Merge branch 'main' of https://github.com/JimLiu/baoyu-skills

* fix(baoyu-post-to-x): render CJK-adjacent bold/italics in articles

md-to-html.ts preprocessed markdown with remark + remark-cjk-friendly, then re-stringified and re-parsed with marked. remark-cjk-friendly parses emphasis adjacent to CJK correctly, but remark-stringify re-emits the **bold*/*italic* markers, which marked then FAILS to parse: its flanking test does not treat a CJK character after the closing delimiter as valid punctuation/whitespace. Result: literal **...** asterisks leak into the pasted X Article body whenever the closing delimiter is directly followed by a CJK character.

Fix: convert strong/emphasis mdast nodes (already identified correctly by remark-cjk-friendly) into raw inline <strong>/<em> HTML before stringify, so marked passes them through untouched. A small recursive serializer preserves nested links, inline code, images, etc.

- add remarkStrongEmToHtml remark plugin + inline mdast serializer
- add unist-util-visit as an explicit dependency
- add regression test covering CJK-adjacent bold and italics

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(baoyu-post-to-x): 修复紧贴中文的加粗/斜体在 X Articles 渲染为字面 ** 的问题

按维护者建议改用 remark-cjk-friendly 标准用法，替代自定义插件/序列化器：

- 删除 remarkStrongEmToHtml 插件与 serializeMdastNode 序列化器
- preprocessCjkMarkdown 简化为 remarkParse + remarkCjkFriendly + remarkStringify 往返
- 移除原 markdown 阶段的 &#x...; 实体解码（该解码会还原 cjk-friendly 为
  骗过 marked flanking 判定而生成的实体，正是 CJK 加粗失败的根因）
- 改在 marked 渲染出最终 HTML 后再 decodeHtmlEntities，此时 markdown 语法
  已不存在，解码安全且保持输出干净
- 移除不再使用的 unist-util-visit 显式依赖
- 测试补充引用式链接 **[docs][d]** 在强调内的回归断言（顺带修复 Codex
  指出的 linkReference 渲染为纯文本的 P2 问题）

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(baoyu-post-to-x): 移除渲染后无差别实体解码，避免篡改作者字面实体

回应 Codex review (P2)：decodeHtmlEntities 对整个渲染 HTML 做无差别
&#x...; 解码，会把作者为显示字面 HTML 而写的实体（如 &#x3C;b&#x3E;）
解码成真实标签，存在内容篡改与 HTML 注入风险。

该解码本就只是为了让输出 HTML「好看」，而 remark-cjk-friendly 引入的
实体是合法 HTML 字符引用，粘进 X 编辑器时能正确渲染；且这些实体只会
出现在强调 span 的外侧（边界字符），不会出现在 span 内部。因此直接
移除解码，既消除过度解码风险，又无功能损失。

- 删除 decodeHtmlEntities 及其在 convertMarkdownToHtml 的调用
- 新增测试：作者字面实体 &#x3C;b&#x3E; 不被解码为真实 <b>，CJK 加粗
  仍正常渲染
- bun test 8 个用例全部通过；真实文章端到端验证 17 处加粗、0 字面 **、
  0 真实标签注入

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: ruofeng <ruofeng.wang@rd.group>
Co-authored-by: Claude <noreply@anthropic.com>

.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
   },
   "metadata": {
     "description": "Skills shared by Baoyu for improving daily work efficiency",
-    "version": "2.5.0"
+    "version": "2.5.2"
   },
   "plugins": [
     {

CHANGELOG.md

@@ -2,6 +2,19 @@
 
 English | [中文](./CHANGELOG.zh.md)
 
+## 2.5.2 - 2026-06-18
+
+### Fixes
+- `codex-imagegen`: stop Codex from satisfying a request by copying an unrelated pre-existing image from `generated_images` instead of generating a new one. Verification now requires real evidence that `image_gen` ran in the current thread — an `image_gen` event in the stream **or** a PNG in this thread's `generated_images` dir — and the spawned-agent instruction forbids reading or reusing history images before `image_gen` is called ([#185](https://github.com/JimLiu/baoyu-skills/issues/185))
+
+### Documentation
+- README: document Codex project-level install (`.agents/skills`) and project- vs user-level WeChat credential scopes
+
+## 2.5.1 - 2026-06-13
+
+### Documentation
+- Image generation skills: document Cursor's native `GenerateImage` backend in the shared backend-selection rule, including prompt-based aspect-ratio guidance, tool-managed output handling, and `reference_image_paths` support.
+
 ## 2.5.0 - 2026-06-12
 
 ### Features

CHANGELOG.zh.md

@@ -2,6 +2,19 @@
 
 [English](./CHANGELOG.md) | 中文
 
+## 2.5.2 - 2026-06-18
+
+### 修复
+- `codex-imagegen`：阻止 Codex 通过复制 `generated_images` 中无关旧图片来代替真正生成的取巧行为。校验改为要求当前线程内 `image_gen` 真正运行的证据——事件流中的 `image_gen` 事件，**或**当前线程 `generated_images` 目录下新生成的 PNG——并在派生 agent 指令中禁止在调用 `image_gen` 之前读取或复用历史图片（[#185](https://github.com/JimLiu/baoyu-skills/issues/185)）
+
+### 文档
+- README：补充 Codex 项目级安装（`.agents/skills`）以及项目级与全局公众号凭证的放置位置
+
+## 2.5.1 - 2026-06-13
+
+### 文档
+- 图片生成类 skills：在共享后端选择规则中补充 Cursor 原生 `GenerateImage` 后端说明，包括通过提示词声明宽高比、处理工具托管输出文件，以及使用 `reference_image_paths` 传入参考图。
+
 ## 2.5.0 - 2026-06-12
 
 ### 新功能

CLAUDE.md

@@ -64,7 +64,7 @@ Skills that prompt users for choices MUST declare the tool-selection convention
 
 ## Image Generation Tools
 
-Skills that render images MUST declare the backend-selection convention **inline** in exactly one place per `SKILL.md` — a `## Image Generation Tools` section near the top (after `## User Input Tools`). Do NOT link out to [docs/image-generation-tools.md](docs/image-generation-tools.md); that doc is the author-side canonical source — copy its body into each SKILL.md. Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) elsewhere in a skill are treated as examples — other runtimes substitute their local equivalent under the rule. The rule is stateless: use whatever backend is available; if multiple, ask the user once; if none, ask how to proceed. Every rendered image's full prompt must be written to a standalone `prompts/NN-*.md` file before any backend is invoked. Backend skills (`baoyu-image-gen`, `baoyu-danger-gemini-web`) are exempt — they render directly rather than selecting a backend.
+Skills that render images MUST declare the backend-selection convention **inline** in exactly one place per `SKILL.md` — a `## Image Generation Tools` section near the top (after `## User Input Tools`). Do NOT link out to [docs/image-generation-tools.md](docs/image-generation-tools.md); that doc is the author-side canonical source — copy its body into each SKILL.md. Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) elsewhere in a skill are treated as examples — other runtimes substitute their local equivalent under the rule. The rule is stateless: use whatever backend is available; if multiple, ask the user once; if none, ask how to proceed. Every rendered image's full prompt must be written to a standalone `prompts/NN-*.md` file before any backend is invoked. Backend skills (`baoyu-image-gen`, `baoyu-danger-gemini-web`) are exempt — they render directly rather than selecting a backend.
 
 ### `codex-imagegen` Backend
 

LICENSE

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

README.md

@@ -19,6 +19,31 @@ Skills shared by Baoyu for improving daily work efficiency with AI Agents (Claud
 npx skills add jimliu/baoyu-skills
 ```
 
+### Codex Project-Level Install
+
+If you only need a subset of skills in one project, you do not need to install the full plugin. Codex scans `.agents/skills` inside a project, so copy or symlink each needed skill as a full directory:
+
+```text
+<project>/.agents/skills/baoyu-cover-image/SKILL.md
+<project>/.agents/skills/baoyu-article-illustrator/SKILL.md
+<project>/.agents/skills/baoyu-post-to-wechat/SKILL.md
+```
+
+For a WeChat Official Account article workflow, the usual minimal set is:
+
+- `baoyu-cover-image`
+- `baoyu-article-illustrator`
+- `baoyu-post-to-wechat`
+
+You do not need to install `baoyu-markdown-to-html` separately. `baoyu-post-to-wechat` already includes the Markdown to WeChat-ready HTML conversion flow. Install `baoyu-format-markdown` only if you need to first turn raw text or drafts into structured Markdown articles with titles, summaries, headings, bold text, lists, and similar formatting.
+
+Place WeChat API credentials according to the scope you want:
+
+- User-level: `~/.baoyu-skills/.env`
+- Project-level: `<project>/.baoyu-skills/.env`
+
+Project-level `.env` files are useful when credentials should apply only to the current project. Do not commit them to Git.
+
 ### Publish to ClawHub / OpenClaw
 
 This repository now supports publishing each `skills/baoyu-*` directory as an individual ClawHub skill.
@@ -1362,7 +1387,9 @@ This project was inspired by and builds upon the following open source projects:
 
 ## License
 
-MIT
+Unless otherwise noted, this repository is licensed under the [MIT License](./LICENSE).
+
+Published ClawHub skills follow ClawHub registry rules and are distributed under `MIT-0`. Third-party code and assets retain their original licenses where noted.
 
 ## Star History
 

README.zh.md

@@ -29,6 +29,31 @@
 npx skills add jimliu/baoyu-skills
 ```
 
+### Codex 项目级安装
+
+如果只在某个项目中使用部分技能，不需要安装整个插件。Codex 会扫描项目里的 `.agents/skills`，可以只把需要的 skill 按整个目录复制或软链接到当前项目：
+
+```text
+<project>/.agents/skills/baoyu-cover-image/SKILL.md
+<project>/.agents/skills/baoyu-article-illustrator/SKILL.md
+<project>/.agents/skills/baoyu-post-to-wechat/SKILL.md
+```
+
+公众号文章发布的最小组合通常是：
+
+- `baoyu-cover-image`
+- `baoyu-article-illustrator`
+- `baoyu-post-to-wechat`
+
+不需要单独安装 `baoyu-markdown-to-html`，`baoyu-post-to-wechat` 已内置 Markdown 到公众号可用 HTML 的转换流程。只有需要先把原始文本或草稿整理成 Markdown 文章结构（标题、摘要、层级标题、加粗、列表等）时，再额外安装 `baoyu-format-markdown`。
+
+公众号 API 凭证按作用范围放置：
+
+- 全局：`~/.baoyu-skills/.env`
+- 项目：`<project>/.baoyu-skills/.env`
+
+项目级 `.env` 适合只给当前项目使用，注意不要提交到 Git。
+
 ### 发布到 ClawHub / OpenClaw
 
 现在这个仓库支持把每个 `skills/baoyu-*` 目录作为独立 ClawHub skill 发布。
@@ -1363,7 +1388,9 @@ HTTP_PROXY=http://127.0.0.1:7890 HTTPS_PROXY=http://127.0.0.1:7890 /baoyu-danger
 
 ## 许可证
 
-MIT
+除另有说明外，本仓库采用 [MIT License](./LICENSE) 授权。
+
+发布到 ClawHub 的 skill 根据 ClawHub registry 规则以 `MIT-0` 分发。第三方代码和素材按其注明的原始许可授权。
 
 ## Star History
 

docs/creating-skills.md

@@ -165,11 +165,11 @@ Standard snippet (copy verbatim):
 
 When this skill needs to render an image:
 
-- **Use whatever image-generation tool or skill is available** in the current runtime — e.g., Codex `imagegen`, Hermes `image_generate`, `baoyu-image-gen`, or any equivalent the user has installed.
+- **Use whatever image-generation tool or skill is available** in the current runtime — e.g., Codex `imagegen`, Cursor `GenerateImage`, Hermes `image_generate`, `baoyu-image-gen`, or any equivalent the user has installed.
 - **If multiple are available**, ask the user **once** at the start which to use (batch with any other initial questions).
 - **If none are available**, tell the user and ask how to proceed.
 
 **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.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
 ```

docs/image-generation-tools.md

@@ -1,6 +1,6 @@
 # Image Generation Tools
 
-Skills in this repo are loaded by multiple agent runtimes (Claude Code, Codex, Hermes, other agents, bare CLI). Each runtime exposes a different image-generation capability — some have a runtime-native tool (Codex `imagegen`, Hermes `image_generate`), others rely on an installed skill (`baoyu-image-gen`, or user-defined). This document defines the canonical **backend-selection rule** every skill that renders images follows so skills stay portable.
+Skills in this repo are loaded by multiple agent runtimes (Claude Code, Codex, Cursor, Hermes, other agents, bare CLI). Each runtime exposes a different image-generation capability — some have a runtime-native tool (Codex `imagegen`, Cursor `GenerateImage`, Hermes `image_generate`), others rely on an installed skill (`baoyu-image-gen`, or user-defined). This document defines the canonical **backend-selection rule** every skill that renders images follows so skills stay portable.
 
 ## The Rule
 
@@ -10,6 +10,7 @@ When a skill needs to render an image, resolve the backend in this order:
 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):
    - **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-image-gen`) unless the user has explicitly pinned a different `preferred_image_backend`.
+   - **Cursor (`GenerateImage`)** — if the runtime exposes a native `GenerateImage` tool, you are running inside Cursor and it outranks any non-native skill the same way Codex `imagegen` does. Two hard caveats: (a) it has no aspect-ratio parameter — state the target aspect ratio / dimensions explicitly in the prompt text passed as `description`; (b) it does not accept an output directory — it saves to a tool-managed location, so after generation copy/move the file to the skill's expected output path (e.g., `outputs/.../NN-xxx.png`). Reference images go in `reference_image_paths`.
    - **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-image-gen`), use it.
    - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
@@ -27,7 +28,7 @@ Each image-consuming skill's `EXTEND.md` carries a single `preferred_image_backe
 |---|---|
 | `auto` (default) | Apply the auto-select rule — runtime-native preferred, fall back to only installed backend, ask if multiple non-native. |
 | `ask` | Always confirm the backend on every run, even when a runtime-native tool exists. |
-| `<backend-id>` (e.g., `codex-imagegen`, `baoyu-image-gen`, `image_generate`) | Pin this backend when available; fall back to `auto` if it isn't. |
+| `<backend-id>` (e.g., `codex-imagegen`, `baoyu-image-gen`, `GenerateImage`, `image_generate`) | Pin this backend when available; fall back to `auto` if it isn't. |
 
 The field is **absent-equals-auto**: older `EXTEND.md` files without this field behave exactly as if `preferred_image_backend: auto` were set. No schema version bump is needed to introduce it.
 
@@ -41,7 +42,7 @@ Each `SKILL.md` that renders images includes **exactly one** `## Image Generatio
 
 Each skill's `references/config/preferences-schema.md` (and its `EXTEND.md` template in `first-time-setup.md`) lists `preferred_image_backend` alongside other preference fields. First-time setup does NOT ask the user about the backend — `auto` is set silently. Users who want to pin a specific backend edit `EXTEND.md` later, and each skill's `## Changing Preferences` section documents the common one-line edits.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) in this document and in SKILL.md are **examples** — agents in other runtimes apply the rule above and substitute the local equivalent. Skill-specific parameters for these backends are illustrative; runtimes without those knobs can omit them.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) in this document and in SKILL.md are **examples** — agents in other runtimes apply the rule above and substitute the local equivalent. Skill-specific parameters for these backends are illustrative; runtimes without those knobs can omit them.
 
 ## Backend Skills Are Exempt
 

package-lock.json

@@ -5,6 +5,7 @@
   "packages": {
     "": {
       "name": "baoyu-skills",
+      "license": "MIT",
       "workspaces": [
         "packages/*"
       ],
@@ -5018,12 +5019,14 @@
     },
     "packages/baoyu-chrome-cdp": {
       "version": "0.1.1",
+      "license": "MIT",
       "engines": {
         "bun": ">=1.2.0"
       }
     },
     "packages/baoyu-codex-imagegen": {
       "version": "0.1.0",
+      "license": "MIT",
       "bin": {
         "codex-imagegen": "src/main.ts"
       },
@@ -5033,6 +5036,7 @@
     },
     "packages/baoyu-fetch": {
       "version": "0.1.2",
+      "license": "MIT",
       "dependencies": {
         "@mozilla/readability": "^0.6.0",
         "chrome-launcher": "^1.2.1",
@@ -5061,6 +5065,7 @@
     },
     "packages/baoyu-md": {
       "version": "0.1.1",
+      "license": "MIT",
       "dependencies": {
         "fflate": "^0.8.2",
         "front-matter": "^4.0.2",

package.json

@@ -1,6 +1,7 @@
 {
   "name": "baoyu-skills",
   "private": true,
+  "license": "MIT",
   "type": "module",
   "workspaces": [
     "packages/*"

packages/baoyu-chrome-cdp/package.json

@@ -1,6 +1,7 @@
 {
   "name": "baoyu-chrome-cdp",
   "version": "0.1.1",
+  "license": "MIT",
   "type": "module",
   "files": [
     "dist",

packages/baoyu-codex-imagegen/package.json

@@ -1,6 +1,7 @@
 {
   "name": "baoyu-codex-imagegen",
   "version": "0.1.0",
+  "license": "MIT",
   "type": "module",
   "description": "Generate images via Codex CLI's built-in image_gen tool from non-Codex runtimes (Claude Code, Hermes, …).",
   "bin": {

packages/baoyu-codex-imagegen/src/main.ts

@@ -6,7 +6,7 @@ import process from "node:process";
 import { setTimeout as delay } from "node:timers/promises";
 import { GenError, type CliOptions, type GenerateResult } from "./types.ts";
 import { runCodexExec } from "./spawn.ts";
-import { findCpToTarget, verifyImageGenWasInvoked, verifyOutput } from "./validator.ts";
+import { hasImageGenEvidence, verifyImageGenWasInvoked, verifyOutput } from "./validator.ts";
 import { cacheKey, lookupCache, storeCache, FileLock } from "./cache.ts";
 import { JsonLogger } from "./logger.ts";
 
@@ -125,7 +125,7 @@ function buildInstruction(prompt: string, opts: CliOptions): string {
   const refHint = opts.refImages.length > 0
     ? `\nREFERENCE IMAGES (attached above): ${opts.refImages.length} image(s) provided for style/composition guidance.\n`
     : "";
-  return `You have an internal tool called image_gen for image generation. Use it.
+  return `You have an internal tool called image_gen for image generation. You MUST call it before doing anything else.
 
 TASK: Generate an image with the spec below, then save to disk.
 
@@ -137,12 +137,15 @@ OUTPUT PATH: ${opts.outputPath}
 ${refHint}
 STEPS:
 1. Call image_gen with the prompt and aspect ratio above${opts.refImages.length > 0 ? " (using the attached reference images for guidance)" : ""}.
-2. Move or copy the resulting image from Codex default location ($CODEX_HOME/generated_images/...) to: ${opts.outputPath}
+2. Move or copy ONLY the image produced by that image_gen call from Codex default location ($CODEX_HOME/generated_images/...) to: ${opts.outputPath}
 3. Verify with: ls -la ${opts.outputPath}
 4. Reply with ONLY this JSON line (no markdown fences, no other text):
    {"status":"ok","path":"${opts.outputPath}","bytes":<file_size_in_bytes>}
 
 HARD CONSTRAINTS:
+- Do NOT search for, find, inspect, reuse, or copy any pre-existing files from $CODEX_HOME/generated_images/ or any other directory.
+- Do NOT run ls/find/rg/grep/glob over $CODEX_HOME/generated_images/ before image_gen has been called.
+- You MUST call image_gen first. Only after image_gen completes may you copy the newly created file from this turn.
 - Do NOT use curl, wget, Python, or any external API.
 - Do NOT use bash to fabricate an image; only image_gen produces real pixels.
 - Use ONLY the image_gen internal tool.`;
@@ -175,13 +178,16 @@ async function attemptGenerate(
     throw new GenError("agent_refused", "No thread id in event stream");
   }
 
-  // verify: image_gen was actually invoked (check $CODEX_HOME/generated_images/{threadId}/)
+  // verify image_gen ran in THIS thread. A PNG in this thread's
+  // generated_images dir is the real signal (image_gen does not surface as a
+  // stream item); the stream check is a forward-compatible fallback. The #185
+  // shortcut (copying an unrelated history image) yields neither.
   const ver = await verifyImageGenWasInvoked(run.threadId);
-  if (!ver.ok) {
-    // secondary verify: did tool_calls include cp/mv from generated_images to our target
-    if (!findCpToTarget(run.toolCalls, opts.outputPath)) {
-      throw new GenError("no_image_gen_tool_use", `image_gen was not invoked: ${ver.reason}`);
-    }
+  if (!hasImageGenEvidence(run.toolCalls, ver.ok)) {
+    throw new GenError(
+      "no_image_gen_tool_use",
+      `image_gen was not invoked (no image_gen event in stream; ${ver.reason})`,
+    );
   }
 
   // verify output

packages/baoyu-codex-imagegen/src/parser.test.ts

@@ -33,17 +33,26 @@ test("parseEventStream tolerates malformed lines", () => {
   expect(r.usage?.input).toBe(1);
 });
 
-test("hasImageGenInvocation finds shell calls touching generated_images", () => {
+test("hasImageGenInvocation does not infer image_gen from shell copies", () => {
   const r = parseEventStream(REAL_PoC_STREAM);
-  // image_gen itself is not an event; inferred via generated_images cp path
-  // this test only verifies parser behavior; driver logic lives in validator
   const hasCp = r.toolCalls.some((tc) => tc.command?.includes("generated_images"));
   expect(hasCp).toBe(true);
+  expect(hasImageGenInvocation(r.toolCalls)).toBe(false);
 });
 
-test("hasImageGenInvocation (proper) returns false when no image_gen tool", () => {
+test("hasImageGenInvocation detects real image_gen tool calls only", () => {
   expect(hasImageGenInvocation([{ id: "1", tool: "shell", status: "completed" }])).toBe(false);
   expect(
     hasImageGenInvocation([{ id: "1", tool: "image_gen", status: "completed" }]),
   ).toBe(true);
 });
+
+test("parseEventStream normalizes an image_generation item to image_gen", () => {
+  const stream = `{"type":"thread.started","thread_id":"t1"}
+{"type":"item.started","item":{"id":"g0","type":"image_generation","status":"in_progress"}}
+{"type":"item.completed","item":{"id":"g0","type":"image_generation","status":"completed"}}
+{"type":"item.completed","item":{"id":"m0","type":"agent_message","text":"done"}}`;
+  const r = parseEventStream(stream);
+  expect(r.toolCalls.some((tc) => tc.tool === "image_gen")).toBe(true);
+  expect(hasImageGenInvocation(r.toolCalls)).toBe(true);
+});

packages/baoyu-codex-imagegen/src/validator.test.ts

@@ -2,11 +2,52 @@ import { test, expect } from "bun:test";
 import { mkdtemp, writeFile, rm, mkdir } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import path from "node:path";
-import { verifyOutput, verifyImageGenWasInvoked, findCpToTarget } from "./validator.ts";
+import { verifyOutput, verifyImageGenWasInvoked, hasImageGenEvidence } from "./validator.ts";
+import { parseEventStream } from "./parser.ts";
 import { GenError } from "./types.ts";
 
 const PNG_HEADER = Buffer.from([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]);
 
+// Condensed + sanitized capture of a REAL successful `codex exec --json` run
+// (thread 019edc1c…, 16:9 maple-tree image, single attempt). The image_gen
+// tool leaves NO stream item — success shows only reasoning / command_execution
+// / agent_message, and a `cp` from generated_images/<FULL-thread-id>/ig_*.png.
+const REAL_SUCCESS_STREAM = `{"type":"thread.started","thread_id":"019edc1c-e7a3-74f0-a276-13bea71d32d6"}
+{"type":"turn.started"}
+{"type":"item.completed","item":{"id":"r0","type":"reasoning"}}
+{"type":"item.completed","item":{"id":"r1","type":"reasoning"}}
+{"type":"item.completed","item":{"id":"m0","type":"agent_message","text":"Image generation is complete. Locating the newly produced image and copying it to the requested path."}}
+{"type":"item.started","item":{"id":"c0","type":"command_execution","command":"ls $CODEX_HOME/generated_images","status":"in_progress"}}
+{"type":"item.completed","item":{"id":"c0","type":"command_execution","command":"ls $CODEX_HOME/generated_images","exit_code":1,"status":"failed"}}
+{"type":"item.started","item":{"id":"c1","type":"command_execution","command":"cp $CODEX_HOME/generated_images/019edc1c-e7a3-74f0-a276-13bea71d32d6/ig_03eda661.png /Users/x/out/maple.png","status":"in_progress"}}
+{"type":"item.completed","item":{"id":"c1","type":"command_execution","command":"cp $CODEX_HOME/generated_images/019edc1c-e7a3-74f0-a276-13bea71d32d6/ig_03eda661.png /Users/x/out/maple.png","exit_code":0,"status":"completed"}}
+{"type":"item.completed","item":{"id":"m1","type":"agent_message","text":"{\\"status\\":\\"ok\\",\\"path\\":\\"/Users/x/out/maple.png\\",\\"bytes\\":1317377}"}}
+{"type":"turn.completed","usage":{"input_tokens":49489,"cached_input_tokens":34432,"output_tokens":2463,"reasoning_output_tokens":1990}}`;
+
+test("real success stream carries no image_gen item — gating on the stream alone would false-negative (#185)", () => {
+  const r = parseEventStream(REAL_SUCCESS_STREAM);
+  expect(r.threadId).toBe("019edc1c-e7a3-74f0-a276-13bea71d32d6");
+  // success path tools: reasoning / shell / agent_message — never image_gen
+  expect(r.toolCalls.map((tc) => tc.tool).sort()).toEqual([
+    "agent_message",
+    "agent_message",
+    "reasoning",
+    "reasoning",
+    "shell",
+    "shell",
+  ]);
+  expect(r.toolCalls.some((tc) => tc.tool === "image_gen")).toBe(false);
+  // filesystem evidence (PNG in this thread's dir) is what must let it through
+  expect(hasImageGenEvidence(r.toolCalls, true)).toBe(true);
+  // with no filesystem evidence and no stream item, it is rejected (the shortcut)
+  expect(hasImageGenEvidence(r.toolCalls, false)).toBe(false);
+});
+
+test("hasImageGenEvidence accepts a real image_gen stream item even without a dir PNG", () => {
+  expect(hasImageGenEvidence([{ id: "1", tool: "image_gen", status: "completed" }], false)).toBe(true);
+  expect(hasImageGenEvidence([{ id: "1", tool: "shell", status: "completed" }], false)).toBe(false);
+});
+
 test("verifyOutput passes for valid PNG", async () => {
   const dir = await mkdtemp(path.join(tmpdir(), "cig-val-"));
   try {
@@ -73,26 +114,3 @@ test("verifyImageGenWasInvoked true when PNG exists in thread dir", async () =>
     await rm(tempHome, { recursive: true, force: true });
   }
 });
-
-test("findCpToTarget detects cp from generated_images", () => {
-  expect(
-    findCpToTarget(
-      [
-        {
-          id: "1",
-          tool: "shell",
-          status: "completed",
-          command: "cp ~/.codex/generated_images/thread/ig_x.png /tmp/out.png",
-        },
-      ],
-      "/tmp/out.png",
-    ),
-  ).toBe(true);
-
-  expect(
-    findCpToTarget(
-      [{ id: "1", tool: "shell", status: "completed", command: "ls /tmp" }],
-      "/tmp/out.png",
-    ),
-  ).toBe(false);
-});

packages/baoyu-codex-imagegen/src/validator.ts

@@ -3,6 +3,7 @@ import { homedir } from "node:os";
 import path from "node:path";
 import { GenError } from "./types.ts";
 import type { ToolCall } from "./types.ts";
+import { hasImageGenInvocation } from "./parser.ts";
 
 const PNG_MAGIC = Buffer.from([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]);
 
@@ -23,15 +24,15 @@ export async function verifyImageGenWasInvoked(threadId: string | null): Promise
   }
 }
 
-export function findCpToTarget(toolCalls: ToolCall[], target: string): boolean {
-  return toolCalls.some(
-    (tc) =>
-      tc.tool === "shell" &&
-      typeof tc.command === "string" &&
-      (tc.command.includes(target) || tc.command.includes(path.basename(target))) &&
-      /\b(cp|mv|cat)\b/.test(tc.command) &&
-      tc.command.includes("generated_images"),
-  );
+// Real evidence that image_gen ran in THIS thread. Codex's image_gen tool does
+// not surface as a stream item, so a successful run shows only reasoning/shell/
+// agent_message — `dirHasImage` (a PNG in this thread's generated_images dir) is
+// what proves it. The stream check is kept as a forward-compatible signal in
+// case a future Codex version emits the item. The #185 shortcut (copying an
+// unrelated history image, which lives under a different thread id) yields
+// neither, so it is correctly rejected.
+export function hasImageGenEvidence(toolCalls: ToolCall[], dirHasImage: boolean): boolean {
+  return dirHasImage || hasImageGenInvocation(toolCalls);
 }
 
 export async function verifyOutput(outputPath: string): Promise<{ bytes: number }> {

packages/baoyu-fetch/package.json

@@ -1,6 +1,7 @@
 {
   "name": "baoyu-fetch",
   "version": "0.1.2",
+  "license": "MIT",
   "description": "Read URLs into high-quality Markdown or JSON with Chrome CDP and site adapters.",
   "type": "module",
   "bin": {

packages/baoyu-md/package.json

@@ -1,6 +1,7 @@
 {
   "name": "baoyu-md",
   "version": "0.1.1",
+  "license": "MIT",
   "type": "module",
   "main": "./dist/index.cjs",
   "module": "./dist/index.js",

skills/baoyu-article-illustrator/SKILL.md

@@ -30,6 +30,7 @@ When this skill needs to render an image, resolve the backend in this order:
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
    - **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-image-gen`) unless the user has explicitly pinned a different `preferred_image_backend`.
    - **Codex via `codex exec` (`codex-imagegen`)** — if the current runtime exposes no native `imagegen` skill but the `codex` CLI is on `PATH` with an active `codex login`, route through `baoyu-image-gen --provider codex-cli` (preferred), or — if baoyu-image-gen is unavailable — invoke the bundled wrapper directly. Details, parameters, and the runtime-discovery procedure live in [references/codex-imagegen.md](references/codex-imagegen.md) — load that file only when this branch is selected.
+   - **Cursor (`GenerateImage`)** — if the runtime exposes a native `GenerateImage` tool, you are running inside Cursor and it outranks any non-native skill the same way Codex `imagegen` does. Two hard caveats: (a) it has no aspect-ratio parameter — state the target aspect ratio / dimensions explicitly in the prompt text passed as `description`; (b) it does not accept an output directory — it saves to a tool-managed location, so after generation copy/move the file to the skill's expected output path (e.g., `outputs/.../NN-xxx.png`). Reference images go in `reference_image_paths`.
    - **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-image-gen`), use it.
    - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
@@ -43,7 +44,7 @@ Setting `preferred_image_backend: ask` forces the step-3 prompt every run regard
 
 **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.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
 
 ## Batch Generation Policy
 

skills/baoyu-comic/SKILL.md

@@ -34,6 +34,7 @@ When this skill needs to render an image, resolve the backend in this order:
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
    - **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-image-gen`) unless the user has explicitly pinned a different `preferred_image_backend`.
    - **Codex via `codex exec` (`codex-imagegen`)** — if the current runtime exposes no native `imagegen` skill but the `codex` CLI is on `PATH` with an active `codex login`, route through `baoyu-image-gen --provider codex-cli` (preferred), or — if baoyu-image-gen is unavailable — invoke the bundled wrapper directly. Details, parameters, and the runtime-discovery procedure live in [references/codex-imagegen.md](references/codex-imagegen.md) — load that file only when this branch is selected.
+   - **Cursor (`GenerateImage`)** — if the runtime exposes a native `GenerateImage` tool, you are running inside Cursor and it outranks any non-native skill the same way Codex `imagegen` does. Two hard caveats: (a) it has no aspect-ratio parameter — state the target aspect ratio / dimensions explicitly in the prompt text passed as `description`; (b) it does not accept an output directory — it saves to a tool-managed location, so after generation copy/move the file to the skill's expected output path (e.g., `outputs/.../NN-xxx.png`). Reference images go in `reference_image_paths`.
    - **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-image-gen`), use it.
    - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
@@ -47,7 +48,7 @@ Setting `preferred_image_backend: ask` forces the step-3 prompt every run regard
 
 **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.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
 
 ## Batch Generation Policy
 

skills/baoyu-cover-image/SKILL.md

@@ -30,6 +30,7 @@ When this skill needs to render an image, resolve the backend in this order:
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
    - **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-image-gen`) unless the user has explicitly pinned a different `preferred_image_backend`.
    - **Codex via `codex exec` (`codex-imagegen`)** — if the current runtime exposes no native `imagegen` skill but the `codex` CLI is on `PATH` with an active `codex login`, route through `baoyu-image-gen --provider codex-cli` (preferred), or — if baoyu-image-gen is unavailable — invoke the bundled wrapper directly. Details, parameters, and the runtime-discovery procedure live in [references/codex-imagegen.md](references/codex-imagegen.md) — load that file only when this branch is selected.
+   - **Cursor (`GenerateImage`)** — if the runtime exposes a native `GenerateImage` tool, you are running inside Cursor and it outranks any non-native skill the same way Codex `imagegen` does. Two hard caveats: (a) it has no aspect-ratio parameter — state the target aspect ratio / dimensions explicitly in the prompt text passed as `description`; (b) it does not accept an output directory — it saves to a tool-managed location, so after generation copy/move the file to the skill's expected output path (e.g., `outputs/.../NN-xxx.png`). Reference images go in `reference_image_paths`.
    - **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-image-gen`), use it.
    - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
@@ -43,7 +44,7 @@ Setting `preferred_image_backend: ask` forces the step-3 prompt every run regard
 
 **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.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
 
 ## Confirmation Policy
 

skills/baoyu-danger-gemini-web/scripts/gemini-webapi/client.ts

@@ -456,19 +456,19 @@ export class GeminiClient extends GemMixin {
           const generated_images: GeneratedImage[] = [];
           const wants_generated =
             get_nested_value(candidate, [12, 7, 0], null) != null ||
-            /http:\/\/googleusercontent\.com\/image_generation_content\/\d+/.test(text);
+            /http:\/\/googleusercontent\.com\/image_generation_content\/\d+/.test(text) ||
+            // Gemini 3 no longer emits the legacy marker in the candidate text; the generated
+            // image arrives in a later response part. Detect it from the raw response instead.
+            /\/gg-dl\//.test(txt) ||
+            /image_generation_content\/\d+/.test(txt);
 
           if (wants_generated) {
             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(
-                'Failed to parse generated images. Please update gemini_webapi to the latest version. If the error persists and is caused by the package, please report it on GitHub.',
-              );
-            }
-
-            const img_candidate = get_nested_value<unknown[]>(img_body, [4, candidate_index], []);
+            // Not every candidate carries a generated image (e.g. multi-candidate responses).
+            // If this candidate has no image part, skip extraction instead of failing the whole call.
+            const img_candidate = img_body ? get_nested_value<unknown[]>(img_body, [4, candidate_index], []) : [];
             const finished = get_nested_value<string | null>(img_candidate, [1, 0], null);
             if (finished) {
               text = finished.replace(/http:\/\/googleusercontent\.com\/image_generation_content\/\d+/g, '').trimEnd();

skills/baoyu-image-gen/scripts/codex-imagegen/main.ts

@@ -6,7 +6,7 @@ import process from "node:process";
 import { setTimeout as delay } from "node:timers/promises";
 import { GenError, type CliOptions, type GenerateResult } from "./types.ts";
 import { runCodexExec } from "./spawn.ts";
-import { findCpToTarget, verifyImageGenWasInvoked, verifyOutput } from "./validator.ts";
+import { hasImageGenEvidence, verifyImageGenWasInvoked, verifyOutput } from "./validator.ts";
 import { cacheKey, lookupCache, storeCache, FileLock } from "./cache.ts";
 import { JsonLogger } from "./logger.ts";
 
@@ -125,7 +125,7 @@ function buildInstruction(prompt: string, opts: CliOptions): string {
   const refHint = opts.refImages.length > 0
     ? `\nREFERENCE IMAGES (attached above): ${opts.refImages.length} image(s) provided for style/composition guidance.\n`
     : "";
-  return `You have an internal tool called image_gen for image generation. Use it.
+  return `You have an internal tool called image_gen for image generation. You MUST call it before doing anything else.
 
 TASK: Generate an image with the spec below, then save to disk.
 
@@ -137,12 +137,15 @@ OUTPUT PATH: ${opts.outputPath}
 ${refHint}
 STEPS:
 1. Call image_gen with the prompt and aspect ratio above${opts.refImages.length > 0 ? " (using the attached reference images for guidance)" : ""}.
-2. Move or copy the resulting image from Codex default location ($CODEX_HOME/generated_images/...) to: ${opts.outputPath}
+2. Move or copy ONLY the image produced by that image_gen call from Codex default location ($CODEX_HOME/generated_images/...) to: ${opts.outputPath}
 3. Verify with: ls -la ${opts.outputPath}
 4. Reply with ONLY this JSON line (no markdown fences, no other text):
    {"status":"ok","path":"${opts.outputPath}","bytes":<file_size_in_bytes>}
 
 HARD CONSTRAINTS:
+- Do NOT search for, find, inspect, reuse, or copy any pre-existing files from $CODEX_HOME/generated_images/ or any other directory.
+- Do NOT run ls/find/rg/grep/glob over $CODEX_HOME/generated_images/ before image_gen has been called.
+- You MUST call image_gen first. Only after image_gen completes may you copy the newly created file from this turn.
 - Do NOT use curl, wget, Python, or any external API.
 - Do NOT use bash to fabricate an image; only image_gen produces real pixels.
 - Use ONLY the image_gen internal tool.`;
@@ -175,13 +178,16 @@ async function attemptGenerate(
     throw new GenError("agent_refused", "No thread id in event stream");
   }
 
-  // verify: image_gen was actually invoked (check $CODEX_HOME/generated_images/{threadId}/)
+  // verify image_gen ran in THIS thread. A PNG in this thread's
+  // generated_images dir is the real signal (image_gen does not surface as a
+  // stream item); the stream check is a forward-compatible fallback. The #185
+  // shortcut (copying an unrelated history image) yields neither.
   const ver = await verifyImageGenWasInvoked(run.threadId);
-  if (!ver.ok) {
-    // secondary verify: did tool_calls include cp/mv from generated_images to our target
-    if (!findCpToTarget(run.toolCalls, opts.outputPath)) {
-      throw new GenError("no_image_gen_tool_use", `image_gen was not invoked: ${ver.reason}`);
-    }
+  if (!hasImageGenEvidence(run.toolCalls, ver.ok)) {
+    throw new GenError(
+      "no_image_gen_tool_use",
+      `image_gen was not invoked (no image_gen event in stream; ${ver.reason})`,
+    );
   }
 
   // verify output

skills/baoyu-image-gen/scripts/codex-imagegen/validator.ts

@@ -3,6 +3,7 @@ import { homedir } from "node:os";
 import path from "node:path";
 import { GenError } from "./types.ts";
 import type { ToolCall } from "./types.ts";
+import { hasImageGenInvocation } from "./parser.ts";
 
 const PNG_MAGIC = Buffer.from([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]);
 
@@ -23,15 +24,15 @@ export async function verifyImageGenWasInvoked(threadId: string | null): Promise
   }
 }
 
-export function findCpToTarget(toolCalls: ToolCall[], target: string): boolean {
-  return toolCalls.some(
-    (tc) =>
-      tc.tool === "shell" &&
-      typeof tc.command === "string" &&
-      (tc.command.includes(target) || tc.command.includes(path.basename(target))) &&
-      /\b(cp|mv|cat)\b/.test(tc.command) &&
-      tc.command.includes("generated_images"),
-  );
+// Real evidence that image_gen ran in THIS thread. Codex's image_gen tool does
+// not surface as a stream item, so a successful run shows only reasoning/shell/
+// agent_message — `dirHasImage` (a PNG in this thread's generated_images dir) is
+// what proves it. The stream check is kept as a forward-compatible signal in
+// case a future Codex version emits the item. The #185 shortcut (copying an
+// unrelated history image, which lives under a different thread id) yields
+// neither, so it is correctly rejected.
+export function hasImageGenEvidence(toolCalls: ToolCall[], dirHasImage: boolean): boolean {
+  return dirHasImage || hasImageGenInvocation(toolCalls);
 }
 
 export async function verifyOutput(outputPath: string): Promise<{ bytes: number }> {

skills/baoyu-infographic/SKILL.md

@@ -30,6 +30,7 @@ When this skill needs to render an image, resolve the backend in this order:
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
    - **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-image-gen`) unless the user has explicitly pinned a different `preferred_image_backend`.
    - **Codex via `codex exec` (`codex-imagegen`)** — if the current runtime exposes no native `imagegen` skill but the `codex` CLI is on `PATH` with an active `codex login`, route through `baoyu-image-gen --provider codex-cli` (preferred), or — if baoyu-image-gen is unavailable — invoke the bundled wrapper directly. Details, parameters, and the runtime-discovery procedure live in [references/codex-imagegen.md](references/codex-imagegen.md) — load that file only when this branch is selected.
+   - **Cursor (`GenerateImage`)** — if the runtime exposes a native `GenerateImage` tool, you are running inside Cursor and it outranks any non-native skill the same way Codex `imagegen` does. Two hard caveats: (a) it has no aspect-ratio parameter — state the target aspect ratio / dimensions explicitly in the prompt text passed as `description`; (b) it does not accept an output directory — it saves to a tool-managed location, so after generation copy/move the file to the skill's expected output path (e.g., `outputs/.../NN-xxx.png`). Reference images go in `reference_image_paths`.
    - **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-image-gen`), use it.
    - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
@@ -43,7 +44,7 @@ Setting `preferred_image_backend: ask` forces the step-3 prompt every run regard
 
 **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.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
 
 ## Reference Images
 

skills/baoyu-post-to-wechat/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: baoyu-post-to-wechat
 description: Posts content to WeChat Official Account (微信公众号) via API or Chrome CDP. Supports article posting (文章) with HTML, markdown, or plain text input, and image-text posting (贴图, formerly 图文) with multiple images. Markdown article workflows default to converting ordinary external links into bottom citations for WeChat-friendly output. Use when user mentions "发布公众号", "post to wechat", "微信公众号", or "贴图/图文/文章".
-version: 1.118.1
+version: 1.118.2
 metadata:
   openclaw:
     homepage: https://github.com/JimLiu/baoyu-skills#baoyu-post-to-wechat

skills/baoyu-post-to-x/scripts/md-to-html.test.ts

@@ -98,3 +98,68 @@ test('parseMarkdown resolves encoded spaces and literal percent image paths', as
   assert.equal(result.contentImages[0]?.localPath, path.join(root, 'Pasted image.png'));
   assert.equal(result.contentImages[1]?.localPath, path.join(root, '100%.png'));
 });
+
+test('parseMarkdown renders CJK-adjacent bold and italics (no literal asterisks)', async (t) => {
+  const root = await makeTempDir('x-md-to-html-cjk-bold-');
+  t.after(() => fs.rm(root, { recursive: true, force: true }));
+
+  const markdownPath = path.join(root, 'post.md');
+  const tempDir = path.join(root, 'tmp');
+  await fs.mkdir(tempDir, { recursive: true });
+  await fs.writeFile(
+    markdownPath,
+    [
+      '# 标题',
+      '',
+      '分工在变细。**国际大厂卷基础设施，中文项目卷场景落地。**这其实是生态成熟的表现。',
+      '',
+      '半角场景 **Top 10 里平均有 8 个** 项目。',
+      '',
+      '斜体 *数据来源 GitHub* 收尾。',
+      '',
+      '参考 **[docs][d]** 了解更多。',
+      '',
+      '[d]: https://example.com',
+    ].join('\n'),
+  );
+
+  const result = await parseMarkdown(markdownPath, { tempDir });
+
+  // Bold directly adjacent to CJK (closing ** followed by CJK) must render.
+  assert.match(result.html, /<strong>国际大厂卷基础设施，中文项目卷场景落地。<\/strong>/);
+  assert.match(result.html, /<strong>Top 10 里平均有 8 个<\/strong>/);
+  // Italics.
+  assert.match(result.html, /<em>数据来源 GitHub<\/em>/);
+  // Reference-style links inside emphasis must render as links, not plain text.
+  assert.match(result.html, /<strong><a href="https:\/\/example\.com" rel="noopener noreferrer nofollow">docs<\/a><\/strong>/);
+  // No literal emphasis delimiters should leak into the output.
+  assert.doesNotMatch(result.html, /\*\*/);
+  assert.doesNotMatch(result.html, /(?<!\*)\*(?!\*)[^*\n]+\*(?!\*)/);
+});
+
+test('parseMarkdown does not decode author-written literal HTML entities into tags', async (t) => {
+  const root = await makeTempDir('x-md-to-html-entities-');
+  t.after(() => fs.rm(root, { recursive: true, force: true }));
+
+  const markdownPath = path.join(root, 'post.md');
+  const tempDir = path.join(root, 'tmp');
+  await fs.mkdir(tempDir, { recursive: true });
+  await fs.writeFile(
+    markdownPath,
+    [
+      '# 标题',
+      '',
+      '正文中写 &#x3C;b&#x3E;literal&#x3C;/b&#x3E; 想显示字面标签。**加粗**收尾。',
+      '',
+      '代码里写 `&#x3C;b&#x3E;` 同样保留。',
+    ].join('\n'),
+  );
+
+  const result = await parseMarkdown(markdownPath, { tempDir });
+
+  // CJK-adjacent bold still renders.
+  assert.match(result.html, /<strong>加粗<\/strong>/);
+  // Author-written literal entities must NOT be decoded into real tags.
+  assert.doesNotMatch(result.html, /<b>literal<\/b>/);
+  assert.match(result.html, /&lt;b&gt;literal&lt;\/b&gt;/);
+});

skills/baoyu-post-to-x/scripts/md-to-html.ts

@@ -136,6 +136,21 @@ function highlightCode(code: string, lang: string): string {
   }
 }
 
+// Normalize CJK-adjacent emphasis so `marked` renders it correctly.
+//
+// `marked`'s emphasis tokenizer treats a closing `**`/`*` directly followed by a
+// CJK character as not right-flanking, so it leaves the delimiters literal
+// (e.g. `**加粗**这` renders as plain text with the asterisks intact). We round-trip
+// the markdown through `remark-cjk-friendly`, whose stringify serializes the
+// boundary character as an HTML entity (`这`); the entity is treated as
+// punctuation by `marked`'s flanking rules, so emphasis parses as expected.
+//
+// We deliberately do NOT decode the entities afterward. They are valid HTML
+// character references that render correctly when the article HTML is pasted into
+// the X editor, and `marked` only emits them for characters outside the emphasis
+// span (the boundary char), never inside it. A blanket decode of the rendered
+// HTML would risk turning author-written literal entities (e.g. `<b>`
+// meant to display `<b>` as text) into real tags, so we leave them intact.
 function preprocessCjkMarkdown(markdown: string): string {
   try {
     const processor = unified()
@@ -143,8 +158,7 @@ function preprocessCjkMarkdown(markdown: string): string {
       .use(remarkCjkFriendly)
       .use(remarkStringify);
 
-    const result = String(processor.processSync(markdown));
-    return result.replace(/&#x([0-9A-Fa-f]+);/g, (_, hex: string) => String.fromCodePoint(parseInt(hex, 16)));
+    return String(processor.processSync(markdown));
   } catch {
     return markdown;
   }

skills/baoyu-slide-deck/SKILL.md

@@ -34,6 +34,7 @@ When this skill needs to render an image, resolve the backend in this order:
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
    - **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-image-gen`) unless the user has explicitly pinned a different `preferred_image_backend`.
    - **Codex via `codex exec` (`codex-imagegen`)** — if the current runtime exposes no native `imagegen` skill but the `codex` CLI is on `PATH` with an active `codex login`, route through `baoyu-image-gen --provider codex-cli` (preferred), or — if baoyu-image-gen is unavailable — invoke the bundled wrapper directly. Details, parameters, and the runtime-discovery procedure live in [references/codex-imagegen.md](references/codex-imagegen.md) — load that file only when this branch is selected.
+   - **Cursor (`GenerateImage`)** — if the runtime exposes a native `GenerateImage` tool, you are running inside Cursor and it outranks any non-native skill the same way Codex `imagegen` does. Two hard caveats: (a) it has no aspect-ratio parameter — state the target aspect ratio / dimensions explicitly in the prompt text passed as `description`; (b) it does not accept an output directory — it saves to a tool-managed location, so after generation copy/move the file to the skill's expected output path (e.g., `outputs/.../NN-xxx.png`). Reference images go in `reference_image_paths`.
    - **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-image-gen`), use it.
    - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
@@ -47,7 +48,7 @@ Setting `preferred_image_backend: ask` forces the step-3 prompt every run regard
 
 **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.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
 
 ## Batch Generation Policy
 

skills/baoyu-translate/SKILL.md

@@ -6,7 +6,7 @@ description: >-
   "localize", "本地化", "refined translation", "精细翻译", "proofread translation", "快速翻译", "快翻",
   "这篇文章翻译一下", or provides a URL/file with translation intent. Supports three modes
   (quick/normal/refined) with custom glossary support.
-version: 1.59.0
+version: 1.117.3
 metadata:
   openclaw:
     homepage: https://github.com/JimLiu/baoyu-skills#baoyu-translate

skills/baoyu-wechat-summary/SKILL.md

@@ -1,7 +1,7 @@
 ---
 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), per-user profiles, and per-group fact memory (memory.md) 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: 1.117.3
+version: 1.118.0
 metadata:
   openclaw:
     homepage: https://github.com/JimLiu/baoyu-skills#baoyu-wechat-summary
@@ -146,15 +146,29 @@ where `data_root` is from EXTEND.md (default `{project_root}/wechat`).
 
 **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 2.5: Look up the group owner（群主）
+
+群主是谁**必须有据可查**，不能凭历史摘要、群友玩笑或印象推断（群主可能换届，历史摘要里的说法会过期）：
+
+```bash
+wx members "<group_name_or_id>" --json
+```
+
+- 检查输出中是否有 owner / role 字段标识群主；有则以此为准
+- 如果 wx-cli 版本不暴露群主信息，则查 memory.md「群基本档案」里有出处的记录；两处都没有 → **摘要里不要断言谁是群主**
+- 查到的结果与「群基本档案」不一致时以本次查询为准，更新档案并追加修订记录（注明查询日期）
+
 ### Step 3: Fetch messages
 
-For small batches (single-day digest, typically < 200 messages), pipe JSON into the agent directly:
+**Always redirect the fetch to a `$TMPDIR` file** — this file is the single source of truth for the whole run: Round 3's attribution audit greps it, and the statistics are computed from it. Never write the digest purely from conversation memory.
+
+For small batches (single-day digest, typically < 200 messages), you may additionally pipe JSON into the agent directly for reading:
 
 ```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 `$TMPDIR` first so the raw payload never sits in conversation context:
+For **large batches** (weekly / monthly digests, > 200 messages), the `$TMPDIR` redirect also keeps the raw payload out of conversation context:
 
 ```bash
 wx history "<group_name_or_id>" --since YYYY-MM-DD --until YYYY-MM-DD -n 5000 --json > "$TMPDIR/wx-messages.json"
@@ -171,7 +185,7 @@ Notes:
 - 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.
+**Incremental mode**: after the fetch, drop any message whose `timestamp` is `<=` the `last_message_time` from `history.json`, and write the filtered set back to the `$TMPDIR` file (so audits and stats run on exactly what the digest covers). Caution: `last_message_time` is `MM-DD HH:MM` — plain string comparison breaks across a year boundary (12-31 vs 01-01); compare by date semantics there. 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
 
@@ -195,6 +209,7 @@ Notes:
 
 - 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.
+- **硬规则**：`nil`、空白、单标点这类占位符样式的名字**绝不允许原样出现在摘要里**。contacts 查不到 remark 时，用「昵称（wxid 后 4 位）」形式区分（如 `nil（…n77g）`），确保读者知道这是谁、且与其他人不混淆。已解析过的映射写入 memory.md「群基本档案」，下期直接复用不再重查。
 
 ### Step 3.7: Load user profiles
 
@@ -224,7 +239,7 @@ See [references/profiles.md](references/profiles.md) for the full file format.
 
 除了按人的 profiles，每个群还有一份全局事实记忆 `{folder}/memory.md`，记录群友指正过、确认过的客观事实（如"某个报错提示的真实原因"、"某产品名的正确写法"、"某事件的实际经过"）。
 
-1. 如果 `memory.md` 存在，读入作为内部背景知识（不写入最终摘要）
+1. 如果 `memory.md` 存在，读入作为内部背景知识（不写入最终摘要）。「群基本档案」小节记录群主、昵称映射等长期事实，写摘要时直接引用（群主以 Step 2.5 的查证结果为最终依据）
 2. **写摘要时必须遵守其中的事实修正**——上一期摘要里说错、已被群友指正的说法，这一期绝不能再犯。例如记忆中有"『当前微信版本不支持』是 AI Agent 无法获取微信链接导致的提示，普通用户可正常打开"，就不能再把它当成"骗点击"的梗来写
 3. 记忆条目是事实约束，不是风格指令——它只纠正"说什么"，不改变 normal/roast 两个版本各自的语气和写法
 4. 标注为「群友说法（未验证）」的条目，引用时保留这个限定，不当成已证实的事实陈述
@@ -280,7 +295,7 @@ 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）
+1. [HH:MM-HH:MM] 话题名称（参与者：A, B, C）— 一句话概括（锚点：54052 宝玉:"原话片段" → 54063 鸭哥:"回应片段"）
 2. [HH:MM-HH:MM] 话题名称（参与者：D, E）— 一句话概括（锚点 id：54100-54112）
 ...
 
@@ -300,7 +315,7 @@ 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.
+- **Carry anchor IDs with verbatim quotes**: for key messages, record `id 发言人:"原话片段"` — sender and quote fragment **copied verbatim from the raw messages**, not paraphrased. In Round 2, jump back to these anchors 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. Pinning "who said what" at the skeleton stage is the first line of defense against misattribution (张冠李戴).
 
 **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).
 
@@ -340,6 +355,17 @@ Walk the Round 1 skeleton against the finished digest. Check:
 - Quotes, names, product/tool names preserved verbatim?
 - Categorization makes sense — is anything in the wrong bucket?
 
+**Attribution audit (mandatory — never skip)**: for every direct quote (text in quotation marks) and every "X 说 / X 发 / X 分享" attribution in the draft, grep the raw `$TMPDIR` messages file and confirm the words actually came from that sender:
+
+```bash
+grep "原话片段" "$TMPDIR/wx-messages.json"   # or jq 'map(select(.content | contains("原话片段")))'
+```
+
+- Quote not found in the file → paraphrase drift or invented memory; restore the original wording or cut it
+- Quote found but sender doesn't match → misattribution; fix the name
+- Audit BOTH versions (normal + roast) if both were generated
+- Record a one-line verdict in working notes: `归因校验：共 N 处引用，通过 X 处，修正 Y 处`
+
 Fix in place. When clean, confirm and proceed.
 
 ### Step 7: Save the digest file(s)
@@ -409,6 +435,18 @@ Counts, frontmatter updates, append-only rules for quotes and events, and privac
 
 更新画像后，扫描本期消息，看是否有需要写入/修订 `{folder}/memory.md` 的事实修正。这一步要**保守**：宁可漏记，不可乱记。
 
+**这一步必须执行、必须留痕，不允许静默跳过。** 按以下流程扫描：
+
+1. 关键词初筛（对 `$TMPDIR` 消息文件跑一遍，圈出候选消息）：
+```bash
+grep -nE "错了|不对|纠正|搞错|其实是|不是.*是|瞎说|胡说|张冠李戴|谁是群主|群主是" "$TMPDIR/wx-messages.json"
+```
+2. 补充人工检查两类高概率位置：
+   - 所有**回复摘要消息**的 reply（Step 3.8 检测到的 in-chat digest，指向它的引用都是候选）
+   - @bot 请求里带指正性质的（Step 3.9 清单）
+3. 逐条按下方门槛判断是否写入
+4. 无论写入几条，最终报告里必须有一行结论：`memory 扫描：候选 N 条 → 写入 M 条`（0 也要写）——强制留痕是为了防止这一步被习惯性跳过
+
 #### 什么算"值得记的事实修正"
 
 典型场景：上一期摘要里有个说法（梗、归因、解释），群友在本期指出它不对，并给出了正确解释。例如摘要把"当前微信版本不支持"写成骗点击的链接，群友指正这其实是 AI Agent 无法获取微信链接时才出现的提示，普通人能正常打开——这就该记。
@@ -448,6 +486,11 @@ Counts, frontmatter updates, append-only rules for quotes and events, and privac
 ```markdown
 # 群级事实记忆 — {群名}
 
+## 群基本档案
+- 群主：{昵称}（{wxid}，查证于 YYYY-MM-DD，来源 wx members / 群友确认）
+- 昵称映射：{占位符昵称} = {remark/真名}（{wxid}）
+- {其他长期有效的群级事实：bot 的称呼、群名由来等}
+
 ## 事实修正
 - "当前微信版本不支持" 是 AI Agent/机器人无法获取微信链接时的提示，普通用户可正常打开，不是骗点击的链接。（指正：消失的大叔，2026-06-12，id 54321；另有 2 人附和）
 
@@ -470,7 +513,8 @@ Profile updates are easy to forget once the digest is on disk. Before reporting
 - [ ] `{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`)
-- [ ] `{folder}/memory.md` checked against this batch's corrections — updated if any passed the Step 8.6 threshold, untouched otherwise
+- [ ] `{folder}/memory.md` checked against this batch's corrections — updated if any passed the Step 8.6 threshold, untouched otherwise; the final report includes the `memory 扫描：候选 N 条 → 写入 M 条` verdict line
+- [ ] Round 3 attribution audit ran, with its `归因校验：…` verdict line in working notes
 
 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.
 

skills/baoyu-wechat-summary/references/output-formats.md

@@ -43,6 +43,8 @@ Example:
   - 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.
+  - **Counts must be computed mechanically** from the `$TMPDIR` messages file (e.g. `jq 'group_by(.from_wxid) | map({name: .[0].from_nickname, n: length}) | sort_by(-.n)'`) — never estimated by eyeballing. Total and per-person counts both.
+- **Incremental runs must show the precise coverage window**: day-granular date ranges share their boundary day with the previous digest, which readers misread as overlap. Add a line right after the message count: `⏱ 覆盖区间: MM-DD HH:MM ~ MM-DD HH:MM` (first and last included message timestamps).
 
 Example:
 

skills/baoyu-xhs-images/SKILL.md

@@ -30,6 +30,7 @@ When this skill needs to render an image, resolve the backend in this order:
 3. **Auto-select** (when the preference is `auto`, unset, or the pinned backend isn't available):
    - **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-image-gen`) unless the user has explicitly pinned a different `preferred_image_backend`.
    - **Codex via `codex exec` (`codex-imagegen`)** — if the current runtime exposes no native `imagegen` skill but the `codex` CLI is on `PATH` with an active `codex login`, route through `baoyu-image-gen --provider codex-cli` (preferred), or — if baoyu-image-gen is unavailable — invoke the bundled wrapper directly. Details, parameters, and the runtime-discovery procedure live in [references/codex-imagegen.md](references/codex-imagegen.md) — load that file only when this branch is selected.
+   - **Cursor (`GenerateImage`)** — if the runtime exposes a native `GenerateImage` tool, you are running inside Cursor and it outranks any non-native skill the same way Codex `imagegen` does. Two hard caveats: (a) it has no aspect-ratio parameter — state the target aspect ratio / dimensions explicitly in the prompt text passed as `description`; (b) it does not accept an output directory — it saves to a tool-managed location, so after generation copy/move the file to the skill's expected output path (e.g., `outputs/.../NN-xxx.png`). Reference images go in `reference_image_paths`.
    - **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-image-gen`), use it.
    - Otherwise (multiple non-native backends with no runtime-native tool), ask the user once — batch with any other initial questions.
@@ -43,7 +44,7 @@ Setting `preferred_image_backend: ask` forces the step-3 prompt every run regard
 
 **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.
 
-Concrete tool names (`imagegen`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
+Concrete tool names (`imagegen`, `GenerateImage`, `image_generate`, `baoyu-image-gen`) above are examples — substitute the local equivalents under the same rule.
 
 ## Batch Generation Policy