
        "./skills/baoyu-translate",
        "./skills/baoyu-url-to-markdown",
        "./skills/baoyu-image-cards",
-        "./skills/baoyu-youtube-transcript"
+        "./skills/baoyu-youtube-transcript",
+        "./skills/baoyu-wechat-summary"
      ]
    }
  ]
@@ -173,3 +173,6 @@ diagram/
 .worktrees/
 youtube-transcript/
 .omx/
+.codex-tmp/
+outputs/
+wechat/
@@ -2,6 +2,58 @@

 English | [中文](./CHANGELOG.zh.md)

+## 1.116.3 - 2026-05-13
+
+### Documentation
+- Replace Claude Code references with generic Agent wording in READMEs to reflect multi-agent support (Claude Code, Codex, etc.)
+
+## 1.116.2 - 2026-05-13
+
+### Documentation
+- `baoyu-wechat-summary`: update example group name in SKILL.md
+
+## 1.116.1 - 2026-05-13
+
+### Features
+- `baoyu-wechat-summary`: add `data_root` option to first-time setup flow, allowing users to customize the digest output directory during initialization
+
+## 1.116.0 - 2026-05-13
+
+### Features
+- Add `baoyu-wechat-summary` skill: summarize WeChat group chat highlights into structured digests with topic extraction, message leaderboards, and per-user profiles. Supports normal and roast (毒舌) versions, incremental mode, and profile backfill. Requires [wx-cli](https://github.com/jackwener/wx-cli).
+
+## 1.115.4 - 2026-05-11
+
+### Documentation
+- Image generation backend selection: emphasize Codex `imagegen` as the priority runtime-native tool (invoke via the `Skill` tool with `skill: "imagegen"`) and forbid SVG/HTML/canvas substitution when no raster backend can be resolved — fall through to asking the user instead of silently emitting code-based art. Updated in `docs/image-generation-tools.md` and inlined into `baoyu-article-illustrator`, `baoyu-comic`, `baoyu-cover-image`, `baoyu-image-cards`, `baoyu-infographic`, `baoyu-slide-deck`, and `baoyu-xhs-images`.
+
+## 1.115.3 - 2026-05-11
+
+### Fixes
+- `baoyu-post-to-wechat`: ensure tab activation before copy/paste in WeChat editor (by @fengxiaodong28)
+- `baoyu-post-to-x`: use toolbar media upload instead of image clipboard paste for X Articles
+
+## 1.115.2 - 2026-05-10
+
+### Fixes
+- `baoyu-post-to-x`: honor explicit Codex Chrome plugin requests as a distinct browser-control mode, keep Chrome Computer Use and CDP fallbacks from silently taking over, and improve X Articles draft creation detection.
+
+## 1.115.1 - 2026-05-10
+
+### Fixes
+- `baoyu-imagine`: change the default MiniMax image API endpoint to `https://api.minimaxi.com` to match the current official image generation documentation, while keeping `https://api.minimax.io` available through `MINIMAX_BASE_URL` overrides.
+- `baoyu-image-gen`: sync the deprecated image-generation entrypoint with the same MiniMax default endpoint and regression coverage.
+
+## 1.115.0 - 2026-05-09
+
+### Features
+- `baoyu-post-to-x`: add Chrome Computer Use as the preferred execution mode in Codex. When Computer Use tools are available, all X UI actions (compose, article, quote, video) go through the user's real Chrome window instead of CDP scripts. CDP scripts become a fallback when Computer Use is unavailable or explicitly not requested.
+
+## 1.114.1 - 2026-05-08
+
+### Fixes
+- `baoyu-danger-gemini-web`: restore generated-image extraction for current Gemini Web responses where image URLs appear as `https://lh3.googleusercontent.com/gg-dl/` without the legacy generated-image markers. Adds regression coverage for the fallback response shape. (by @evilstar2016)
+
 ## 1.114.0 - 2026-05-05

 ### Features
@@ -2,6 +2,58 @@

 [English](./CHANGELOG.md) | 中文

+## 1.116.3 - 2026-05-13
+
+### 文档
+- README 中将 Claude Code 替换为通用的 Agent 表述，体现多 Agent 支持（Claude Code、Codex 等）
+
+## 1.116.2 - 2026-05-13
+
+### 文档
+- `baoyu-wechat-summary`：更新 SKILL.md 中的示例群名
+
+## 1.116.1 - 2026-05-13
+
+### 新功能
+- `baoyu-wechat-summary`：初始化设置流程中新增 `data_root` 选项，允许用户在首次配置时自定义摘要输出目录
+
+## 1.116.0 - 2026-05-13
+
+### 新功能
+- 新增 `baoyu-wechat-summary` 技能：将微信群聊精华提炼为结构化简报，支持话题提取、发言排行榜和群友画像。可生成正常版和毒舌版，支持增量模式和画像回溯初始化。需安装 [wx-cli](https://github.com/jackwener/wx-cli)。
+
+## 1.115.4 - 2026-05-11
+
+### 文档
+- 图片生成后端选择规则强化：明确将 Codex `imagegen` 作为运行时原生工具的优先项（通过 `Skill` 工具调用，`skill: "imagegen"`），并禁止在无可用光栅后端时降级为 SVG/HTML/canvas 等代码渲染 —— 应退回到询问用户，而非静默输出代码绘图。规则同步更新到 `docs/image-generation-tools.md`，并按自包含规则内联到 `baoyu-article-illustrator`、`baoyu-comic`、`baoyu-cover-image`、`baoyu-image-cards`、`baoyu-infographic`、`baoyu-slide-deck`、`baoyu-xhs-images`。
+
+## 1.115.3 - 2026-05-11
+
+### 修复
+- `baoyu-post-to-wechat`：修复微信编辑器中复制粘贴前未激活标签页的问题 (by @fengxiaodong28)
+- `baoyu-post-to-x`：X 文章图片插入改用工具栏媒体上传替代剪贴板粘贴方式
+
+## 1.115.2 - 2026-05-10
+
+### 修复
+- `baoyu-post-to-x`：将显式请求 Codex Chrome 插件的场景作为独立浏览器控制模式处理，避免 Chrome Computer Use 或 CDP 回退流程静默接管；同时改进 X Articles 草稿创建按钮检测。
+
+## 1.115.1 - 2026-05-10
+
+### 修复
+- `baoyu-imagine`：将默认 MiniMax 图片 API 端点改为 `https://api.minimaxi.com`，与当前官方图片生成文档保持一致；仍可通过 `MINIMAX_BASE_URL` 覆盖为 `https://api.minimax.io`。
+- `baoyu-image-gen`：同步已废弃图片生成入口的 MiniMax 默认端点和回归测试。
+
+## 1.115.0 - 2026-05-09
+
+### 新功能
+- `baoyu-post-to-x`：新增 Chrome Computer Use 作为 Codex 环境下的首选执行模式。当 Computer Use 工具可用时，所有 X 界面操作（发帖、文章、引用、视频）均通过用户真实 Chrome 窗口完成，不再使用 CDP 脚本。CDP 脚本降级为 Computer Use 不可用或用户明确要求时的回退方案。
+
+## 1.114.1 - 2026-05-08
+
+### 修复
+- `baoyu-danger-gemini-web`：修复当前 Gemini Web 响应中生成图 URL 以 `https://lh3.googleusercontent.com/gg-dl/` 形式出现、但不再包含旧版生成图 marker 时的图片提取失败问题。补充该响应形态的回归测试。 (by @evilstar2016)
+
 ## 1.114.0 - 2026-05-05

 ### 新功能
@@ -2,7 +2,7 @@

 English | [中文](./README.zh.md)

-Skills shared by Baoyu for improving daily work efficiency with Claude Code.
+Skills shared by Baoyu for improving daily work efficiency with AI Agents (Claude Code, Codex, etc.).

 ## Prerequisites

@@ -40,7 +40,7 @@ Publishing to ClawHub releases the published skill under `MIT-0`, per ClawHub's

 ### Register as Plugin Marketplace

-Run the following command in Claude Code:
+Run the following command in the Agent:

 ```bash
 /plugin marketplace add JimLiu/baoyu-skills
@@ -64,7 +64,7 @@ Run the following command in Claude Code:

 **Option 3: Ask the Agent**

-Simply tell Claude Code:
+Simply tell the Agent:

 > Please install Skills from github.com/JimLiu/baoyu-skills

@@ -80,7 +80,7 @@ The marketplace now exposes a single plugin so each skill is registered exactly

 To update skills to the latest version:

-1. Run `/plugin` in Claude Code
+1. Run `/plugin` in the Agent
 2. Switch to **Marketplaces** tab (use arrow keys or Tab)
 3. Select **baoyu-skills**
 4. Choose **Update marketplace**
@@ -579,7 +579,7 @@ Plain text input is treated as a regular post. Markdown files are treated as X A

 ```bash
 # Post with text
-/baoyu-post-to-x "Hello from Claude Code!"
+/baoyu-post-to-x "Hello from AI Agent!"

 # Post with images
 /baoyu-post-to-x "Check this out" --image photo.png
@@ -834,7 +834,7 @@ AI SDK-based image generation using OpenAI GPT Image 2, Azure OpenAI, Google, Op
 | `DASHSCOPE_BASE_URL` | Custom DashScope endpoint | - |
 | `ZAI_BASE_URL` | Custom Z.AI endpoint | `https://api.z.ai/api/paas/v4` |
 | `BIGMODEL_BASE_URL` | Backward-compatible alias for Z.AI endpoint | - |
-| `MINIMAX_BASE_URL` | Custom MiniMax endpoint | `https://api.minimax.io` |
+| `MINIMAX_BASE_URL` | Custom MiniMax endpoint | `https://api.minimaxi.com` |
 | `REPLICATE_BASE_URL` | Custom Replicate endpoint | - |
 | `JIMENG_BASE_URL` | Custom Jimeng endpoint | `https://visual.volcengineapi.com` |
 | `JIMENG_REGION` | Jimeng region | `cn-north-1` |
@@ -1112,6 +1112,36 @@ Custom style descriptions are also accepted, e.g., `--style "poetic and lyrical"
 - Translator's notes for cultural/domain-specific references
 - Output directory with all intermediate files preserved

+#### baoyu-wechat-summary
+
+Summarize WeChat group chat highlights into a structured digest. Extracts topics, quotes, and stats from group messages using [wx-cli](https://github.com/jackwener/wx-cli). Maintains per-group history and per-user profiles across runs. Supports normal and roast (毒舌) versions.
+
+```bash
+# Summarize a group's recent messages
+/baoyu-wechat-summary 相亲相爱一家人 最近 1 天
+
+# Weekly summary
+/baoyu-wechat-summary AI 技术群 最近 7 天
+
+# Incremental (since last digest)
+/baoyu-wechat-summary 相亲相爱一家人
+
+# Roast version
+/baoyu-wechat-summary 相亲相爱一家人 最近 3 天 毒舌版
+```
+
+**Requirements**:
+- [wx-cli](https://github.com/jackwener/wx-cli) installed (`npm install -g @jackwener/wx-cli`)
+- WeChat 4.x running and logged in on macOS
+
+**Features**:
+- Topic extraction with attribution and quotes
+- Message leaderboard and per-user profiles
+- Incremental mode (picks up where last digest left off)
+- Multi-day range splitting for large batches
+- Normal and roast (毒舌) digest versions
+- Profile backfill from historical digests
+
 ## Environment Configuration

 Some skills require API keys or custom configuration. Environment variables can be set in `.env` files:
@@ -1167,7 +1197,7 @@ ZAI_IMAGE_MODEL=glm-image
 # MiniMax
 MINIMAX_API_KEY=xxx
 MINIMAX_IMAGE_MODEL=image-01
-# MINIMAX_BASE_URL=https://api.minimax.io
+# MINIMAX_BASE_URL=https://api.minimaxi.com

 # Replicate
 REPLICATE_API_TOKEN=r8_xxx
@@ -2,7 +2,7 @@

 [English](./README.md) | 中文

-宝玉分享的 Claude Code 技能集，提升日常工作效率。
+宝玉分享的 AI Agent 技能集（适用于 Claude Code、Codex 等），提升日常工作效率。

 ## 前置要求

@@ -40,7 +40,7 @@ clawhub install baoyu-markdown-to-html

 ### 注册插件市场

-在 Claude Code 中运行：
+在 Agent 中运行：

 ```bash
 /plugin marketplace add JimLiu/baoyu-skills
@@ -64,7 +64,7 @@ clawhub install baoyu-markdown-to-html

 **方式三：告诉 Agent**

-直接告诉 Claude Code：
+直接告诉 Agent：

 > 请帮我安装 github.com/JimLiu/baoyu-skills 中的 Skills

@@ -80,7 +80,7 @@ clawhub install baoyu-markdown-to-html

 更新技能到最新版本：

-1. 在 Claude Code 中运行 `/plugin`
+1. 在 Agent 中运行 `/plugin`
 2. 切换到 **Marketplaces** 标签页（使用方向键或 Tab）
 3. 选择 **baoyu-skills**
 4. 选择 **Update marketplace**
@@ -579,7 +579,7 @@ clawhub install baoyu-markdown-to-html

 ```bash
 # 发布文字
-/baoyu-post-to-x "Hello from Claude Code!"
+/baoyu-post-to-x "Hello from AI Agent!"

 # 发布带图片
 /baoyu-post-to-x "看看这个" --image photo.png
@@ -834,7 +834,7 @@ AI 驱动的生成后端。
 | `DASHSCOPE_BASE_URL` | 自定义 DashScope 端点 | - |
 | `ZAI_BASE_URL` | 自定义 Z.AI 端点 | `https://api.z.ai/api/paas/v4` |
 | `BIGMODEL_BASE_URL` | Z.AI 端点向后兼容别名 | - |
-| `MINIMAX_BASE_URL` | 自定义 MiniMax 端点 | `https://api.minimax.io` |
+| `MINIMAX_BASE_URL` | 自定义 MiniMax 端点 | `https://api.minimaxi.com` |
 | `REPLICATE_BASE_URL` | 自定义 Replicate 端点 | - |
 | `JIMENG_BASE_URL` | 自定义即梦端点 | `https://visual.volcengineapi.com` |
 | `JIMENG_REGION` | 即梦区域 | `cn-north-1` |
@@ -1112,6 +1112,36 @@ AI 驱动的生成后端。
 - 为文化/专业术语添加译注
 - 输出目录保留所有中间文件

+#### baoyu-wechat-summary
+
+微信群聊精华提取。使用 [wx-cli](https://github.com/jackwener/wx-cli) 从群消息中提取话题、引言和统计数据，生成结构化简报。支持跨次运行的群聊历史和群友画像维护，可生成正常版和毒舌版。
+
+```bash
+# 总结群最近消息
+/baoyu-wechat-summary 相亲相爱一家人 最近 1 天
+
+# 周报
+/baoyu-wechat-summary AI 技术群 最近 7 天
+
+# 增量模式（从上次摘要继续）
+/baoyu-wechat-summary 相亲相爱一家人
+
+# 毒舌版
+/baoyu-wechat-summary 相亲相爱一家人 最近 3 天 毒舌版
+```
+
+**前置要求**：
+- 安装 [wx-cli](https://github.com/jackwener/wx-cli)（`npm install -g @jackwener/wx-cli`）
+- macOS 上运行并登录 WeChat 4.x
+
+**特性**：
+- 话题提取，带归属和引言
+- 发言排行榜和群友画像
+- 增量模式（从上次摘要断点继续）
+- 大批量消息自动按天分割
+- 正常版和毒舌版两种风格
+- 支持从历史摘要回溯初始化画像
+
 ## 环境配置

 部分技能需要 API 密钥或自定义配置。环境变量可以在 `.env` 文件中设置：
@@ -1167,7 +1197,7 @@ ZAI_IMAGE_MODEL=glm-image
 # MiniMax
 MINIMAX_API_KEY=xxx
 MINIMAX_IMAGE_MODEL=image-01
-# MINIMAX_BASE_URL=https://api.minimax.io
+# MINIMAX_BASE_URL=https://api.minimaxi.com

 # Replicate
 REPLICATE_API_TOKEN=r8_xxx
@@ -9,11 +9,14 @@ When a skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if the skill's `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends.

 ## The Preference Field
@@ -28,11 +28,14 @@ When this skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends. Users change the pinned backend via the `## Changing Preferences` section below.

 **Prompt file requirement (hard)**: write each image's full, final prompt to a standalone file under `prompts/` (naming: `NN-{type}-[slug].md`) BEFORE invoking any backend. The backend receives the prompt file (or its content); the file is the reproducibility record and lets you switch backends without regenerating prompts.
@@ -345,7 +345,14 @@ Prompt Files:

 ### 5.2 Select Generation Skill

-Check available skills. If multiple, ask user.
+Follow the `## Image Generation Tools` rule at the top of `SKILL.md`. Concretely:
+
+- If `imagegen` is in your available-skills list (Codex), use it — invoke via the `Skill` tool with `skill: "imagegen"`.
+- Else if the EXTEND.md pin is available, use it.
+- Else if exactly one non-native backend is installed, use it.
+- Else, ask the user.
+
+**Do not generate SVG, HTML, or any code-based vector as a substitute for the raster image.** If no raster backend can be resolved, ask the user how to proceed.

 ### 5.3 Process References ⚠️ REQUIRED if references saved in Step 1.0

@@ -32,11 +32,14 @@ When this skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends. Users change the pinned backend via the `## Changing Preferences` section below.

 **Prompt file requirement (hard)**: write each image's full, final prompt to a standalone file under `prompts/` (naming: `NN-{type}-[slug].md`) BEFORE invoking any backend. The backend receives the prompt file (or its content); the file is the reproducibility record and lets you switch backends without regenerating prompts.
@@ -28,11 +28,14 @@ When this skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends. Users change the pinned backend via the `## Changing Preferences` section below.

 **Prompt file requirement (hard)**: write each image's full, final prompt to a standalone file under `prompts/` (naming: `NN-{type}-[slug].md`) BEFORE invoking any backend. The backend receives the prompt file (or its content); the file is the reproducibility record and lets you switch backends without regenerating prompts.
@@ -0,0 +1,25 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { collect_generated_image_urls_from_response_parts } from "./client.ts";
+
+test("response part fallback finds generated images when legacy generated markers are absent", () => {
+  const generatedUrl = "https://lh3.googleusercontent.com/gg-dl/example-generated-image";
+  const initialCandidate = ["rcid-1", ["image generated successfully"]];
+  const imageCandidate = [
+    "rcid-1",
+    ["image generated successfully"],
+    { nestedPayload: [{ media: generatedUrl }] },
+  ];
+  const responseJson = [
+    ["wrb.fr", null, JSON.stringify([null, [], null, null, [initialCandidate]])],
+    ["wrb.fr", null, JSON.stringify([null, [], null, null, [imageCandidate]])],
+  ];
+
+  assert.equal(initialCandidate[12], undefined);
+  assert.equal(
+    /http:\/\/googleusercontent\.com\/image_generation_content\/\d+/.test(String(initialCandidate[1]?.[0])),
+    false,
+  );
+  assert.deepEqual(collect_generated_image_urls_from_response_parts(responseJson, 0, 0), [generatedUrl]);
+});
@@ -37,6 +37,8 @@ type InitOptions = {

 type RequestKwargs = RequestInit & { timeout_ms?: number };

+const GENERATED_IMAGE_URL_PREFIX = 'https://lh3.googleusercontent.com/gg-dl/';
+
 function normalize_headers(h?: HeadersInit): Record<string, string> {
  if (!h) return {};
  if (Array.isArray(h)) return Object.fromEntries(h.map(([k, v]) => [k, v]));
@@ -78,6 +80,59 @@ function collect_strings(root: unknown, accept: (s: string) => boolean, limit: n
  return out;
 }

+function collect_generated_image_urls(root: unknown, limit: number = 4): string[] {
+  return collect_strings(root, (s) => s.startsWith(GENERATED_IMAGE_URL_PREFIX), limit);
+}
+
+function parse_response_part_body(part: unknown): unknown[] | null {
+  if (!Array.isArray(part)) return null;
+  const part_body = get_nested_value<string | null>(part, [2], null);
+  if (!part_body) return null;
+  try {
+    const part_json = JSON.parse(part_body) as unknown;
+    return Array.isArray(part_json) ? part_json : null;
+  } catch {
+    return null;
+  }
+}
+
+function find_generated_image_part(
+  response_json: unknown[],
+  body_index: number,
+  candidate_index: number,
+  limit: number = 4,
+): { body: unknown[]; urls: string[] } | null {
+  for (let part_index = body_index; part_index < response_json.length; part_index++) {
+    const part_json = parse_response_part_body(response_json[part_index]);
+    if (!part_json) continue;
+    const cand = get_nested_value<unknown>(part_json, [4, candidate_index], null);
+    if (!cand) continue;
+    const urls = collect_generated_image_urls(cand, limit);
+    if (urls.length > 0) return { body: part_json, urls };
+  }
+  return null;
+}
+
+export function collect_generated_image_urls_from_response_parts(
+  response_json: unknown[],
+  body_index: number,
+  candidate_index: number,
+  limit: number = 4,
+): string[] {
+  return find_generated_image_part(response_json, body_index, candidate_index, limit)?.urls ?? [];
+}
+
+function push_generated_images(
+  generated_images: GeneratedImage[],
+  urls: string[],
+  proxy: string | null,
+  cookies: Record<string, string>,
+): void {
+  for (const url of urls) {
+    generated_images.push(new GeneratedImage(url, '[Generated Image]', '', proxy, cookies));
+  }
+}
+
 export class GeminiClient extends GemMixin {
  public cookies: Record<string, string> = {};
  public proxy: string | null = null;
@@ -404,24 +459,8 @@ export class GeminiClient extends GemMixin {
            /http:\/\/googleusercontent\.com\/image_generation_content\/\d+/.test(text);

          if (wants_generated) {
-            let img_body: unknown[] | null = null;
-            for (let part_index = body_index; part_index < (response_json as unknown[]).length; part_index++) {
-              const part = (response_json as unknown[])[part_index];
-              if (!Array.isArray(part)) continue;
-              const part_body = get_nested_value<string | null>(part, [2], null);
-              if (!part_body) continue;
-              try {
-                const part_json = JSON.parse(part_body) as unknown[];
-                const cand = get_nested_value<unknown>(part_json, [4, candidate_index], null);
-                if (!cand) continue;
-
-                const urls = collect_strings(cand, (s) => s.startsWith('https://lh3.googleusercontent.com/gg-dl/'), 1);
-                if (urls.length > 0) {
-                  img_body = part_json;
-                  break;
-                }
-              } catch {}
-            }
+            const image_part = find_generated_image_part(response_json as unknown[], body_index, candidate_index, 1);
+            const img_body = image_part?.body ?? null;

            if (!img_body) {
              throw new ImageGenerationError(
@@ -452,13 +491,22 @@ export class GeminiClient extends GemMixin {
            }

            if (generated_images.length === 0) {
-              const urls = collect_strings(img_candidate, (s) => s.startsWith('https://lh3.googleusercontent.com/gg-dl/'), 4);
-              for (const url of urls) {
-                generated_images.push(new GeneratedImage(url, '[Generated Image]', '', this.proxy, this.cookies));
-              }
+              push_generated_images(generated_images, collect_generated_image_urls(img_candidate), this.proxy, this.cookies);
            }
          }

+          // Fallback: unconditionally scan all response parts for generated image URLs.
+          // The `wants_generated` detection above relies on `candidate[12][7][0]` and an old
+          // `googleusercontent.com/image_generation_content/` URL pattern, both of which no
+          // longer appear in the current Gemini Web API response format.
+          // When Gemini does generate images, their URLs now start with
+          // `https://lh3.googleusercontent.com/gg-dl/` and are present somewhere in the
+          // response parts — this fallback finds them so images are not silently dropped.
+          if (generated_images.length === 0) {
+            const urls = collect_generated_image_urls_from_response_parts(response_json as unknown[], body_index, candidate_index);
+            push_generated_images(generated_images, urls, this.proxy, this.cookies);
+          }
+
          out.push(new Candidate({ rcid, text, thoughts, web_images, generated_images }));
        }

@@ -28,11 +28,14 @@ When this skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends. Users change the pinned backend via the `## Changing Preferences` section below.

 **Prompt file requirement (hard)**: write each image's full, final prompt to a standalone file under `prompts/` (naming: `NN-{type}-[slug].md`) BEFORE invoking any backend. The file is the reproducibility record and lets you switch backends without regenerating prompts.
@@ -24,6 +24,6 @@ Read when the user picks `--provider minimax` or sets `default_model.minimax`. D

 ## Official References

- [Image Generation Guide](https://platform.minimax.io/docs/guides/image-generation)
- [Text-to-Image API](https://platform.minimax.io/docs/api-reference/image-generation-t2i)
- [Image-to-Image API](https://platform.minimax.io/docs/api-reference/image-generation-i2i)
+- [Image Generation Guide](https://platform.minimaxi.com/docs/guides/image-generation)
+- [Text-to-Image API](https://platform.minimaxi.com/docs/api-reference/image-generation-t2i)
+- [Image-to-Image API](https://platform.minimaxi.com/docs/api-reference/image-generation-i2i)
@@ -60,8 +60,11 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
  };
 }

-test("MiniMax URL builder normalizes /v1 suffixes", (t) => {
-  useEnv(t, { MINIMAX_BASE_URL: "https://api.minimax.io" });
+test("MiniMax URL builder uses documented default and normalizes /v1 suffixes", (t) => {
+  useEnv(t, { MINIMAX_BASE_URL: null });
+  assert.equal(buildMinimaxUrl(), "https://api.minimaxi.com/v1/image_generation");
+
+  process.env.MINIMAX_BASE_URL = "https://api.minimax.io";
  assert.equal(buildMinimaxUrl(), "https://api.minimax.io/v1/image_generation");

  process.env.MINIMAX_BASE_URL = "https://proxy.example.com/custom/v1/";
@@ -44,7 +44,7 @@ function getApiKey(): string | null {
 }

 export function buildMinimaxUrl(): string {
-  const base = (process.env.MINIMAX_BASE_URL || "https://api.minimax.io").replace(/\/+$/g, "");
+  const base = (process.env.MINIMAX_BASE_URL || "https://api.minimaxi.com").replace(/\/+$/g, "");
  return base.endsWith("/v1") ? `${base}/image_generation` : `${base}/v1/image_generation`;
 }

@@ -197,7 +197,7 @@ export async function generateImage(
 ): Promise<Uint8Array> {
  const apiKey = getApiKey();
  if (!apiKey) {
-    throw new Error("MINIMAX_API_KEY is required. Get one from https://platform.minimax.io/");
+    throw new Error("MINIMAX_API_KEY is required. Get one from https://platform.minimaxi.com/");
  }

  const body = await buildRequestBody(prompt, model, args);
@@ -24,6 +24,6 @@ Read when the user picks `--provider minimax` or sets `default_model.minimax`. D

 ## Official References

- [Image Generation Guide](https://platform.minimax.io/docs/guides/image-generation)
- [Text-to-Image API](https://platform.minimax.io/docs/api-reference/image-generation-t2i)
- [Image-to-Image API](https://platform.minimax.io/docs/api-reference/image-generation-i2i)
+- [Image Generation Guide](https://platform.minimaxi.com/docs/guides/image-generation)
+- [Text-to-Image API](https://platform.minimaxi.com/docs/api-reference/image-generation-t2i)
+- [Image-to-Image API](https://platform.minimaxi.com/docs/api-reference/image-generation-i2i)
@@ -61,8 +61,11 @@ function makeArgs(overrides: Partial<CliArgs> = {}): CliArgs {
  };
 }

-test("MiniMax URL builder normalizes /v1 suffixes", (t) => {
-  useEnv(t, { MINIMAX_BASE_URL: "https://api.minimax.io" });
+test("MiniMax URL builder uses documented default and normalizes /v1 suffixes", (t) => {
+  useEnv(t, { MINIMAX_BASE_URL: null });
+  assert.equal(buildMinimaxUrl(), "https://api.minimaxi.com/v1/image_generation");
+
+  process.env.MINIMAX_BASE_URL = "https://api.minimax.io";
  assert.equal(buildMinimaxUrl(), "https://api.minimax.io/v1/image_generation");

  process.env.MINIMAX_BASE_URL = "https://proxy.example.com/custom/v1/";
@@ -44,7 +44,7 @@ function getApiKey(): string | null {
 }

 export function buildMinimaxUrl(): string {
-  const base = (process.env.MINIMAX_BASE_URL || "https://api.minimax.io").replace(/\/+$/g, "");
+  const base = (process.env.MINIMAX_BASE_URL || "https://api.minimaxi.com").replace(/\/+$/g, "");
  return base.endsWith("/v1") ? `${base}/image_generation` : `${base}/v1/image_generation`;
 }

@@ -197,7 +197,7 @@ export async function generateImage(
 ): Promise<Uint8Array> {
  const apiKey = getApiKey();
  if (!apiKey) {
-    throw new Error("MINIMAX_API_KEY is required. Get one from https://platform.minimax.io/");
+    throw new Error("MINIMAX_API_KEY is required. Get one from https://platform.minimaxi.com/");
  }

  const body = await buildRequestBody(prompt, model, args);
@@ -28,11 +28,14 @@ When this skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends. Users change the pinned backend via the `## Changing Preferences` section below.

 **Prompt file requirement (hard)**: write each image's full, final prompt to a standalone file under `prompts/` (naming: `NN-{type}-[slug].md`) BEFORE invoking any backend. The backend receives the prompt file (or its content); the file is the reproducibility record and lets you switch backends without regenerating prompts.
@@ -180,6 +180,10 @@ async function copyHtmlFromBrowser(cdp: CdpConnection, htmlFilePath: string, con
  }, { sessionId });
  await sleep(300);

+  console.log('[wechat] Activating HTML tab for copy...');
+  await cdp.send('Target.activateTarget', { targetId });
+  await sleep(300);
+
  console.log('[wechat] Copying content...');
  await sendCopy(cdp, sessionId);
  await sleep(1000);
@@ -189,6 +193,11 @@ async function copyHtmlFromBrowser(cdp: CdpConnection, htmlFilePath: string, con
 }

 async function pasteFromClipboardInEditor(session: ChromeSession): Promise<void> {
+  console.log('[wechat] Activating editor tab for paste...');
+  if (session.targetId) {
+    await session.cdp.send('Target.activateTarget', { targetId: session.targetId });
+    await sleep(300);
+  }
  console.log('[wechat] Pasting content...');
  await sendPaste(session.cdp, session.sessionId);
  await sleep(1000);
@@ -1,7 +1,7 @@
 ---
 name: baoyu-post-to-x
-description: Posts content and articles to X (Twitter). Supports regular posts with images/videos and X Articles (long-form Markdown). Uses real Chrome with CDP to bypass anti-automation. Use when user asks to "post to X", "tweet", "publish to Twitter", or "share on X".
-version: 1.56.2
+description: Posts content and articles to X (Twitter). Supports regular posts with images/videos and X Articles (long-form Markdown). In Codex, honor explicit requests for the Codex Chrome plugin/@chrome by using the Chrome Extension workflow; otherwise use Chrome Computer Use when available and fall back to real Chrome CDP scripts only when allowed. Use when user asks to "post to X", "tweet", "publish to Twitter", or "share on X".
+version: 1.57.2
 metadata:
  openclaw:
    homepage: https://github.com/JimLiu/baoyu-skills#baoyu-post-to-x
@@ -13,7 +13,12 @@ metadata:

 # Post to X (Twitter)

-Posts text, images, videos, and long-form articles to X via real Chrome browser (bypasses anti-bot detection).
+Posts text, images, videos, and long-form articles to X via a real Chrome browser.
+
+In Codex, do not conflate these browser paths:
+- **Codex Chrome plugin / `@chrome` / Chrome Extension**: use the bundled `chrome:Chrome` skill and its Node REPL browser client. This is required whenever the user says "Codex Chrome plugin", "Codex 自带的 Chrome 插件", `@chrome`, or similar.
+- **Chrome Computer Use**: use `mcp__computer_use__.*` against the visible Google Chrome UI only when the user asks for Computer Use or no Chrome-plugin preference is stated and Computer Use is available.
+- **CDP script mode**: use only as a fallback when the selected mode is unavailable or the user explicitly asks for CDP/script mode.

 ## Script Directory

@@ -28,15 +33,67 @@ Posts text, images, videos, and long-form articles to X via real Chrome browser
 **Script Reference**:
 | Script | Purpose |
 |--------|---------|
-| `scripts/x-browser.ts` | Regular posts (text + images) |
-| `scripts/x-video.ts` | Video posts (text + video) |
-| `scripts/x-quote.ts` | Quote tweet with comment |
-| `scripts/x-article.ts` | Long-form article publishing (Markdown) |
+| `scripts/x-browser.ts` | Regular posts (text + images), CDP fallback |
+| `scripts/x-video.ts` | Video posts (text + video), CDP fallback |
+| `scripts/x-quote.ts` | Quote tweet with comment, CDP fallback |
+| `scripts/x-article.ts` | Long-form article publishing (Markdown), CDP fallback |
 | `scripts/md-to-html.ts` | Markdown → HTML conversion |
 | `scripts/copy-to-clipboard.ts` | Copy content to clipboard |
 | `scripts/paste-from-clipboard.ts` | Send real paste keystroke |
 | `scripts/check-paste-permissions.ts` | Verify environment & permissions |

+## Execution Mode Selection (Required)
+
+Choose exactly one mode before interacting with X:
+
+1. If the user explicitly asks for the Codex Chrome plugin, `@chrome`, the Chrome extension, or "Codex 自带的 Chrome 插件", use **Codex Chrome Plugin Mode**. Do not call Computer Use first.
+2. If the user explicitly asks for Chrome Computer Use, use **Chrome Computer Use Mode**. Do not fall back to CDP, Playwright, the in-app Browser, or the Chrome plugin without telling the user and getting approval.
+3. If the user explicitly asks for CDP/script mode, use **CDP Script Mode**.
+4. Otherwise, prefer **Chrome Computer Use Mode**. For Markdown **X Articles with local content images**, use the tested X editor flow: insert each body image from the toolbar (`Insert` -> `Media` -> dialog icon button `Add photos or video`) at its placeholder, then delete the placeholder text. Use CDP Script Mode only when the selected browser-control mode is unavailable or the UI upload/selection flow is unreliable.
+
+Never use the in-app Browser for X publishing workflows.
+
+## Codex Chrome Plugin Mode
+
+Use this mode whenever the user requests the Codex Chrome plugin, `@chrome`, or the Chrome Extension path. This uses the user's real Chrome profile and X login through the bundled Chrome plugin, not Computer Use and not CDP.
+
+**Setup**
+1. Load the `chrome:Chrome` skill before browser work.
+2. Use `tool_search` for `node_repl js` if the Node REPL `js` tool is not already visible.
+3. Initialize the Chrome browser client exactly as the Chrome skill specifies, then run a lightweight call such as `browser.user.openTabs()` to verify the extension connection.
+4. If the first lightweight call fails, wait 2 seconds and retry once. If it still fails, follow the Chrome skill's extension checks and recovery steps. If checks pass but communication still fails, ask the user before opening a new Chrome window. Do not switch to Computer Use or CDP silently.
+
+**General rules**
+- Use the Chrome plugin's `browser.tabs.*`, `tab.playwright.*`, `tab.cua.*`, and file chooser APIs for X UI actions.
+- Shell commands are allowed for Markdown preprocessing and rich-HTML clipboard preparation. For X Article body images, do not rely on image clipboard paste; use the editor's `Insert` -> `Media` upload flow.
+- If a file upload fails with `Not allowed`, tell the user: `To enable file upload, go to chrome://extensions in Chrome, click Details under the Codex extension, and enable "Allow access to file URLs." See https://developers.openai.com/codex/app/chrome-extension#upload-files for details.`
+- If the Chrome plugin reports `native pipe is closed`, retry the lightweight browser call once after 2 seconds, then run the Chrome skill health checks. If Chrome is running, the extension is enabled, and the native host manifest is correct, ask permission to open a new Chrome window and retry. Do not keep sending browser actions through the broken pipe.
+- Never click `Publish`, `Post`, or any externally visible submit action without explicit final confirmation from the user in the current conversation.
+
+**X Articles**
+1. Convert Markdown and keep the image map:
+   ```bash
+   ${BUN_X} {baseDir}/scripts/md-to-html.ts article.md --save-html /tmp/x-article-body.html > /tmp/x-article.json
+   ```
+2. Read the JSON output for `title`, `coverImage`, and `contentImages` (`placeholder` → `localPath`).
+3. Open or create the article draft at `https://x.com/compose/articles`.
+4. Upload the cover with the Chrome plugin file chooser flow. If upload is blocked by extension permissions, stop and report the exact permission fix above.
+5. Fill the title, then copy rich HTML:
+   ```bash
+   ${BUN_X} {baseDir}/scripts/copy-to-clipboard.ts html --file /tmp/x-article-body.html
+   ```
+6. Paste into the article body with a real paste keystroke through the Chrome plugin. On macOS use `Meta+V`.
+7. Verify the editor text contains the article body and `XIMGPH_` placeholders. Do not rely on `tab.clipboard.readText()` as proof of the system clipboard after shell clipboard writes; on macOS verify with `pbpaste` if needed.
+8. For each `contentImages` item in placeholder order:
+   - Locate the visible placeholder text (`XIMGPH_N`) and click it to place the caret there.
+   - Open the toolbar menu `Insert` -> `Media`.
+   - In the modal, click the icon button with `aria-label="Add photos or video"`; do not click the text/dropzone or hidden file input.
+   - Use the file chooser to upload that image's `localPath`.
+   - After the image appears, if `XIMGPH_N` remains above it, select exactly that placeholder and press `Delete` first. Use `Backspace` only if `Delete` fails and the selected text is confirmed to be exactly the placeholder.
+   - Verify the placeholder count for that `XIMGPH_N` is `0`.
+9. Open Preview and verify title, cover, body, links, and images.
+10. Ask for explicit confirmation before clicking `Publish`.
+
 ## Preferences (EXTEND.md)

 Check EXTEND.md in priority order — the first one found wins:
@@ -86,6 +143,73 @@ Checks: Chrome, profile isolation, Bun, Accessibility, clipboard, paste keystrok

 ---

+## Chrome Computer Use Mode
+
+Use this mode when the user explicitly asks for Chrome Computer Use, or when no Chrome-plugin preference is stated and Codex can control `Google Chrome` with Computer Use. This uses the user's existing Chrome window, cookies, login, extensions, and X session.
+
+**General rules**:
+- Start each assistant turn that controls Chrome by calling `get_app_state` for `Google Chrome`.
+- Prefer element-index actions when available; use coordinates only for editor text selection or drag selection.
+- Do not use the in-app Browser, the Chrome plugin, Playwright, or CDP for X UI actions in this mode unless the user approves a mode change.
+- Never click `Publish`, `Post`, or any externally visible submit action without an explicit final confirmation from the user in the current conversation.
+
+**Regular posts**:
+1. Open or navigate Chrome to `https://x.com/compose/post`.
+2. Type the post text into the composer using Computer Use.
+3. For each image, run:
+   ```bash
+   ${BUN_X} {baseDir}/scripts/copy-to-clipboard.ts image /absolute/path/to/image.png
+   ```
+4. Paste with Computer Use (`super+v` on macOS, `control+v` on Windows/Linux), then wait until X finishes uploading media.
+5. Ask for confirmation before clicking `Post`.
+
+**Video posts**:
+1. Open or navigate Chrome to `https://x.com/compose/post`.
+2. Type the post text into the composer.
+3. Use the visible media upload/file picker UI to attach the video.
+4. Wait for upload and processing to complete.
+5. Ask for confirmation before clicking `Post`.
+
+**Quote tweets**:
+1. Open the tweet URL in Chrome.
+2. Use the visible quote/repost UI to choose Quote.
+3. Type the comment.
+4. Ask for confirmation before clicking `Post`.
+
+**X Articles**:
+1. Convert Markdown and keep the image map:
+   ```bash
+   ${BUN_X} {baseDir}/scripts/md-to-html.ts article.md --save-html /tmp/x-article-body.html > /tmp/x-article.json
+   ```
+2. Read the JSON output for `title`, `coverImage`, and `contentImages` (`placeholder` → `localPath`).
+3. In Chrome, open `https://x.com/compose/articles`, create or open the draft, upload the cover if present, and fill the title.
+4. Copy rich HTML to the clipboard:
+   ```bash
+   ${BUN_X} {baseDir}/scripts/copy-to-clipboard.ts html --file /tmp/x-article-body.html
+   ```
+5. Paste into the article body with Computer Use.
+6. For each `contentImages` entry in placeholder order:
+   - Locate the exact visible placeholder text such as `XIMGPH_3` and click it to set the insertion point.
+   - Open the toolbar `Insert` dropdown, choose `Media`, then click the modal's icon button labeled `Add photos or video`.
+   - Use the native file picker to choose the image's `localPath`.
+   - Wait until the image block appears and any upload activity is finished.
+   - If the placeholder remains above the inserted image, reselect exactly that placeholder text and press `Delete` first. Use `Backspace` only if `Delete` fails and the selected text is confirmed to be exactly the placeholder.
+7. Verify no `XIMGPH_` placeholders remain and the expected images appear.
+8. Open Preview and verify title, cover, body, links, and images.
+9. Ask for explicit confirmation before clicking `Publish`.
+
+If Computer Use selection, toolbar upload, or file-picker control becomes unreliable, stop and report the blocker instead of switching to the Chrome plugin or CDP silently.
+
+---
+
+## CDP Script Mode (Fallback)
+
+Use the script sections below only when the selected browser-control mode is unavailable, unreliable, or explicitly not requested. These scripts launch or reuse a real Chrome instance via CDP and keep the browser open for review.
+
+Do not use CDP Script Mode when the user explicitly requires the Codex Chrome plugin or Chrome Computer Use unless the user approves the fallback after you explain the blocker.
+
+---
+
 ## Post Type Selection

 Unless the user explicitly specifies the post type:
@@ -107,6 +231,8 @@ ${BUN_X} {baseDir}/scripts/x-browser.ts "Hello!" --image ./photo.png

 **Note**: Script opens browser with content filled in. User reviews and publishes manually.

+**Codex mode note**: If the user explicitly requested the Codex Chrome plugin, use **Codex Chrome Plugin Mode**. Otherwise, if Chrome Computer Use is enabled, use **Chrome Computer Use Mode** instead of running `x-browser.ts`.
+
 ---

 ## Video Posts
@@ -126,6 +252,8 @@ ${BUN_X} {baseDir}/scripts/x-video.ts "Check this out!" --video ./clip.mp4

 **Note**: Script opens browser with content filled in. User reviews and publishes manually.

+**Codex mode note**: If the user explicitly requested the Codex Chrome plugin, use **Codex Chrome Plugin Mode**. Otherwise, if Chrome Computer Use is enabled, use **Chrome Computer Use Mode** instead of running `x-video.ts`.
+
 **Limits**: Regular 140s max, Premium 60min. Processing: 30-60s.

 ---
@@ -147,6 +275,8 @@ ${BUN_X} {baseDir}/scripts/x-quote.ts https://x.com/user/status/123 "Great insig

 **Note**: Script opens browser with content filled in. User reviews and publishes manually.

+**Codex mode note**: If the user explicitly requested the Codex Chrome plugin, use **Codex Chrome Plugin Mode**. Otherwise, if Chrome Computer Use is enabled, use **Chrome Computer Use Mode** instead of running `x-quote.ts`.
+
 ---

 ## X Articles
@@ -167,7 +297,11 @@ ${BUN_X} {baseDir}/scripts/x-article.ts article.md --cover ./cover.jpg

 **Frontmatter**: `title`, `cover_image` supported in YAML front matter.

-**Note**: Script opens browser with article filled in. User reviews and publishes manually.
+**Codex mode note**: If the user explicitly requested the Codex Chrome plugin, follow **Codex Chrome Plugin Mode** above. If the user explicitly requested Chrome Computer Use, follow **Chrome Computer Use Mode**. Otherwise, prefer Chrome Computer Use; for Markdown articles with local content images, use the toolbar `Insert` -> `Media` image-upload workflow before falling back to `x-article.ts` in **CDP Script Mode**.
+
+**CDP fallback note**: The script opens browser with article filled in. User reviews and publishes manually unless `--submit` is used.
+
+**Publish safety**: Do not use `--submit` or click `Publish` unless the user explicitly confirms the final public publish action.

 **Post-Composition Check**: The script automatically verifies after all images are inserted:
 - Remaining `XIMGPH_` placeholders in editor content
@@ -181,7 +315,7 @@ If the check fails (warnings in output), alert the user with the specific issues

 ### Chrome debug port not ready

-If a script fails with `Chrome debug port not ready` or `Unable to connect`, kill existing Chrome CDP instances first, then retry:
+CDP fallback only: if a script fails with `Chrome debug port not ready` or `Unable to connect`, kill existing Chrome CDP instances first, then retry:

 ```bash
 pkill -f "Chrome.*remote-debugging-port" 2>/dev/null; pkill -f "Chromium.*remote-debugging-port" 2>/dev/null; sleep 2
@@ -192,7 +326,8 @@ pkill -f "Chrome.*remote-debugging-port" 2>/dev/null; pkill -f "Chromium.*remote
 ## Notes

 - First run: manual login required (session persists)
- All scripts only fill content into the browser, user must review and publish manually
+- In Codex Chrome Plugin Mode and Chrome Computer Use Mode, use the user's existing Chrome session and do not launch a separate CDP profile
+- CDP scripts only fill content into the browser by default; user must review and publish manually unless `--submit` is explicitly used
 - Cross-platform: macOS, Linux, Windows

 ## Extension Support
@@ -2,6 +2,17 @@

 Publish Markdown articles to X Articles editor with rich text formatting and images.

+## Mode Selection
+
+In Codex, choose the browser-control mode from the user's wording:
+
+1. If the user says "Codex Chrome plugin", "Codex 自带的 Chrome 插件", `@chrome`, or Chrome Extension, use **Codex Chrome Plugin Workflow**. Do not try Computer Use first.
+2. If the user explicitly asks for Chrome Computer Use, use **Computer Use Workflow**.
+3. If the user explicitly asks for CDP/script mode, use **CDP Script Workflow**.
+4. Otherwise, use Computer Use when available; if unavailable or blocked, use CDP Script Workflow.
+
+Never use the in-app Browser for X Article publishing. Never switch away from an explicitly requested mode without explaining the blocker and getting approval.
+
 ## Prerequisites

 - X Premium subscription (required for Articles)
@@ -10,6 +21,42 @@ Publish Markdown articles to X Articles editor with rich text formatting and ima

 ## Usage

+### Codex Chrome Plugin (When Requested)
+
+Use the `chrome:Chrome` skill and its Node REPL browser client. Verify the connection with a lightweight call such as `browser.user.openTabs()`. If it fails, wait 2 seconds and retry once, then follow the Chrome skill's health checks.
+
+Prepare the article HTML and image map:
+
+```bash
+${BUN_X} {baseDir}/scripts/md-to-html.ts article.md --save-html /tmp/x-article-body.html > /tmp/x-article.json
+```
+
+Copy generated HTML as rich text:
+
+```bash
+${BUN_X} {baseDir}/scripts/copy-to-clipboard.ts html --file /tmp/x-article-body.html
+```
+
+Use the Chrome plugin's tab, Playwright-wrapper, CUA, clipboard, and file chooser APIs for all X UI operations. If upload fails with `Not allowed`, stop and tell the user to enable file URL access for the Codex Chrome Extension in `chrome://extensions` → Details.
+
+### Chrome Computer Use
+
+Prepare the article HTML and image map:
+
+```bash
+${BUN_X} {baseDir}/scripts/md-to-html.ts article.md --save-html /tmp/x-article-body.html > /tmp/x-article.json
+```
+
+Copy the generated HTML as rich text:
+
+```bash
+${BUN_X} {baseDir}/scripts/copy-to-clipboard.ts html --file /tmp/x-article-body.html
+```
+
+Then use Codex Computer Use against `Google Chrome` for all X UI operations.
+
+### CDP Script Fallback
+
 ```bash
 # Publish markdown article (preview mode)
 ${BUN_X} {baseDir}/scripts/x-article.ts article.md
@@ -21,6 +68,8 @@ ${BUN_X} {baseDir}/scripts/x-article.ts article.md --cover ./cover.jpg
 ${BUN_X} {baseDir}/scripts/x-article.ts article.md --submit
 ```

+Do not use `--submit` unless the user has explicitly confirmed the final public publish action.
+
 ## Markdown Format

 ```markdown
@@ -118,7 +167,63 @@ JSON output:
 | `1. item` | `<ol><li>` |
 | `![](img)` | Image placeholder |

-## Workflow
+## Codex Chrome Plugin Workflow
+
+1. **Load Chrome skill**: use `chrome:Chrome`, not Computer Use.
+2. **Connect**: initialize the Chrome plugin browser client and verify with `browser.user.openTabs()`.
+3. **Parse Markdown**: run `md-to-html.ts --save-html /tmp/x-article-body.html > /tmp/x-article.json`.
+4. **Read the map**: use `/tmp/x-article.json` for `title`, `coverImage`, and `contentImages`.
+5. **Open X Articles**: open or claim a Chrome tab for `https://x.com/compose/articles`.
+6. **Create Draft**: click the create/write button if needed, or open the target draft.
+7. **Upload Cover**: use the Chrome plugin file chooser flow. If file upload returns `Not allowed`, report the Chrome Extension file-access fix and stop.
+8. **Fill Title**: fill the title field.
+9. **Paste Content**:
+   - Run `copy-to-clipboard.ts html --file /tmp/x-article-body.html`.
+   - Click the article body.
+   - Press `Meta+V` on macOS or `Control+V` on Windows/Linux through the Chrome plugin.
+   - Verify the article body appeared and contains `XIMGPH_` placeholders. On macOS, use `pbpaste` to verify shell-written system clipboard contents if paste is suspicious; `tab.clipboard.readText()` may not reflect the system clipboard after shell writes.
+10. **Insert Images**: for each `contentImages` item in placeholder order:
+   - Locate the exact visible placeholder text (`XIMGPH_N`) and click it to put the insertion point there.
+   - Open the editor toolbar dropdown `Insert` and choose `Media`.
+   - In the `Insert` modal, click the icon button with `aria-label="Add photos or video"`; do not click the "Choose a file or drag it here" text/dropzone or hidden file input.
+   - Use the Chrome plugin file chooser flow to upload that image's `localPath`.
+   - Wait until the image block appears. If `XIMGPH_N` remains above the image, select exactly that placeholder and press `Delete` first; use `Backspace` only if `Delete` fails and the selected text is confirmed to be exactly the placeholder.
+   - Verify that placeholder's count is `0` before continuing.
+11. **Verify**:
+   - Inspect the editor for `XIMGPH_` residue.
+   - Confirm the expected number of image blocks is visible.
+   - Open Preview and verify title, cover, body, links, and images.
+12. **Publish Safety**: ask the user for explicit final confirmation before clicking `Publish`.
+
+If the Chrome plugin reports `native pipe is closed`, retry one lightweight browser call after 2 seconds, then run the Chrome skill health checks. If Chrome, the extension, and native host are healthy, ask the user before opening a new Chrome window and retrying.
+
+## Computer Use Workflow
+
+1. **Detect Computer Use**: call `get_app_state` for `Google Chrome`; use `tool_search` first if the tools are not visible.
+2. **Parse Markdown**: run `md-to-html.ts --save-html /tmp/x-article-body.html > /tmp/x-article.json`.
+3. **Read the map**: use `/tmp/x-article.json` for `title`, `coverImage`, and `contentImages`.
+4. **Open X Articles**: use Chrome Computer Use to navigate to `https://x.com/compose/articles`.
+5. **Create Draft**: click the create/write button if needed, or open the target draft.
+6. **Upload Cover**: if `coverImage` exists, use Chrome's visible upload/file picker UI. If the file picker cannot be operated reliably, stop and ask for help rather than switching to CDP silently.
+7. **Fill Title**: type the title into the title field.
+8. **Paste Content**:
+   - Run `copy-to-clipboard.ts html --file /tmp/x-article-body.html`.
+   - Click the article body.
+   - Press `super+v` on macOS or `control+v` on Windows/Linux with Computer Use.
+9. **Insert Images**: for each `contentImages` item in placeholder order:
+   - Locate the exact visible placeholder text (`XIMGPH_N`) and click it to put the insertion point there.
+   - Open the editor toolbar dropdown `Insert`, choose `Media`, then click the icon button with `aria-label="Add photos or video"` inside the modal.
+   - Use the native file picker to choose that image's `localPath`.
+   - Wait until the image block appears and upload activity is complete.
+   - If `XIMGPH_N` remains above the inserted image, reselect exactly that placeholder text and press `Delete` first; use `Backspace` only if `Delete` fails and the Computer Use state confirms the selected text is exactly the placeholder.
+   - Confirm that placeholder is gone before continuing.
+10. **Verify**:
+   - Inspect the Computer Use state for `XIMGPH_` residue.
+   - Confirm the expected number of image blocks is visible.
+   - Open Preview and verify title, cover, body, links, and images.
+11. **Publish Safety**: ask the user for explicit final confirmation before clicking `Publish`.
+
+## CDP Script Workflow (Fallback)

 1. **Parse Markdown**: Extract title, cover, content images, generate HTML
 2. **Launch Chrome**: Real browser with CDP, persistent login
@@ -127,17 +232,18 @@ JSON output:
 5. **Upload Cover**: Use file input for cover image
 6. **Fill Title**: Type title into title field
 7. **Paste Content**: Copy HTML to clipboard, paste into editor
-8. **Insert Images**: For each placeholder (reverse order):
-   - Find placeholder text in editor
-   - Select the placeholder
-   - Copy image to clipboard
-   - Paste to replace selection
+8. **Insert Images**: For each placeholder in placeholder order:
+   - Find and click the placeholder text in the editor
+   - Use `Insert` -> `Media`
+   - Click the modal's icon button labeled `Add photos or video`
+   - Upload the matching image file
+   - Delete the leftover placeholder text with `Delete` after the image appears
 9. **Post-Composition Check** (automatic):
   - Scan editor for remaining `XIMGPH_` placeholders
   - Compare expected vs actual image count
   - Warn if issues found
 10. **Review**: Browser stays open for 60s preview
-11. **Publish**: Only with `--submit` flag
+11. **Publish**: Only with `--submit` flag and explicit user confirmation

 ## Example Session

@@ -145,23 +251,27 @@ JSON output:
 User: /post-to-x article ./blog/my-post.md --cover ./thumbnail.png

 Claude:
-1. Parses markdown: title="My Post", 3 content images
-2. Launches Chrome with CDP
-3. Navigates to x.com/compose/articles
-4. Clicks create button
+1. Detects that the user requested the Codex Chrome plugin
+2. Parses markdown: title="My Post", 3 content images
+3. Saves `/tmp/x-article-body.html` and `/tmp/x-article.json`
+4. Uses the Chrome plugin to open X Articles and create a draft
 5. Uploads thumbnail.png as cover
 6. Fills title "My Post"
-7. Pastes HTML content
+7. Pastes HTML content with a real Chrome paste
 8. Inserts 3 images at placeholder positions
-9. Reports: "Article composed. Review and use --submit to publish."
+9. Opens Preview and asks before publishing
 ```

 ## Troubleshooting

 - **No create button**: Ensure X Premium subscription is active
 - **Cover upload fails**: Check file path and format (PNG, JPEG)
- **Images not inserting**: Verify placeholders exist in pasted content
+- **Images not inserting**: Verify placeholders exist in pasted content; use `Insert` -> `Media` -> modal icon button `Add photos or video`, not image clipboard paste, the dropzone text, or the hidden file input.
 - **Content not pasting**: Check HTML clipboard: `${BUN_X} {baseDir}/scripts/copy-to-clipboard.ts html --file /tmp/test.html`
+- **Chrome plugin `native pipe is closed`**: retry once after 2 seconds, then run Chrome skill checks; ask before opening a new Chrome window if checks pass.
+- **Chrome plugin upload `Not allowed`**: enable file URL access for the Codex Chrome Extension in `chrome://extensions` → Details.
+- **Computer Use unavailable**: Use the CDP fallback script, unless the user explicitly required Chrome Computer Use.
+- **Placeholder remains after upload**: Select only the placeholder text and press `Delete` after upload completes. Use `Backspace` only if `Delete` fails and the selection is exactly the placeholder.

 ## How It Works

@@ -172,7 +282,21 @@ Claude:
   - Downloads remote images locally
   - Returns structured JSON

-2. `x-article.ts` publishes via CDP:
+2. The Codex Chrome plugin publishes through the user's real Chrome session when explicitly requested:
+   - Uses the user's active Chrome profile and logged-in X session
+   - Uses the Chrome Extension browser client rather than Computer Use or CDP
+   - Uses `copy-to-clipboard.ts` for rich HTML body paste
+   - Inserts body images through X's toolbar `Insert` -> `Media` modal and its `Add photos or video` icon button
+   - Keeps the final publish click under user confirmation
+
+3. Chrome Computer Use publishes through the user's visible Chrome UI:
+   - Uses the user's active Chrome profile and logged-in X session
+   - Uses `copy-to-clipboard.ts` for rich HTML body paste
+   - Inserts body images through X's toolbar `Insert` -> `Media` modal and its `Add photos or video` icon button
+   - Uses real keystrokes (`super+v`/`control+v`) through Codex Computer Use
+   - Keeps the final publish click under user confirmation
+
+4. `x-article.ts` publishes via CDP as a fallback:
   - Launches real Chrome (bypasses detection)
   - Uses persistent profile (saved login)
   - Navigates and fills editor via DOM manipulation
@@ -6,6 +6,17 @@ Detailed documentation for posting text and images to X.

 If you prefer step-by-step control:

+### Step 0: Prefer Chrome Computer Use in Codex
+
+When running inside Codex, first detect whether Chrome Computer Use is enabled:
+
+1. If Computer Use tools are already visible, call `get_app_state` for `Google Chrome`.
+2. If not, use `tool_search` for `computer-use get_app_state click press_key drag scroll Google Chrome`, then call `get_app_state`.
+3. If `get_app_state` succeeds, use the user's real Chrome with Computer Use for all X UI actions.
+4. Use CDP scripts only when Computer Use is unavailable or explicitly requested.
+
+If the user explicitly asks for Chrome Computer Use, do not use Playwright, the in-app Browser, or CDP without approval.
+
 ### Step 1: Copy Image to Clipboard

 ```bash
@@ -25,27 +36,15 @@ ${BUN_X} {baseDir}/scripts/paste-from-clipboard.ts --app "Google Chrome" --retri
 ${BUN_X} {baseDir}/scripts/paste-from-clipboard.ts --delay 200
 ```

-### Step 3: Use Playwright MCP (if Chrome session available)
+### Step 3: Use Chrome Computer Use (Preferred)

-```bash
-# Navigate
-mcp__playwright__browser_navigate url="https://x.com/compose/post"
-
-# Get element refs
-mcp__playwright__browser_snapshot
-
-# Type text
-mcp__playwright__browser_click element="editor" ref="<ref>"
-mcp__playwright__browser_type element="editor" ref="<ref>" text="Your content"
-
-# Paste image (after copying to clipboard)
-mcp__playwright__browser_press_key key="Meta+v"  # macOS
-# or
-mcp__playwright__browser_press_key key="Control+v"  # Windows/Linux
-
-# Screenshot to verify
-mcp__playwright__browser_take_screenshot filename="preview.png"
-```
+1. Use `get_app_state` for `Google Chrome`.
+2. Navigate Chrome to `https://x.com/compose/post` if needed.
+3. Click the composer and type the post text.
+4. Copy each image to the clipboard with `copy-to-clipboard.ts image <path>`.
+5. Press `super+v` on macOS or `control+v` on Windows/Linux with Computer Use.
+6. Wait until X finishes media upload.
+7. Ask for explicit confirmation before clicking `Post`.

 ## Image Support

@@ -59,12 +58,12 @@ mcp__playwright__browser_take_screenshot filename="preview.png"
 User: /post-to-x "Hello from Claude!" --image ./screenshot.png

 Claude:
-1. Runs: ${BUN_X} {baseDir}/scripts/x-browser.ts "Hello from Claude!" --image ./screenshot.png
-2. Chrome opens with X compose page
-3. Text is typed into editor
-4. Image is copied to clipboard and pasted
-5. Browser stays open 30s for preview
-6. Reports: "Post composed. Use --submit to post."
+1. Detects Chrome Computer Use
+2. Opens X compose in the user's real Chrome
+3. Types text into editor
+4. Copies image to clipboard and pastes with Computer Use
+5. Waits for upload and verifies the preview
+6. Asks before clicking Post
 ```

 ## Troubleshooting
@@ -80,7 +79,13 @@ Claude:

 ## How It Works

-The `x-browser.ts` script uses Chrome DevTools Protocol (CDP) to:
+In Chrome Computer Use mode:
+1. Codex controls the user's visible Google Chrome window
+2. Text is typed through the real UI
+3. Images are copied to the system clipboard and pasted with real keystrokes
+4. The user confirms before the final public post
+
+The `x-browser.ts` script is the CDP fallback. It uses Chrome DevTools Protocol (CDP) to:
 1. Launch real Chrome (not Playwright) with `--disable-blink-features=AutomationControlled`
 2. Use persistent profile directory for saved login sessions
 3. Interact with X via CDP commands (Runtime.evaluate, Input.dispatchKeyEvent)
@@ -164,7 +164,7 @@ export async function publishArticle(options: ArticleOptions): Promise<void> {

    // Check if we're on the articles list page (has Write button)
    console.log('[x-article] Looking for Write button...');
-    const writeButtonFound = await waitForElement('[data-testid="empty_state_button_text"]', 10_000);
+    const writeButtonFound = await waitForElement('[data-testid="empty_state_button_text"]', 30_000);

    if (writeButtonFound) {
      console.log('[x-article] Clicking Write button...');
@@ -172,6 +172,40 @@ export async function publishArticle(options: ArticleOptions): Promise<void> {
        expression: `document.querySelector('[data-testid="empty_state_button_text"]')?.click()`,
      }, { sessionId });
      await sleep(2000);
+    } else {
+      console.log('[x-article] Write button not found, looking for create button...');
+      const createClicked = await cdp.send<{ result: { value: boolean } }>('Runtime.evaluate', {
+        expression: `(() => {
+          const selectors = [
+            'button[aria-label="create"]',
+            'button[aria-label="Create"]',
+            'button[aria-label*="create" i]',
+            'button[aria-label*="write" i]',
+            'a[href="/compose/articles"]'
+          ];
+          for (const sel of selectors) {
+            const el = document.querySelector(sel);
+            if (el instanceof HTMLElement) {
+              el.click();
+              return true;
+            }
+          }
+          const buttons = [...document.querySelectorAll('button')];
+          const byText = buttons.find((el) => /^(create|write)$/i.test((el.textContent || '').trim()));
+          if (byText instanceof HTMLElement) {
+            byText.click();
+            return true;
+          }
+          return false;
+        })()`,
+        returnByValue: true,
+      }, { sessionId });
+      if (createClicked.result.value) {
+        console.log('[x-article] Create button clicked');
+        await sleep(2000);
+      } else {
+        console.log('[x-article] Create button not found');
+      }
    }

    // Wait for editor (title textarea)
@@ -218,6 +218,7 @@ async function launchChromeOnce(

  try {
    await waitForChromeDebugPort(port, 30_000, { includeLastError: true });
+    chrome.unref();
    return { chrome, port };
  } catch (error) {
    killChrome(chrome);
@@ -32,11 +32,14 @@ When this skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends. Users change the pinned backend via the `## Changing Preferences` section below.

 **Prompt file requirement (hard)**: write each image's full, final prompt to a standalone file under `prompts/` (naming: `NN-slide-[slug].md`) BEFORE invoking any backend. The file is the reproducibility record and lets you switch backends without regenerating prompts.
@@ -0,0 +1,33 @@
+# baoyu-wechat-summary preferences
+#
+# Copy this file to one of:
+#   .baoyu-skills/baoyu-wechat-summary/EXTEND.md           (project-local, takes precedence)
+#   $XDG_CONFIG_HOME/baoyu-skills/baoyu-wechat-summary/EXTEND.md   (XDG, falls back to ~/.config)
+#   ~/.baoyu-skills/baoyu-wechat-summary/EXTEND.md         (per-user home)
+#
+# First match wins. Keys are case-insensitive. Use `key: value` or `key=value`.
+# Blank lines and lines starting with `#` are ignored.
+
+# REQUIRED — your own wxid. Used to recognize your own messages so they show up
+# as `self_display` (below) in digests instead of the raw wxid string.
+# Find via:  wx contacts --query <你的昵称> --json    (look for chat_type=private, your own row)
+self_wxid: wxid_xxxxxxxxxxxx
+
+# REQUIRED — the display name to substitute for your wxid in digest text.
+self_display: 宝玉
+
+# OPTIONAL — which version(s) to generate when the user doesn't explicitly say.
+# Values: normal | roast | both
+# Default: normal
+default_version: normal
+
+# OPTIONAL — default time range when the user gives none and there's no prior
+# digest to incrementally extend from. Forms: `Nd` (days), `Nh` (hours), or a
+# fixed `YYYY-MM-DD` (treated as both --since and --until).
+# Default: 24h (i.e., today)
+# default_time_range: 7d
+
+# OPTIONAL — override the root directory where digest folders live. Default
+# resolves to `{project_root}/wechat`. Useful if you want a shared archive
+# outside the current project.
+# data_root: ~/Documents/wechat-digests
@@ -0,0 +1,456 @@
+---
+name: baoyu-wechat-summary
+description: Summarizes WeChat group chat highlights into a structured digest using the local wx-cli binary (https://github.com/jackwener/wx-cli). Generates a normal digest by default; a roast (毒舌) version is opt-in. Maintains per-group history (history.json + history-digests.jsonl) and per-user profiles across runs, with privacy guardrails baked in. Use when the user asks to "总结群聊", "群聊精华", "群聊摘要", "summarize group chat", "group chat digest", mentions a WeChat group name with a time range, says "帮我看看 XX 群最近聊了什么", "XX 群有什么值得看的", or asks to "回溯画像" / "初始化画像" / "backfill profiles". Adds the roast version when the user says "毒舌版", "roast 版", "再来个毒舌的", or similar.
+version: 0.1.0
+metadata:
+  openclaw:
+    homepage: https://github.com/JimLiu/baoyu-skills#baoyu-wechat-summary
+    requires:
+      anyBins:
+        - wx
+---
+
+# WeChat Group Summary
+
+群聊精华提取专家。把零散的微信群聊记录提炼成结构化、可读性强的简报，并维护跨次运行的群聊历史与群友画像。底层依赖外部 [wx-cli](https://github.com/jackwener/wx-cli) 二进制（`wx` 命令），不打包脚本。
+
+> **⚠️ Sandbox restriction**
+>
+> wx-cli reads from `~/.wx-cli/` (config, cache, daemon socket) and from WeChat's data directory (`~/Library/Containers/com.tencent.xinWeChat/` on macOS). Both paths are outside Claude Code's default sandbox. Every `wx` command in this skill needs to run with `dangerouslyDisableSandbox: true` from the start — don't waste a sandbox attempt first. The user can use `/sandbox` to view/edit restrictions.
+
+## User Input Tools
+
+When this skill prompts the user, follow this tool-selection rule (priority order):
+
+1. **Prefer built-in user-input tools** exposed by the current agent runtime — e.g., `AskUserQuestion`, `request_user_input`, `clarify`, `ask_user`, or any equivalent.
+2. **Fallback**: if no such tool exists, emit a numbered plain-text message and ask the user to reply with the chosen number/answer for each question.
+3. **Batching**: if the tool supports multiple questions per call, combine all applicable questions into a single call; if only single-question, ask them one at a time in priority order.
+
+Concrete `AskUserQuestion` references below are examples — substitute the local equivalent in other runtimes.
+
+## Prerequisites
+
+Before invoking the workflow, verify the environment. Run these checks in order; stop at the first failure and surface the exact next command the user needs.
+
+1. **wx-cli installed** — run `wx --version`. If missing, tell the user to install it themselves (`npm install -g @jackwener/wx-cli` or use one of the alternatives at https://github.com/jackwener/wx-cli). **Do NOT auto-install** — this repo forbids piped/silent installs.
+2. **`~/.wx-cli` directory owned by the current user** — `sudo wx init` historically chowned this directory to root, which breaks every subsequent non-sudo `wx` call. Check:
+   ```bash
+   ls -la ~/.wx-cli/ 2>/dev/null | head -5
+   ```
+   If the directory exists but the owner is `root` (or anything other than `$(whoami)`), tell the user to repair it themselves:
+   ```bash
+   sudo chown -R $(whoami) ~/.wx-cli
+   sudo rm -f ~/.wx-cli/daemon.pid ~/.wx-cli/daemon.sock
+   wx daemon start
+   ```
+   The skill should NOT run `sudo` on the user's behalf.
+3. **wx-cli initialized** — `wx sessions` should return data. If it fails with "no keys" / "init required", instruct the user to run `wx init` while WeChat is running (on macOS, `codesign --force --deep --sign - /Applications/WeChat.app` first). Prefer non-sudo init; only fall back to `sudo wx init` if the user's wx-cli version requires it — and warn them that they'll need step 2's chown after.
+4. **WeChat 4.x running and logged in** — required for the daemon to find data files.
+
+## Preferences (EXTEND.md)
+
+Check EXTEND.md in priority order — the first one found wins:
+
+| Priority | Path | Scope |
+|----------|------|-------|
+| 1 | `.baoyu-skills/baoyu-wechat-summary/EXTEND.md` (relative to project root) | Project |
+| 2 | `${XDG_CONFIG_HOME:-$HOME/.config}/baoyu-skills/baoyu-wechat-summary/EXTEND.md` | XDG |
+| 3 | `$HOME/.baoyu-skills/baoyu-wechat-summary/EXTEND.md` | User home |
+
+| Result | Action |
+|--------|--------|
+| Found | Read, parse, apply. On first use in session, briefly remind: "Using preferences from [path]. Edit it to change defaults." |
+| Not found | **MUST** run first-time setup (BLOCKING) before generating any digest — do NOT silently use defaults. |
+
+### Supported keys
+
+EXTEND.md is plain text with `key: value` or `key=value` lines, `#` for comments, case-insensitive keys.
+
+| Key | Type | Default | Purpose |
+|-----|------|---------|---------|
+| `self_wxid` | string | (required) | The owning account's wxid. Messages whose `from_wxid` matches this are attributed to the user. |
+| `self_display` | string | (required) | Display name to substitute for the user's own messages in digest text. |
+| `default_version` | `normal` / `roast` / `both` | `normal` | Which version(s) to generate when the user doesn't say otherwise. |
+| `default_time_range` | string (e.g. `7d`, `24h`, `1d`) | (none) | Default range when the user omits time and there's no incremental anchor. |
+| `data_root` | path | `{project_root}/wechat` | Override where digest folders live. |
+
+A starter template lives at [EXTEND.md.example](EXTEND.md.example).
+
+### First-Time Setup (BLOCKING)
+
+If no EXTEND.md is found, do NOT silently proceed.
+
+**Step A — Try to auto-discover `self_wxid` and `self_display` first.** Run (in order, stop at the first that succeeds):
+
+```bash
+# 1. If wx-cli exposes a whoami, use it
+wx whoami --json 2>/dev/null
+
+# 2. Otherwise, find self-sent messages in recent sessions
+wx sessions --json --limit 20 2>/dev/null
+```
+
+For option 2, scan the sessions for any private/group thread the user has sent into and read one of their own `from_wxid` / `from_nickname` pairs. If you can confidently pre-fill both values, use them as defaults in the question below; otherwise leave the fields blank for the user to fill in.
+
+**Step B — Confirm with one `AskUserQuestion` call (batched), pre-filling whatever auto-discovery found:**
+
+- `self_wxid` (e.g., `wxid_abc123`) — fall-back hint: the user can find it with `wx contacts --query "<own nickname>"`, or by inspecting any of their own sent messages in `wx sessions --json`
+- `self_display` (e.g., `宝玉`) — how they want their messages attributed
+- `default_version` — pick one of `normal` / `roast` / `both`
+- `data_root` — where digest folders live. Default: `{project_root}/wechat`. Enter a custom absolute path (e.g. `~/Documents/wechat-digests`) or leave blank for default.
+- Save location — pick one of project / XDG / home
+
+Write EXTEND.md to the chosen path. If the user provided a non-default `data_root`, include it as an uncommented line; otherwise omit it (the default applies automatically). Confirm "Preferences saved to [path]. Edit it any time to change defaults.", then continue with the digest workflow.
+
+## Workflow
+
+### Step 1: Parse the user's request
+
+Extract:
+
+- **Group name** (or partial name for fuzzy matching)
+- **Time range** — interpret flexibly:
+  - "最近 1 天" / "今天" / "last 24 hours" → 1 day
+  - "最近 3 天" → 3 days
+  - "最近 7 天" / "这周" → 7 days
+  - "最近 30 天" / "最近一个月" → 30 days
+  - "某天" (e.g. "3 月 5 号") → that specific date
+  - "某天到某天" (e.g. "3 月 1 号到 3 月 5 号") → date range
+  - "从上次开始" / "继续" / "接着上次" / "since last" → **incremental mode**: read `history.json` for this group, use `last_digest.last_message_time` as the start
+  - No time specified → **incremental mode**. If no `history.json` exists yet, fall back to `default_time_range` from EXTEND.md if set, else last 24 hours.
+- **Version(s) to generate**:
+  - Start from `default_version` in EXTEND.md.
+  - User request overrides: keywords "毒舌"/"roast"/"挑衅"/"再来个毒的"/"sass" → force `include_roast=true`. Keywords "只要正经的"/"normal only"/"不要毒舌" → force `include_normal=true, include_roast=false`. "都来一份"/"两个版本都要"/"both" → both.
+  - At least one of `include_normal`/`include_roast` must end up true.
+
+Convert relative ranges into absolute `--since YYYY-MM-DD --until YYYY-MM-DD` pairs using today's local date.
+
+### Step 2: Find the group + resolve folder path
+
+```bash
+wx contacts --query "<group_name>" --json
+```
+
+Filter for entries whose `username` ends in `@chatroom`. If multiple groups match, use `AskUserQuestion` to disambiguate. If none match, fall back to `wx sessions --json` and search there before asking the user.
+
+Once resolved, compute the folder path:
+
+```
+{data_root}/{group_id}-{sanitized_group_name}/
+```
+
+where `data_root` is from EXTEND.md (default `{project_root}/wechat`).
+
+**Sanitize the group name** — replace any of `/ \ : * ? " < > | NUL` and control characters with `_`. Trim trailing dots and whitespace. Don't strip emoji or Chinese characters.
+
+**Group-rename detection**: list existing folders under `{data_root}/` and find any folder whose name starts with `{group_id}-`. If one exists but the suffix differs (group was renamed), rename the existing folder to the new `{group_id}-{sanitized_new_name}` form. If a target with the new name already exists (rare), keep both and prefer the existing one for this run.
+
+### Step 3: Fetch messages
+
+For small batches (single-day digest, typically < 200 messages), pipe JSON into the agent directly:
+
+```bash
+wx history "<group_name_or_id>" --since YYYY-MM-DD --until YYYY-MM-DD -n 5000 --json
+```
+
+For **large batches** (weekly / monthly digests, > 200 messages), redirect to `$TMPDIR` first so the raw payload never sits in conversation context:
+
+```bash
+wx history "<group_name_or_id>" --since YYYY-MM-DD --until YYYY-MM-DD -n 5000 --json > "$TMPDIR/wx-messages.json"
+wc -c "$TMPDIR/wx-messages.json"
+jq 'length' "$TMPDIR/wx-messages.json"
+```
+
+Then read the file in slices via `Read` with `offset` + `limit`, or process with `jq` queries (e.g. `jq '.[0:200]'`, `jq '[.[] | {id, from_nickname, timestamp, content: (.content | .[0:50])}]'` for a lightweight skeleton pass). Reading all 500+ messages at once will burn token budget unnecessarily.
+
+Notes:
+
+- `--since` is inclusive; `--until` is interpreted as a date (the whole day). If the user asked for "today only", set both to today.
+- `-n 5000` is a defensive cap; for very active groups, raise it and re-fetch.
+- Filter the returned messages by their `timestamp` to be safe (some daemons may return adjacent days).
+- **Range splitting**: for ranges > 7 days OR > 500 messages, prefer generating per-3-day digests and then a meta-summary over forcing one giant digest — the categorization quality degrades sharply past a week's worth of unrelated topics.
+
+**Incremental mode**: after the fetch, drop any message whose `timestamp` is `<=` the `last_message_time` from `history.json`. If zero messages remain, tell the user "上次摘要后没有新消息，已跳过生成" and exit.
+
+### Step 3.5: Parse the message schema
+
+`wx history --json` returns an array of message objects. Use the fields that are present; tolerate missing fields:
+
+- **`id` / `msg_id` / `local_id`** — message identifier (use whichever wx-cli emits). Reference IDs in working notes as anchors when building the skeleton.
+- **`from_wxid`** — stable sender identifier
+- **`from_nickname`** — display name (may be the group remark or original nickname)
+- **`content`** — text payload. Examples:
+  - Plain text → use as-is
+  - `[图片]` → opaque placeholder; see image handling below
+  - `[表情]` → emoji/sticker; skip in body unless surrounded by discussion
+  - `[视频]` / `[文件]` → media reference; skip unless discussed
+  - `[链接] <title>` or `[链接/文件] <title>` → shared article; the title IS the information — quote it and credit the sharer
+  - `[系统] ... revokemsg` → revoked; exclude from digest and from leaderboard
+- **`timestamp`** — convert to `MM-DD HH:MM` for display (and use full ISO for `generated_at`)
+- **`chat_type`** — sanity-check `group`
+- **Quote/reply** — try `quote_id`, `reply_to`, `quoted_msg_id`, or any nested `quote` object. If present, use it as strong attribution. If absent, fall back to context but flag the inferred link as uncertain.
+
+### Step 3.6: Resolve self + ambiguous nicknames
+
+- Substitute `self_display` for every message whose `from_wxid` matches `self_wxid` (from EXTEND.md). Apply this in the leaderboard, portraits, and body text. The user MUST appear under their real display name and count toward stats — never skip them.
+- Scan all unique senders for ambiguous handles: ≤2 characters, common programming words (`nil`, `null`, `test`, `admin`, `user`, `undefined`), single emoji, or otherwise low-information. For each, run `wx contacts --query "<nick>" --json --limit 5` and pick a meaningful name in this priority: remark > nickname > wxid. Apply the substitution everywhere in the digest.
+
+### Step 3.7: Load user profiles
+
+For each unique sender appearing in this batch:
+
+- Look in `{folder}/profiles/{wxid}-*.md` by `wxid` prefix match. Read the matched file if found.
+- If `include_roast`, **also** look in `{folder}/profiles-roast/{wxid}-*.md` for the roast pass.
+
+Compile a condensed **profile context block** as internal working memory — do NOT write it into the final digest. Example shape:
+
+```
+== 群友历史画像（来自 profiles/）==
+K. H：空中直播员 / 生活百科全书。常见话题：旅行、金融、美食。经典金句："要不要买moderna"。
+可可苏玛：...
+```
+
+Rules:
+
+- Only load profiles for users active in this batch — never preload everyone.
+- Profile is **background**, not template. Current messages are still the primary source.
+- Use historical labels for **continuity** ("又双叒叕化身空中直播员") or **contrast** ("一向省钱的 XX 今天居然...").
+- **Strict separation**: normal pass reads only `profiles/`, roast pass reads only `profiles-roast/`. Never cross-load.
+
+See [references/profiles.md](references/profiles.md) for the full file format.
+
+### Step 3.8: Detect existing in-chat digests (optional)
+
+Some users (e.g., the original 宝玉 workflow) post digests directly into the group as messages. If we don't notice these, the new digest will re-cover the same ground.
+
+Scan the fetched messages for signals of a prior in-chat digest:
+
+- `from_wxid == self_wxid` AND
+- `content` contains `群聊精华` OR `消息统计:` OR `📊 消息统计` OR a leaderboard pattern (e.g. `^\d+\. .+: \d+ 条`), AND
+- `content` length > 1500 chars.
+
+If a match is found:
+
+1. Extract the digest's covered date or range from the title line (e.g., `xxx 群聊精华 · 2026-05-12` or `... · 2026-05-10 ~ 2026-05-12`).
+2. Surface the finding to the user via `AskUserQuestion`:
+   - "Detected an in-chat digest by you covering {范围}. Use {范围 end + 1} as the start instead of `history.json`?"
+   - Options: `Yes, skip up to {end of detected range}` / `No, use history.json` / `No, cover everything in the requested range`.
+3. Apply the chosen anchor.
+
+This is a heuristic — when uncertain (multiple matches, malformed title), default to `history.json` and tell the user what was skipped.
+
+Generate the digest in three rounds so nothing slips through. The methodology stays here in SKILL.md; the content/style rules live in [references/output-formats.md](references/output-formats.md) — read that file in Round 2 before drafting.
+
+#### Round 1 — Build the skeleton
+
+Read every message in order. **Skip image fetching/decoding** in this round. List every distinct discussion topic. Bias toward over-listing — trim in Round 3.
+
+Internal working format (not written to the final file):
+
+```
+== 话题清单（共 N 条消息）==
+1. [HH:MM-HH:MM] 话题名称（参与者：A, B, C）— 一句话概括（锚点 id：54052, 54055, 54063）
+2. [HH:MM-HH:MM] 话题名称（参与者：D, E）— 一句话概括（锚点 id：54100-54112）
+...
+
+== 可能需要图片上下文的话题 ==
+- 话题 3：锚点 id=49661（图片是讨论主体）
+
+== 发言统计 ==
+1. XXX — N 条  2. YYY — N 条  ...
+```
+
+Topic principles:
+
+- Topic-switch signals: time gap > 30 min, participant change, content jump.
+- 2+ participants OR substantive content qualifies as a topic; pure emoji-banter does not.
+- **Strict attribution**: each topic must record "who said what". Don't fuse adjacent messages from different senders just because they're close in time — when minutes apart or interleaved with others, split into separate topics. Prefer two topics over one wrongly-merged topic.
+- **Carry anchor IDs**: list the key message IDs for each topic. In Round 2, jump back to these IDs in the raw messages and verify content, don't guess from context. If `quote_id` / `reply_to` is present, use the ID chain — that's the most reliable attribution.
+
+**Flag-for-images criteria** (any one triggers): an explicit comment on an image (`看发型是X？`, `这是谁？`, `笑死`), multiple people piling onto the same image without saying what it is, an image as the core information (晒单/截图/资料), an explanatory line right after an image (`gpt-image-2`, `太可怕了`), or cross-sender ambiguity (B says "这个看着像 X" but the previous image is from A).
+
+#### Round 2 — Flesh out + write the digest
+
+For each topic in the skeleton, jump back to its anchor IDs and expand into full content with quotes and clear attribution. Then write the digest file.
+
+**Image handling** (limited — wx-cli does not decode chat images):
+
+For each flagged topic, check whether a description file already exists at `{folder}/imgs/{message_id}.txt`. If yes, read it (one-line plain text) and weave its content into the topic. If no, treat the image as opaque (`[图片]`) and write around it — describe what the surrounding messages tell us, but don't invent visual content.
+
+The `imgs/` directory exists as an **extension point**: a user (or a future wx-cli capability) can drop `{message_id}.txt` files with one-line descriptions, and the skill will pick them up. The skill itself does NOT generate these files in this version.
+
+**Use the profile context block** (from Step 3.7):
+
+- Echo continuity for matching behavior ("又双叒叕直播飞行体验")
+- Highlight contrast for departures ("一向话少的 XX 今天突然爆发")
+- Callback past quotes ("继上次'要不要买 moderna'之后，这次又...")
+- Don't sacrifice current material to force a callback.
+
+**Writing order**: write the body categories first, then the opening overview based on the finished body (so the hook is accurate).
+
+Detailed structure, voice, formatting rules, and content guidelines are in [references/output-formats.md](references/output-formats.md). Load that file now if not already loaded.
+
+#### Round 3 — Audit
+
+Walk the Round 1 skeleton against the finished digest. Check:
+
+- Any listed topic missing from the digest?
+- Quotes, names, product/tool names preserved verbatim?
+- Categorization makes sense — is anything in the wrong bucket?
+
+Fix in place. When clean, confirm and proceed.
+
+### Step 7: Save the digest file(s)
+
+If `include_normal`:
+
+- Single date → `{folder}/YYYY-MM-DD.md`
+- Date range → `{folder}/YYYY-MM-DD_YYYY-MM-DD.md`
+- Overwrite if the same date/range already exists.
+
+If `include_roast`:
+
+- Same naming, but with `-roast` suffix: `YYYY-MM-DD-roast.md` or `YYYY-MM-DD_YYYY-MM-DD-roast.md`.
+
+Both versions share the same statistics (message count, leaderboard) and the same underlying skeleton.
+
+### Step 8: Save history (two files)
+
+Maintain two files in the group folder:
+
+#### `history.json` — single record, fast read
+
+Always reflects only the most recent normal digest. Overwrite on each run when `include_normal=true`.
+
+```json
+{
+  "group_id": "12345678901@chatroom",
+  "group_name": "相亲相爱一家人",
+  "folder": "12345678901@chatroom-相亲相爱一家人",
+  "last_digest": {
+    "file": "2026-03-12.md",
+    "date_range": "2026-03-12",
+    "generated_at": "2026-03-12T10:30:00+08:00",
+    "message_count": 150,
+    "last_message_time": "03-12 18:45"
+  }
+}
+```
+
+- `group_name` updates on every run (handles renames).
+- `folder` records the current folder basename for cross-reference.
+- `last_message_time` is the timestamp of the most recent message included, in `MM-DD HH:MM` — used by incremental mode.
+- Roast-only runs do NOT touch this file.
+
+#### `history-digests.jsonl` — append-only archive
+
+One JSON object per line, same shape as `last_digest`. Every normal-version run appends one line (in chronological order). Used by backfill and historical lookups. Never read for incremental mode (which only needs the latest).
+
+```jsonl
+{"file":"2026-03-10.md","date_range":"2026-03-10","generated_at":"2026-03-10T09:00:00+08:00","message_count":420,"last_message_time":"03-10 22:30"}
+{"file":"2026-03-11.md","date_range":"2026-03-11","generated_at":"2026-03-11T09:05:00+08:00","message_count":312,"last_message_time":"03-11 23:10"}
+{"file":"2026-03-12.md","date_range":"2026-03-12","generated_at":"2026-03-12T10:30:00+08:00","message_count":150,"last_message_time":"03-12 18:45"}
+```
+
+If a normal digest with the same `file` name is regenerated, append a new line anyway (the JSONL is a strict log; readers can dedupe by `file` if they need to).
+
+### Step 8.5: Update user profiles
+
+For each user with 3+ messages in this batch who appeared in the 群友画像 section:
+
+- If `include_normal`, update `{folder}/profiles/{wxid}-{nickname}.md`.
+- If `include_roast`, update `{folder}/profiles-roast/{wxid}-{nickname}.md`.
+
+Counts, frontmatter updates, append-only rules for quotes and events, and privacy guardrails are detailed in [references/profiles.md](references/profiles.md). Load that file when running this step.
+
+### Completion checklist
+
+Profile updates are easy to forget once the digest is on disk. Before reporting the run as "done", verify every applicable file:
+
+- [ ] `{folder}/YYYY-MM-DD.md` written (if `include_normal`)
+- [ ] `{folder}/YYYY-MM-DD-roast.md` written (if `include_roast`)
+- [ ] `{folder}/history.json` overwritten with the new `last_digest` (if `include_normal`)
+- [ ] `{folder}/history-digests.jsonl` appended one line (if `include_normal`)
+- [ ] `{folder}/profiles/{wxid}-*.md` updated for every user with 3+ messages (if `include_normal`)
+- [ ] `{folder}/profiles-roast/{wxid}-*.md` updated for every user with 3+ messages (if `include_roast`)
+
+If any item is unchecked, finish it before declaring success. Don't ship a digest with a stale `history.json` — incremental mode depends on it.
+
+### Step 9: Backfill (user-triggered)
+
+When the user says "回溯画像" / "初始化画像" / "backfill profiles":
+
+1. Confirm the target group (if not specified, ask which one).
+2. List all digest files in `{folder}/` and `history-digests.jsonl`.
+3. Read existing digests in batches of 10–15 to avoid context blowup.
+4. For users appearing in 3+ digests, seed profile files using their leaderboard counts, portrait paragraphs, and quoted lines from the historical digests.
+5. Write to `profiles/` (and `profiles-roast/` if any `-roast.md` files exist).
+6. Report back: how many profiles were created, how many users covered.
+
+Full procedure in [references/profiles.md](references/profiles.md).
+
+## Storage layout
+
+```
+{data_root}/                                        # default: {project_root}/wechat/
+└── {group_id}-{group_name}/                        # e.g. 12345678901@chatroom-相亲相爱一家人/
+    ├── history.json                                # last digest pointer (fast)
+    ├── history-digests.jsonl                       # append-only archive
+    ├── 2026-03-12.md                               # normal digest, single date
+    ├── 2026-03-12-roast.md                         # roast digest (only if generated)
+    ├── 2026-03-10_2026-03-12.md                    # normal digest, date range
+    ├── profiles/                                   # normal user profiles
+    │   ├── onlytiancai-胡浩🐸.md
+    │   └── ...
+    ├── profiles-roast/                             # roast user profiles (only if any roast generated)
+    │   ├── onlytiancai-胡浩🐸.md
+    │   └── ...
+    └── imgs/                                       # optional image-description files
+        ├── 49661.txt                               # one-line plain text description
+        └── ...
+```
+
+## wx-cli quick reference
+
+| Command | Purpose |
+|---------|---------|
+| `wx --version` | Sanity-check that wx-cli is installed |
+| `wx sessions --json` | List recent sessions; useful for verifying init and finding the user's own wxid |
+| `wx contacts --query "<name>" --json` | Fuzzy-match contacts/groups by display name, remark, or wxid |
+| `wx history "<group>" --since DATE --until DATE -n N --json` | Pull a group's messages within a date range as JSON |
+| `wx members "<group>" --json` | List a group's members (rarely needed; mostly for completeness) |
+| `wx stats "<group>" --since DATE` | wx-cli's built-in stats; we compute our own from `wx history` JSON so the format matches our digest |
+| `wx daemon status` / `wx daemon stop` / `wx daemon logs --follow` | Daemon lifecycle (troubleshooting) |
+
+All `wx` commands accept `--json` for machine-readable output. Default output is YAML — only use it for human eyeballing during debugging.
+
+## Troubleshooting
+
+When a `wx` command fails, diagnose by the symptom, not by retrying blindly. Common patterns:
+
+| Symptom | Cause | Fix (tell the user to run these — do NOT run `sudo` for them) |
+|---------|-------|----------------------------------------------------------------|
+| `Operation not permitted` / `Access denied to ~/.wx-cli` | Sandbox is on | Re-run the command with `dangerouslyDisableSandbox: true`. Persistent fix: `/sandbox` to allow `~/.wx-cli` and the WeChat data dir. |
+| `无法写入 /Users/<u>/.wx-cli` / `Permission denied` | `~/.wx-cli` is owned by root (legacy `sudo wx init`) | `sudo chown -R $(whoami) ~/.wx-cli && sudo rm -f ~/.wx-cli/daemon.{pid,sock} && wx daemon start` |
+| `wx history` hangs / times out / returns nothing | Daemon is stuck | `wx daemon stop && rm -f ~/.wx-cli/daemon.{pid,sock} && wx daemon start`, then retry |
+| `no keys` / `init required` after the daemon was working | Keys went stale (WeChat restart, version upgrade) | Make sure WeChat is running, then `wx init --force` (non-sudo first; only `sudo` if your wx-cli version requires it) |
+| `wx contacts` returns zero rows for a group you know exists | Group is folded into 折叠群 or the daemon hasn't indexed it yet | `wx sessions --json` and search there; if missing, run `wx daemon stop && wx daemon start` and retry |
+| Messages returned but `--since` / `--until` window looks wrong | Date string not in `YYYY-MM-DD` format, or off-by-one timezone | Confirm the dates are local-time `YYYY-MM-DD`. Re-filter the JSON by `timestamp` locally as a belt-and-suspenders step. |
+| Empty result for a chat that should have activity | `-n` cap too low for a noisy group | Raise `-n` (e.g. to 20000) and re-fetch |
+
+**Recovery order when nothing makes sense:**
+
+1. Is WeChat running?
+2. Is `~/.wx-cli` owned by `$(whoami)`?
+3. Is the daemon healthy? (`wx daemon status`)
+4. Restart the daemon (`wx daemon stop && wx daemon start`)
+5. Last resort: `wx init --force` (while WeChat is running)
+
+Never auto-retry inside the skill — every failure should produce a clear diagnostic plus the exact command the user needs to run.
+
+## Notes and limitations
+
+- **Image content is opaque**. wx-cli does not decode chat images. The skill respects an `imgs/{message_id}.txt` extension point but does not auto-populate it. When a topic depends heavily on an image with no description file, the digest should say so honestly rather than invent visual content.
+- **Reply attribution is best-effort**. If wx-cli's output exposes a quote/reply field, use it. Otherwise fall back to context and flag uncertain inferences in working notes.
+- **Local time only**. Date parsing uses the agent's local time zone. Cross-time-zone group members may show timestamps that don't match their wall clock. Per the format rules, never use timestamps to infer sleep or location.
+- **wx-cli reinit**. If `wx history` suddenly returns nothing after a WeChat restart, the keys may be stale. Tell the user to run `sudo wx init --force` (while WeChat is running) and retry.
@@ -0,0 +1,349 @@
+# Output formats — normal & roast digest
+
+This reference defines the two digest variants the skill produces: the **normal** version (default, sober summary) and the **roast** version (毒舌，sarcastic critique, opt-in). Load this file during Step 4 (skeleton) and keep it open through Step 6 (audit).
+
+Both versions share the same overall layout and writing rules; the differences are tone, the leaderboard annotations, the portraits, and the footer. Write the normal version first when both are requested — it's the anchor for incremental mode and the source of truth for the profile updates.
+
+---
+
+## 1. Normal version
+
+### 1.1 Five-part structure
+
+```
+[Title line]
+[📊 Stats block + Top 10 leaderboard]
+[Opening summary — 1-2 paragraphs of prose]
+[群友画像 — one entry per active user (3+ msgs)]
+[Categorized body — 3-6 self-named sections per day]
+[Optional pain-point section]
+[Fixed footer]
+```
+
+### 1.2 Title line
+
+- Single line, no markdown heading.
+- Form: `{群名} 群聊精华 · {日期或日期区间}`
+- Date single day: `2026-03-12`. Date range: `2026-03-12 ~ 2026-03-15`.
+
+Example:
+
+```
+相亲相爱一家人 群聊精华 · 2026-03-12
+```
+
+### 1.3 Statistics block
+
+- Starts with `📊 消息统计: 共 N 条消息`.
+- Followed by a leaderboard, top 10 senders by message count, one per line.
+- Form per line: `{排名}. {昵称}: {消息数} 条`
+- Counting rules:
+  - Include images, emojis, links, voice transcripts — anything that occupies a chat row is one message.
+  - Exclude system messages and revoked messages (`[系统]`, `revokemsg`).
+  - For the `self_wxid` user, substitute `self_display` from EXTEND.md before counting/displaying.
+  - Resolve ambiguous nicknames (per SKILL.md Step 3.6) before tallying so the same person isn't double-counted.
+
+Example:
+
+```
+📊 消息统计: 共 387 条消息
+1. 蛙总: 92 条
+2. 老王: 58 条
+3. 阿喵: 41 条
+...
+```
+
+### 1.4 Opening summary
+
+- 1-2 paragraphs, plain prose, no headings, no bullets.
+- Hook the reader: lead with the most distinctive thread of the day (a heated debate, a surprising announcement, a market move someone reacted to).
+- Reference 2-4 of the day's category titles in the prose so the reader knows what's coming.
+- Mention 1-2 specific people only if their contribution is central; otherwise stay topic-focused.
+- No timestamps, no message counts (those live in the stats block).
+
+### 1.5 群友画像 section
+
+- Heading line: `群友画像`
+- One entry per user with 3+ messages this batch.
+- Order: by message count, descending.
+- Entry header: `{昵称}（{角色标签}）` — the role tag is your one-line read on this person *today*. Examples: `做空美股的乐子人`, `深夜技术指导`, `论坛级吐槽担当`.
+- Body: 2-5 bullets with `•` prefix. Each bullet states one observation. Quote evidence inline where natural.
+- Continuity: if you loaded a prior profile in Step 3.7, carry forward the established tags/observations that still apply, and call out *change* explicitly (`今天罕见地没提空头`, `从昨天的乐观转向今天的焦虑`).
+- Don't invent backstory — only what's in the messages or the prior profile.
+
+Example:
+
+```
+群友画像
+
+蛙总（做空美股的乐子人）
+• 全天反复提"做空 SPY"，被群友提醒已连续三周看错方向
+• 难得正面回应技术问题："我那个脚本是用 Bun 跑的，慢得跟蜗牛似的"
+• 临近收盘转为沉默，与昨日大放厥词的状态对比明显
+```
+
+### 1.6 Categorized body
+
+- 3-6 self-named categories per day.
+- Each category is a thematic bucket — name it for the *topic*, not generic ("讨论"、"闲聊" are forbidden labels).
+- Category header: `{emoji} {标题}` — one emoji prefix, then a short noun phrase.
+  - Suggested emoji: 🛠 工具/技术，📦 产品发布，📰 新闻/市场，💬 观点辩论，😄 笑料/段子，📚 学习分享，💸 钱与消费，🍜 生活日常。
+- Body inside each category: prose with embedded quotes. Use `•` bullets when listing 3+ parallel items; otherwise paragraphs.
+- Attribution: name the speaker on first mention in a thread (`蛙总说他...`). For follow-on lines in the same thread, attribution can be implicit if the chain is short and clear.
+- Quotes: use 「」 for direct quotes. Quote when the wording is vivid, surprising, or characteristic; paraphrase otherwise.
+- Merge: a multi-person discussion is one entry, not a list of one-line replies.
+- Links: preserve the full URL inline. Article titles stay verbatim.
+
+Example:
+
+```
+🛠 Claude Code 4.7 实测
+
+蛙总下午把 4.7 装上后第一反应是「比 4.6 慢一倍」，老王跟着复现，怀疑是 Opus 默认配置导致。阿喵贴了官方文档 https://docs.claude.com/.../opus-4-7 ，提到可以切回 Sonnet 4.6 跑速测，三人最终结论：复杂任务 4.7 强，日常用 4.6 更顺手。
+```
+
+### 1.7 Pain-point section (optional)
+
+- Include only when the day's chat contains at least one concrete unresolved or partially-resolved problem.
+- Heading: `今日待解决问题` or `本周悬而未决`.
+- One entry per problem. Format:
+  ```
+  问题：<一句话描述>
+  提出者：<昵称>
+  背景：<1-2 句来龙去脉>
+  状态：<✅ 已解决 / ⚠️ 部分解决 / ❌ 仍未解决>
+  方案：<若有人提了方案，写在这；否则写"暂无方案">
+  ```
+- Skip the section entirely if there are no genuine pain points — don't pad with trivial questions.
+
+### 1.8 Footer
+
+Fixed line, last in file:
+
+```
+本简报由 AI 自动生成
+```
+
+No date, no signature, no version number.
+
+---
+
+## 2. Roast version (毒舌版)
+
+The roast version mirrors the normal version's structure but inverts the tone. Generate only when `include_roast=true` (configured via EXTEND.md `default_version` or triggered by request keywords like 毒舌/roast/挑衅).
+
+### 2.1 Structural parity
+
+```
+[Title line — adds "毒舌版" suffix]
+[📊 Stats block — each leaderboard row gets a roast comment]
+[Opening summary — absurd recap, sarcastic]
+[群友画像 — 不留情面版]
+[Categorized body — louder, more brutal category titles]
+[Fixed footer — roast version]
+```
+
+Pain-point section is **dropped** in the roast version.
+
+### 2.2 Title line
+
+Form: `{群名} 群聊精华 · {日期} · 毒舌版`
+
+Example: `相亲相爱一家人 群聊精华 · 2026-03-12 · 毒舌版`
+
+### 2.3 Statistics block (roast)
+
+- Same `📊 消息统计: 共 N 条消息` opener.
+- Leaderboard: each row gets a parenthetical roast comment.
+- Form: `{排名}. {昵称}: {消息数} 条 ({一句毒舌评语})`
+
+Example:
+
+```
+📊 消息统计: 共 387 条消息
+1. 蛙总: 92 条 (一个人撑起了空头的体面)
+2. 老王: 58 条 (主要功能是给蛙总当反例)
+3. 阿喵: 41 条 (发的链接比发的话还多)
+```
+
+### 2.4 Opening summary (roast)
+
+- 1-2 paragraphs, sarcastic recap tone.
+- Highlight the day's most ridiculous beat (a failed prediction, a heated argument over something trivial, a wild flex).
+- Reference 2-3 of the roast category titles in the prose.
+
+### 2.5 群友画像 (roast — 不留情面版)
+
+- Same per-user entry format as normal, but with the role tag dialed up.
+- Amplify quirks, contradictions, fail moments visible in the messages.
+- 2-5 bullets with `•`. Each bullet is a roastable observation backed by a direct quote.
+- Don't invent — every roast must trace back to something the person actually said in the batch (or in the loaded roast profile from prior batches).
+
+Example:
+
+```
+群友画像
+
+蛙总（做空美股钉子户，三周亏损不改其志）
+• 今天第 47 次预测 SPY 见顶，给出的理由是"我感觉"
+• 被老王翻出上周聊天记录后嘴硬：「上周的不算，那是市场不理性」
+• 收盘前突然安静，疑似刚看完账户
+```
+
+### 2.6 Categorized body (roast)
+
+- Same 3-6 category structure.
+- Titles can be louder, more brutal, mock-headline style.
+- Examples of acceptable roast titles:
+  - `蛙总做空翻车: 一个人对抗整个美股`
+  - `老王再次试图当人生导师, 群友集体打哈欠`
+  - `阿喵又分享了一篇没人读完的长文`
+- Bodies still preserve real quotes and traceability — sarcasm is in framing, not fabrication.
+
+### 2.7 Footer (roast)
+
+Fixed line, last in file:
+
+```
+本简报由一个没有感情的 AI 自动生成,如有冒犯,概不负责
+```
+
+### 2.8 Red lines (non-negotiable)
+
+These rules override style; violate them and the roast becomes harmful. Write the spicy version first, then audit against this list and rewrite anything that crosses a line.
+
+- **Mock public group behavior only.** Never appearance, weight, body, health, mental state, family, relationships (unless openly group-discussed at the person's own initiative), finances beyond what they publicly mention.
+- **调侃 ≠ 人身攻击。** Readers should laugh at the situation, not feel sorry for the target. If you can't think of a joke that doesn't read as cruel, drop the bullet.
+- **No timestamp-based sleep/timezone jokes.** Server time ≠ recipient's local time; also implies surveillance. Forbidden: `凌晨 3 点还在群里发言，怕是没睡好`, `这位是哪个时区的`, etc.
+- **No medical/psychological speculation.** Even joking diagnoses (`这位需要看医生`, `典型 ADHD`) are out.
+- **No outing.** Don't infer identity attributes the person hasn't volunteered (orientation, religion, politics beyond direct quotes).
+- **Roast the take, not the person's right to have a take.** `这个观点错得离谱` is fine. `连这都不懂还敢发言` is not.
+
+If a target hasn't said anything roastable in this batch (3+ messages but all neutral), give them a one-line affectionate ribbing and move on. Don't manufacture conflict.
+
+### 2.9 Writing order
+
+1. Draft the spicy version freely — don't self-censor mid-sentence.
+2. After the body is complete, do a separate audit pass against §2.8.
+3. Rewrite or delete any line that crosses a red line.
+4. Then read the whole thing once more for tone calibration: is it fun, or is it mean?
+
+---
+
+## 3. Common formatting rules (both versions)
+
+- **No markdown.** No `**bold**`, no `# headings`, no `*italic*`, no `[link](url)` syntax. Headings are plain text on their own line.
+- **Bullets use `•`.** Not `-`, not `*`, not `1.` for prose-style bullets.
+- **Numbered lists** (`1.`, `2.`) are reserved for the leaderboard.
+- **Subcategory hints** within a body block are plain text with no symbol prefix.
+- **Links preserved verbatim.** Paste the full URL inline. Don't shorten, don't hide behind text.
+- **One emoji per category title.** Don't stack 🛠💬 etc.
+- **Pain-point statuses** use ✅⚠️❌ verbatim.
+- **Quotes use 「」.** Single quotes for nested.
+- **Names verbatim.** Don't abbreviate `蛙总` to `蛙`, don't translate Chinese names, don't anonymize.
+
+---
+
+## 4. Common content rules (both versions)
+
+- **Filter only pure noise.** Cut: lone emoji reactions, "好的"/"收到"/"哈哈哈" with no follow-on, duplicate forwards.
+- **Keep gossip, anecdotes, signature moments.** These are the highlight reel — the whole point of the digest.
+- **Plain language.** Preserve vivid expressions and idiosyncratic phrasings — that's what makes the speaker recognizable.
+- **Keep real names.** Both for traceability and so the digest is useful as memory.
+- **Tool, product, URL names complete.** `Claude Code 4.7`, not `CC`. `https://github.com/...`, not `GitHub 上那个项目`.
+- **Merge, don't list.** A 30-message debate becomes one paragraph, not 30 bullet points.
+- **Direct-quote deep observations.** When someone says something striking, quote it verbatim with 「」 rather than paraphrase.
+- **Shared articles → title + sharer.** `阿喵分享了《一个 Rust 工程师的反思》` — include the title and who shared.
+- **No timestamp-based sleep/timezone inference.** (Repeated here because it applies to both versions, not just roast — never say `凌晨 3 点还在线` in either.)
+- **No fabricated facts.** Every claim must be supported by an actual message in the batch (or in a loaded profile). If you're tempted to "add color," stop.
+
+---
+
+## 5. Output skeleton — quick reference
+
+When you forget the structure mid-write, this is the skeleton:
+
+### Normal
+
+```
+{群名} 群聊精华 · {日期}
+
+📊 消息统计: 共 N 条消息
+1. {昵称}: N 条
+2. {昵称}: N 条
+...
+10. {昵称}: N 条
+
+{开篇 1-2 段，无标题，直入主题}
+
+群友画像
+
+{昵称}（{角色标签}）
+• {观察 1}
+• {观察 2}
+• {观察 3}
+
+{昵称}（{角色标签}）
+• {观察 1}
+• {观察 2}
+
+🛠 {分类标题 1}
+
+{该分类下的整理过的讨论 / 段落 / 引用}
+
+📦 {分类标题 2}
+
+{...}
+
+今日待解决问题（可选，没有就不写）
+
+问题: {一句话}
+提出者: {昵称}
+背景: {1-2 句}
+状态: ⚠️ 部分解决
+方案: {若有}
+
+本简报由 AI 自动生成
+```
+
+### Roast
+
+```
+{群名} 群聊精华 · {日期} · 毒舌版
+
+📊 消息统计: 共 N 条消息
+1. {昵称}: N 条 ({毒舌评语})
+2. {昵称}: N 条 ({毒舌评语})
+...
+
+{毒舌开篇 1-2 段}
+
+群友画像
+
+{昵称}（{放大的角色标签}）
+• {毒舌观察 1}
+• {毒舌观察 2}
+
+🛠 {更大声的分类标题}
+
+{保留真实引用的毒舌叙述}
+
+本简报由一个没有感情的 AI 自动生成,如有冒犯,概不负责
+```
+
+---
+
+## 6. Self-check before saving
+
+Before writing the digest file, mentally walk through:
+
+1. Stats block accurate? Counts match the filtered message set?
+2. Top 10 names resolved (self_display substituted, ambiguous nicknames disambiguated)?
+3. Opening hooks at least one real category title?
+4. Every active user (3+ msgs) has a 画像 entry?
+5. Every category has a topic-named title (not "讨论")?
+6. Every quote uses 「」 and is traceable to a real message?
+7. Links inline and complete?
+8. No markdown bold/heading/link syntax leaked through?
+9. (Roast only) Every roast bullet would pass the §2.8 red-line audit?
+10. Footer line exact match?
@@ -0,0 +1,273 @@
+# Profiles — user portrait files
+
+This reference defines the per-user profile system. Profiles let the digest carry forward observations across many days so the 群友画像 section in each new digest can show continuity (`蛙总今天罕见地没提空头`) instead of starting from scratch.
+
+Two parallel profile directories live alongside each group's digests:
+
+- `profiles/` — observations sourced from the **normal** version of the digest.
+- `profiles-roast/` — observations sourced from the **roast** version.
+
+They are kept strictly separate. The normal-version generation reads only `profiles/`; the roast-version generation reads only `profiles-roast/`. This prevents roast snark from contaminating the sober summary and vice versa.
+
+Load this file during Step 3.7 (load profiles for active users), Step 8.5 (update profiles after digest is written), and Step 9 (backfill).
+
+---
+
+## 1. File format
+
+### 1.1 Path & naming
+
+- Normal: `wechat/{group_id}-{group_name}/profiles/{wxid}-{nickname}.md`
+- Roast: `wechat/{group_id}-{group_name}/profiles-roast/{wxid}-{nickname}.md`
+
+The **stable** identifier is the `wxid` prefix. The `-{nickname}` suffix is for human browsability — if it changes, rename the file.
+
+Filename sanitization: replace `/`, `\`, `:`, `*`, `?`, `"`, `<`, `>`, `|`, NUL, and control characters with `_`. Trim trailing dots and whitespace. Cap total filename length at 200 chars (rare nicknames can be very long).
+
+### 1.2 Frontmatter
+
+YAML frontmatter at the top of every profile file:
+
+```yaml
+---
+name: "<current display name>"
+wxid: "<wxid>"
+aliases: ["<old nickname>", "<even older nickname>"]
+first_seen: "YYYY-MM-DD"
+last_seen: "YYYY-MM-DD"
+total_messages: N
+digest_appearances: N
+avg_messages_per_digest: N.N
+---
+```
+
+Field rules:
+
+- `name`: the most recent display name from `from_nickname` (or `self_display` for the owning user).
+- `wxid`: stable; never changes once written.
+- `aliases`: append-only; every prior display name we've seen. Don't include the current `name` in this list.
+- `first_seen` / `last_seen`: dates of first/most-recent digest appearance, YYYY-MM-DD.
+- `total_messages`: cumulative count across all digests this profile has been updated from.
+- `digest_appearances`: how many digest files this user has 3+ messages in.
+- `avg_messages_per_digest`: `total_messages / digest_appearances`, one decimal.
+
+### 1.3 Free-form body — normal profile
+
+Section headers are plain text on their own line. Order is fixed.
+
+```
+角色标签
+
+• {4-6 短语标签}
+
+关注领域
+
+• {领域 1}
+• {领域 2}
+
+发言风格
+
+{1-3 句描述，可以多段}
+
+互动模式
+
+• {与某某的互动模式}
+• {另一种互动模式}
+
+经典金句
+
+• [YYYY-MM-DD] 「{直接引用}」
+• [YYYY-MM-DD] 「{直接引用}」
+
+标志性事件
+
+• [YYYY-MM-DD] {事件描述}
+• [YYYY-MM-DD] {事件描述}
+```
+
+### 1.4 Free-form body — roast profile
+
+Same plain-text section header style, different sections.
+
+```
+人设标签
+
+• {4-6 放大版标签}
+
+核心槽点
+
+• {可吐槽点 1}
+• {可吐槽点 2}
+
+毒舌语录库
+
+• [YYYY-MM-DD] 「{该用户说过的话} — {简短毒舌点评}」
+• [YYYY-MM-DD] 「{...}」
+
+经典翻车现场
+
+• [YYYY-MM-DD] {翻车描述 + 引用 / 证据}
+• [YYYY-MM-DD] {...}
+```
+
+---
+
+## 2. Update rules
+
+Rules differ per section. Append-only sections must never lose history; mergeable sections may be rewritten as understanding sharpens.
+
+### 2.1 Normal profile
+
+| Section | Update mode | Notes |
+|---------|-------------|-------|
+| 角色标签 | **Merge** | Cap 4-6 tags. Can replace less representative tags with stronger ones. Always keep the most consistently-supported tag. |
+| 关注领域 | **Merge dedupe** | Add new domains; dedupe by meaning, not exact string. |
+| 发言风格 | **Refine** | Only update when a clearly new pattern emerges. Avoid rewriting on every digest. |
+| 互动模式 | **Merge** | Add new modes; can refine existing ones with more detail. |
+| 经典金句 | **Append-only** | Never delete. No cap. Each entry must be dated and quoted verbatim. |
+| 标志性事件 | **Append-only** | Never delete. No cap. Each entry dated. |
+
+### 2.2 Roast profile
+
+| Section | Update mode | Notes |
+|---------|-------------|-------|
+| 人设标签 | **Merge** | Cap 4-6. Can sharpen tags as patterns repeat. |
+| 核心槽点 | **Append-only** | Never delete; recurring 槽点 build up here. |
+| 毒舌语录库 | **Append-only** | Never delete. No cap. Each entry dated, with both the quote and the roast comment. |
+| 经典翻车现场 | **Append-only** | Never delete. No cap. Each entry dated. |
+
+### 2.3 Frontmatter on every update
+
+- Update `name` if current display name differs from the recorded one. Push the old name onto `aliases` if not already there.
+- If `name` changed, also rename the file from `{wxid}-{old_nickname}.md` to `{wxid}-{new_nickname}.md`.
+- Update `last_seen` to the current digest's end date.
+- Increment `total_messages` by this batch's message count for this user.
+- Increment `digest_appearances` by 1.
+- Recompute `avg_messages_per_digest`.
+
+---
+
+## 3. Step 8.5 — Update procedure
+
+Run after the digest file(s) are written. Iterate over every user with 3+ messages in this batch.
+
+1. **Look up the profile.**
+   - Scan `profiles/` (or `profiles-roast/` for the roast pass) for a file whose name starts with `{wxid}-`.
+   - If found: open it.
+   - If not found: create a new file using the frontmatter template. `first_seen = last_seen = current digest end date`, `total_messages = this batch's count`, `digest_appearances = 1`.
+
+2. **Resolve wxid for new users.** When a new user appears, you already know their `wxid` from the wx-cli message data — use it directly. If for some reason only the nickname is known, run `wx contacts --query "{nickname}" --json` to resolve; if multiple matches, prefer the one currently in the group (cross-check `wx members <group>` if needed).
+
+3. **Update frontmatter.** Per §2.3.
+
+4. **Update body sections.**
+   - For mergeable sections (角色标签，关注领域，发言风格，互动模式 / roast: 人设标签): read the existing content, integrate new observations from this batch, rewrite the section.
+   - For append-only sections (经典金句，标志性事件 / roast: 毒舌语录库，经典翻车现场，核心槽点): append new entries, each dated and verbatim. Never edit or remove prior entries.
+
+5. **Write back.** Overwrite the file.
+
+6. **Source separation.** Pass running for the normal digest writes only to `profiles/`. Pass running for the roast digest writes only to `profiles-roast/`. Even if both versions are generated in the same skill invocation, run two separate update passes.
+
+---
+
+## 4. Step 9 — Backfill procedure
+
+Triggered when the user says `回溯画像`, `初始化画像`, `backfill profiles`, or similar. This builds initial profiles from already-written digest files without re-fetching from wx-cli.
+
+1. **List inputs.**
+   - List every `*.md` digest file under `wechat/{group_id}-{group_name}/` (top level, not inside `profiles/` or `profiles-roast/`).
+   - Partition by filename suffix: `*-roast.md` → roast pass, all others → normal pass.
+   - Optionally also read `history-digests.jsonl` for fast metadata lookup (date, message count) before opening individual files.
+
+2. **Decide whether to run roast backfill.** Only run the roast pass if at least one `*-roast.md` file exists.
+
+3. **Process in batches of 10-15 digest files.** Reading all of them at once will blow context. For each batch:
+   - Read the digests.
+   - For each user appearing in the leaderboard or 群友画像 across the batch, accumulate:
+     - Message counts per digest (from the stats block).
+     - Role tags and observations (from the 群友画像 section).
+     - Quotes (from inline 「」 in the body).
+     - Dated events (from category bodies — when the digest mentions specific incidents).
+   - Resolve wxid for each accumulated user via `wx contacts --query "{nickname}" --json` if not already cached. Cache the wxid↔nickname mapping for the rest of the backfill.
+
+4. **Threshold.** Generate a profile file only for users appearing in **3 or more** digests in the corpus. Below that, skip (probably one-time visitors).
+
+5. **Write profile files.**
+   - For the normal pass, write to `profiles/{wxid}-{nickname}.md`.
+   - For the roast pass, write to `profiles-roast/{wxid}-{nickname}.md`.
+   - Use the most recent nickname as the filename suffix. Push older nicknames into `aliases`.
+   - Sort 经典金句，标志性事件，毒舌语录库，经典翻车现场 entries chronologically by date.
+   - No cap on the size of append-only sections during backfill — let history flow in.
+
+6. **Compute frontmatter.**
+   - `first_seen` = earliest digest date the user appeared in.
+   - `last_seen` = latest digest date the user appeared in.
+   - `total_messages` = sum of per-digest counts.
+   - `digest_appearances` = number of digests the user crossed the 3-message threshold in.
+
+7. **Report.** After both passes complete, print a short summary:
+   - `Backfilled {N} normal profiles from {M} digests.`
+   - `Backfilled {K} roast profiles from {L} roast digests.` (only if roast pass ran)
+   - List any users skipped due to wxid resolution failures so the user can fix manually.
+
+8. **Re-running backfill is safe.** If the user runs backfill twice, treat existing profile files as the prior state and merge — same rules as Step 8.5 updates. Don't blow away existing append-only entries.
+
+---
+
+## 5. Privacy guardrails
+
+These apply to both normal and roast profiles, with an extra layer for roast.
+
+### 5.1 Forbidden (write neither in normal nor roast)
+
+- **Real-world full names** when only a nickname was used in the group. If the person introduced themselves with `我叫王二`, `王二` is on the table; `王晓明` inferred from another channel is not.
+- **Phone numbers, emails, ID numbers, home addresses, employer addresses, exact birth dates** — even if mentioned in the group, don't lift them into profile files.
+- **Health, medical, psychological information.** Even self-disclosed (`我最近有点抑郁`) — don't bake it into a permanent profile.
+- **Private romantic / family details** unless openly group-discussed by the person themselves. A passing mention by another member doesn't count.
+- **Embarrassing private failures.** Public ones (a take that aged badly in front of the group) are fair game; private ones (a job rejection mentioned briefly) are not.
+- **Sleep / timezone inference from timestamps.** Server time ≠ recipient's local time, and it implies surveillance.
+
+### 5.2 Allowed
+
+- **Public group behavior** — what they said, how they argued, what they shared.
+- **Direct quotes** of things said in the group (these are already public to the group).
+- **Interest areas, hobbies, tool preferences** as expressed in group discussion.
+- **Interaction patterns** with other group members.
+- **Publicly mentioned consumption** (`蛙总今天又分享了买了什么书`) — fine if they themselves mentioned it.
+- **Publicly shared travel / life anecdotes** they told the group.
+
+### 5.3 Roast-only extras
+
+In addition to §5.1, the roast profile must **not** include:
+
+- **Anything about appearance, weight, body, looks.**
+- **Anything about family members** (their kids, parents, partners) — only the person themselves.
+- **Mental-health speculation**, even as a joke. No `这位需要看医生`, no `典型 ADHD`.
+- **Identity-based roasts.** No mocking of orientation, religion, ethnicity, nationality, gender.
+
+The roast may mock:
+
+- Stupid takes, contradictions, factual errors.
+- Repetitive behavior (`第 47 次预测见顶`).
+- Self-undermining moments (`昨天说 X，今天说 not X`).
+- Performative flexes that didn't land.
+
+The rule of thumb: **roast the take, not the person.**
+
+---
+
+## 6. Reading profiles during digest generation (Step 3.7)
+
+When loading profile context for a fresh digest:
+
+1. Iterate over users active in this batch (3+ messages).
+2. For the normal pass, read `profiles/{wxid}-*.md` for each. Skip if missing.
+3. If the current run also generates the roast version, **separately** read `profiles-roast/{wxid}-*.md` during the roast generation pass.
+4. Compile a condensed working-memory block:
+   - The user's current `name` and `aliases` (so you can recognize them under different names).
+   - 角色标签 / 人设标签 (so you can carry forward or contrast).
+   - The 3-5 most recent 经典金句 / 毒舌语录 entries (so you can detect callbacks and repeats).
+   - The 3-5 most recent 标志性事件 / 翻车现场 entries (so you can spot recurring themes).
+5. Don't dump the entire profile into the digest — the profile is *context*, the digest is *today*.
+
+If a profile contradicts what you see in today's batch (e.g., the profile says `从不主动发起话题`, but today they started three threads), call that out explicitly in the day's 群友画像 — that's the kind of contrast that makes the digest interesting.
@@ -29,11 +29,14 @@ When this skill needs to render an image, resolve the backend in this order:
 1. **Current-request override** — if the user names a specific backend in the current message, use it.
 2. **Saved preference** — if `EXTEND.md` sets `preferred_image_backend` to a backend available right now, use it.
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
-   - If the current runtime exposes a native image tool (e.g., Codex `imagegen`, Hermes `image_generate`), use it. Runtime-native tools are preferred by default — agents that know their own tool inventory should surface the native one here.
+   - **Codex (`imagegen`)** — first, inspect your available-skills / tool inventory. If a skill named `imagegen` is listed, you are running inside Codex and MUST use it: invoke via the `Skill` tool with `skill: "imagegen"`, passing the saved prompt file's content (plus output path and aspect ratio per Codex `imagegen`'s own args). Codex `imagegen` is the official raster backend in that runtime and outranks any non-native skill (e.g., `baoyu-imagine`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Other runtime-native tools** — if the runtime exposes a different native image tool (e.g., Hermes `image_generate`), use it the same way.
   - Otherwise, if exactly one non-native backend is installed (e.g., `baoyu-imagine`), use it.
   - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
 4. **If none are available**, tell the user and ask how to proceed.

+**⛔ Never substitute SVG, HTML, canvas, or other code-based rendering for raster image generation.** Codex `imagegen`'s own description says it should be used "when the output should be a bitmap asset rather than repo-native code or vector." If you cannot resolve a raster backend via step 3, fall through to step 4 and ask the user — do **not** silently emit SVG, write inline `<svg>` markup, or produce HTML/CSS art as a substitute. This applies even if the article/section seems "diagram-like": the consumer skill calling this rule has already decided that a raster image is what it needs.
+
 Setting `preferred_image_backend: ask` forces the step-3 prompt every run regardless of available backends. Users change the pinned backend via the `## Changing Preferences` section below.

 **Prompt file requirement (hard)**: write each image's full, final prompt to a standalone file under `prompts/` (naming: `NN-{type}-[slug].md`) BEFORE invoking any backend. The file is the reproducibility record and lets you switch backends without regenerating prompts.
Author	SHA1	Message	Date
Jim Liu 宝玉	20ebf6126c	chore: release v1.116.3	2026-05-13 22:33:29 -05:00
Jim Liu 宝玉	64db328e61	docs: replace Claude Code references with Agent in READMEs	2026-05-13 22:33:16 -05:00
Jim Liu 宝玉	a3819b8e30	chore: release v1.116.2	2026-05-13 22:30:54 -05:00
Jim Liu 宝玉	234c2a832b	docs(baoyu-wechat-summary): update example group name in SKILL.md	2026-05-13 22:30:38 -05:00
Jim Liu 宝玉	7943eb7b05	chore: release v1.116.1	2026-05-13 22:28:34 -05:00
Jim Liu 宝玉	4b4a4d3863	feat(baoyu-wechat-summary): add data_root to first-time setup flow	2026-05-13 22:28:13 -05:00
Jim Liu 宝玉	919849c863	chore: release v1.116.0	2026-05-13 22:18:01 -05:00
Jim Liu 宝玉	1faee80da4	feat(baoyu-wechat-summary): add WeChat group chat summary skill	2026-05-13 22:16:39 -05:00
Jim Liu 宝玉	dea9322d0d	chore: release v1.115.4	2026-05-11 18:45:31 -05:00
Jim Liu 宝玉	86b1be83b8	docs(image-generation): emphasize Codex imagegen priority and forbid SVG/HTML substitution Strengthen the shared image-generation backend selection rule: - Codex `imagegen` MUST be used when listed in the available-skills inventory; invoke via the `Skill` tool with `skill: "imagegen"`. - Never substitute SVG, HTML, canvas, or other code-based rendering when no raster backend can be resolved — fall through and ask the user instead. Applied to docs/image-generation-tools.md and inlined into baoyu-article-illustrator (plus its references/workflow.md), baoyu-comic, baoyu-cover-image, baoyu-image-cards, baoyu-infographic, baoyu-slide-deck, baoyu-xhs-images per the self-containment rule.	2026-05-11 18:45:13 -05:00
Jim Liu 宝玉	a1b935b2d8	chore: release v1.115.3	2026-05-11 16:24:14 -05:00
Jim Liu 宝玉	9968d79a26	fix(baoyu-post-to-x): use toolbar media upload instead of image clipboard paste	2026-05-11 16:23:17 -05:00
Jim Liu 宝玉	6f75fb17e4	chore: add .codex-tmp and outputs to .gitignore	2026-05-11 16:23:09 -05:00
Jim Liu 宝玉	adba24281b	Merge pull request #149 from fengxiaodong28/fix-browser-copy-paste fix(browser): ensure tab activation before copy/paste in WeChat editor	2026-05-10 11:37:40 -05:00
Jim Liu 宝玉	61342ecfea	chore: release v1.115.2	2026-05-10 02:49:50 -05:00
Jim Liu 宝玉	5f753dd584	fix(baoyu-post-to-x): respect Chrome plugin mode	2026-05-10 02:49:42 -05:00
Jim Liu 宝玉	076192d58e	chore: release v1.115.1	2026-05-09 22:10:54 -05:00
Jim Liu 宝玉	d6d434e714	fix(image): update MiniMax default endpoint	2026-05-09 22:07:26 -05:00
Jim Liu 宝玉	6f3600d8e5	chore: release v1.115.0	2026-05-09 22:05:27 -05:00
Jim Liu 宝玉	045fe5e57e	feat(baoyu-post-to-x): add Chrome Computer Use as preferred execution mode In Codex, prefer the bundled Chrome Computer Use path for all X UI actions (compose, article, quote, video). CDP scripts become a fallback when Computer Use is unavailable or explicitly not requested.	2026-05-09 22:04:58 -05:00
FENG/XIAODONG	0b3b7d13b5	fix(browser): ensure tab activation before copy/paste in WeChat editor When posting articles via browser automation, the HTML preview tab needs to be active before copying content, and the editor tab needs to be active before pasting. Without explicit Target.activateTarget calls, AppleScript Cmd+C/Cmd+V would act on the wrong tab, causing the editor body to remain empty after paste. This fix adds Target.activateTarget before both copy and paste operations, ensuring the correct tab is in focus for system-level clipboard operations.	2026-05-09 22:13:06 +08:00
Jim Liu 宝玉	aa1a967a9f	chore: release v1.114.1	2026-05-08 17:45:03 -05:00
Jim Liu 宝玉	d643bad53c	test(baoyu-danger-gemini-web): cover generated image response fallback	2026-05-08 17:43:31 -05:00
Jim Liu 宝玉	0d977787b5	Merge pull request #146 from evilstar2016/fix/gemini-generated-image-detection fix(gemini-webapi): add fallback scan for generated images	2026-05-08 17:40:45 -05:00
evilstar2016	516803feb4	fix(gemini-webapi): add fallback scan for generated images when wants_generated fails The `wants_generated` detection checks `candidate[12][7][0]` and an old `googleusercontent.com/image_generation_content/` URL pattern in the response text. Both are no longer present in the current Gemini Web API response format, causing the entire generated-image extraction block to be skipped even when Gemini successfully generates images — resulting in "No image returned in response" errors. Generated image URLs now appear as `https://lh3.googleusercontent.com/gg-dl/` somewhere in the response parts. This commit adds an unconditional fallback that scans all response parts for those URLs when `generated_images` is still empty after the existing `wants_generated` block, reusing the already-present `collect_strings` helper and `GeneratedImage` constructor. The existing code path is untouched — the fallback only runs when no images were found through the original logic, so old response formats continue to work.	2026-05-06 18:48:08 +09:00