chore: re-trigger release for v0.6.0

fix(ci): fix shell script error when commit messages contain backticks
feat: add openwebui_id to the GitHub Copilot SDK files filter plugin metadata.
2026-02-09 19:54:49 +08:00 · 2026-02-09 19:53:23 +08:00 · 2026-02-09 19:48:41 +08:00 · 2026-02-09 19:45:23 +08:00 · 2026-02-09 03:10:29 +00:00 · 2026-02-09 00:44:14 +00:00
69 changed files with 8657 additions and 3015 deletions
--- a/.agent/workflows/plugin-development.md
+++ b/.agent/workflows/plugin-development.md
@@ -11,6 +11,7 @@ This workflow outlines the standard process for developing, documenting, and rel
 Reference: `.github/copilot-instructions.md`

 ### Bilingual Requirement
+
 Every plugin **MUST** have bilingual versions for both code and documentation:

 - **Code**:
@@ -21,6 +22,7 @@ Every plugin **MUST** have bilingual versions for both code and documentation:
  - Chinese: `plugins/{type}/{name}/README_CN.md`

 ### Code Structure
+
 - **Docstring**: Must include `title`, `author`, `version`, `description`, etc.
 - **Valves**: Use `pydantic` for configuration.
 - **Database**: Re-use `open_webui.internal.db` shared connection.
@@ -32,8 +34,9 @@ Every plugin **MUST** have bilingual versions for both code and documentation:
  - Use Chat Persistence API for database storage
  - Always update both `messages[]` and `history.messages`

-### Commit Messages
- **Language**: **English ONLY**. Do not use Chinese in commit messages.
+### Commit Messages & Release Notes
+
+- **Language**: **English ONLY**. Do not use Chinese in commit messages or release notes.
 - **Format**: Conventional Commits (e.g., `feat:`, `fix:`, `docs:`).

 ## 2. Documentation Updates
@@ -41,10 +44,16 @@ Every plugin **MUST** have bilingual versions for both code and documentation:
 When adding or updating a plugin, you **MUST** update the following documentation files to maintain consistency:

 ### Plugin Directory
- `README.md`: Update version, description, and usage. **Explicitly describe new features in a prominent position at the beginning.**
- `README_CN.md`: Update version, description, and usage. **Explicitly describe new features in a prominent position at the beginning.**
+
+- `README.md`: Update version, description, and usage.
+  - **Key Capabilities**: **MUST** include ALL core functionalities and features. Do not only list new features in "What's New".
+  - **What's New**: Explicitly describe only the latest changes/updates in a prominent position at the beginning. This section is dynamic and changes with versions.
+- `README_CN.md`: Update version, description, and usage.
+  - **核心功能 (Key Capabilities)**: **必须**包含所有核心功能和特性，不能只在 "What's New" 中列出。
+  - **最新更新 (What's New)**: 在开头显眼位置明确描述最新的更改/更新。此部分是动态的，随版本变化。

 ### Global Documentation (`docs/`)
+
 - **Index Pages**:
  - `docs/plugins/{type}/index.md`: Add/Update list item with **correct version**.
  - `docs/plugins/{type}/index.zh.md`: Add/Update list item with **correct version**.
@@ -53,6 +62,7 @@ When adding or updating a plugin, you **MUST** update the following documentatio
  - `docs/plugins/{type}/{name}.zh.md`: Ensure content matches README_CN.

 ### Root README
+
 - `README.md`: Add to "Featured Plugins" if applicable.
 - `README_CN.md`: Add to "Featured Plugins" if applicable.

@@ -61,6 +71,7 @@ When adding or updating a plugin, you **MUST** update the following documentatio
 Reference: `.github/workflows/release.yml`

 ### Version Bumping
+
 - **Rule**: Version bump is required **ONLY when the user explicitly requests a release**. Regular code changes do NOT require version bumps.
 - **Format**: Semantic Versioning (e.g., `1.0.0` -> `1.0.1`).
 - **When to Bump**: Only update the version when:
@@ -78,13 +89,14 @@ Reference: `.github/workflows/release.yml`
  8. Docs Detail CN (`docs/.../{name}.zh.md`)

 ### Automated Release Process
-1.  **Trigger**: Push to `main` branch with changes in `plugins/**/*.py`.
-2.  **Detection**: `scripts/extract_plugin_versions.py` detects changed plugins and compares versions.
-3.  **Release**:
+
+1. **Trigger**: Push to `main` branch with changes in `plugins/**/*.py`.
+2. **Detection**: `scripts/extract_plugin_versions.py` detects changed plugins and compares versions.
+3. **Release**:
    - Generates release notes based on changes.
    - Creates a GitHub Release tag (e.g., `v2024.01.01-1`).
    - Uploads individual `.py` files of **changed plugins only** as assets.
-4.  **Market Publishing**:
+4. **Market Publishing**:
    - Workflow: `.github/workflows/publish_plugin.yml`
    - Trigger: Release published.
    - Action: Automatically updates the plugin code and metadata on OpenWebUI.com using `scripts/publish_plugin.py`.
@@ -95,6 +107,7 @@ Reference: `.github/workflows/release.yml`
      - Example: `https://github.com/Fu-Jie/awesome-openwebui/blob/main/plugins/filters/folder-memory/README.md`

 ### Pull Request Check
+
 - Workflow: `.github/workflows/plugin-version-check.yml`
 - Checks if plugin files are modified.
 - **Fails** if version number is not updated.
@@ -114,5 +127,3 @@ Before committing:
 ## 5. Git Operations (Agent Rules)

 Strictly follow the rules defined in `.github/copilot-instructions.md` → **Git Operations (Agent Rules)** section.
-
-
--- a/.github/copilot-instructions.md
+++ b/.github/copilot-instructions.md
@@ -889,6 +889,9 @@ Filter 实例是**单例 (Singleton)**。
 - [ ] 使用 logging 而非 print
 - [ ] 测试双语界面
 - [ ] **一致性检查**: 确保文档、代码、README 同步
+- [ ] **README 结构**: 
+    - **Key Capabilities** (英文) / **核心功能** (中文): 必须包含所有核心功能
+    - **What's New** (英文) / **最新更新** (中文): 仅包含最新版本的变更信息

 ### 2. 🔄 一致性维护 (Consistency Maintenance)

--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -324,13 +324,13 @@ jobs:

      - name: Generate release notes
        id: notes
+        env:
+          VERSION: ${{ steps.version.outputs.version }}
+          TITLE: ${{ github.event.inputs.release_title }}
+          NOTES: ${{ github.event.inputs.release_notes }}
+          DETECTED_CHANGES: ${{ needs.check-changes.outputs.release_notes }}
+          COMMITS: ${{ steps.commits.outputs.commits }}
        run: |
-          VERSION="${{ steps.version.outputs.version }}"
-          TITLE="${{ github.event.inputs.release_title }}"
-          NOTES="${{ github.event.inputs.release_notes }}"
-          DETECTED_CHANGES="${{ needs.check-changes.outputs.release_notes }}"
-          COMMITS="${{ steps.commits.outputs.commits }}"
-          
          echo "# ${VERSION} Release / 发布" > release_notes.md
          echo "" >> release_notes.md
          
--- a/README.md
+++ b/README.md
@@ -10,28 +10,28 @@ A collection of enhancements, plugins, and prompts for [OpenWebUI](https://githu
 <!-- STATS_START -->
 ## 📊 Community Stats

-> 🕐 Auto-updated: 2026-02-07 09:38
+> 🕐 Auto-updated: 2026-02-09 11:10

 | 👤 Author | 👥 Followers | ⭐ Points | 🏆 Contributions |
 |:---:|:---:|:---:|:---:|
-| [Fu-Jie](https://openwebui.com/u/Fu-Jie) | **203** | **246** | **40** |
+| [Fu-Jie](https://openwebui.com/u/Fu-Jie) | **210** | **259** | **43** |

 | 📝 Posts | ⬇️ Downloads | 👁️ Views | 👍 Upvotes | 💾 Saves |
 |:---:|:---:|:---:|:---:|:---:|
-| **20** | **3675** | **43326** | **213** | **253** |
+| **20** | **3909** | **45507** | **221** | **264** |

 ### 🔥 Top 6 Popular Plugins

-> 🕐 Auto-updated: 2026-02-07 09:38
+> 🕐 Auto-updated: 2026-02-09 11:10

 | Rank | Plugin | Version | Downloads | Views | Updated |
 |:---:|------|:---:|:---:|:---:|:---:|
-| 🥇 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | 0.9.2 | 877 | 7765 | 2026-01-28 |
-| 🥈 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | 1.5.0 | 631 | 5825 | 2026-01-30 |
-| 🥉 | [Export to Word (Enhanced)](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | 0.4.3 | 343 | 2659 | 2026-01-28 |
-| 4️⃣ | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | 0.3.7 | 327 | 1519 | 2026-01-29 |
-| 5️⃣ | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | 1.2.2 | 327 | 3412 | 2026-01-28 |
-| 6️⃣ | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | 1.2.4 | 293 | 4243 | 2026-01-29 |
+| 🥇 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | 0.9.2 | 916 | 8070 | 2026-01-28 |
+| 🥈 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | 1.5.0 | 665 | 6205 | 2026-01-30 |
+| 🥉 | [Export to Word Enhanced](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | 0.4.4 | 365 | 2841 | 2026-02-07 |
+| 4️⃣ | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | 1.2.2 | 350 | 3580 | 2026-01-28 |
+| 5️⃣ | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | 0.3.7 | 334 | 1569 | 2026-01-29 |
+| 6️⃣ | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | 1.2.4 | 310 | 4433 | 2026-01-29 |

 *See full stats in [Community Stats Report](./docs/community-stats.md)*
 <!-- STATS_END -->
--- a/README_CN.md
+++ b/README_CN.md
@@ -7,28 +7,28 @@ OpenWebUI 增强功能集合。包含个人开发与收集的插件、提示词
 <!-- STATS_START -->
 ## 📊 社区统计

-> 🕐 自动更新于 2026-02-07 09:38
+> 🕐 自动更新于 2026-02-09 11:10

 | 👤 作者 | 👥 粉丝 | ⭐ 积分 | 🏆 贡献 |
 |:---:|:---:|:---:|:---:|
-| [Fu-Jie](https://openwebui.com/u/Fu-Jie) | **203** | **246** | **40** |
+| [Fu-Jie](https://openwebui.com/u/Fu-Jie) | **210** | **259** | **43** |

 | 📝 发布 | ⬇️ 下载 | 👁️ 浏览 | 👍 点赞 | 💾 收藏 |
 |:---:|:---:|:---:|:---:|:---:|
-| **20** | **3675** | **43326** | **213** | **253** |
+| **20** | **3909** | **45507** | **221** | **264** |

 ### 🔥 热门插件 Top 6

-> 🕐 自动更新于 2026-02-07 09:38
+> 🕐 自动更新于 2026-02-09 11:10

 | 排名 | 插件 | 版本 | 下载 | 浏览 | 更新日期 |
 |:---:|------|:---:|:---:|:---:|:---:|
-| 🥇 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | 0.9.2 | 877 | 7765 | 2026-01-28 |
-| 🥈 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | 1.5.0 | 631 | 5825 | 2026-01-30 |
-| 🥉 | [Export to Word (Enhanced)](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | 0.4.3 | 343 | 2659 | 2026-01-28 |
-| 4️⃣ | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | 0.3.7 | 327 | 1519 | 2026-01-29 |
-| 5️⃣ | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | 1.2.2 | 327 | 3412 | 2026-01-28 |
-| 6️⃣ | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | 1.2.4 | 293 | 4243 | 2026-01-29 |
+| 🥇 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | 0.9.2 | 916 | 8070 | 2026-01-28 |
+| 🥈 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | 1.5.0 | 665 | 6205 | 2026-01-30 |
+| 🥉 | [Export to Word Enhanced](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | 0.4.4 | 365 | 2841 | 2026-02-07 |
+| 4️⃣ | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | 1.2.2 | 350 | 3580 | 2026-01-28 |
+| 5️⃣ | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | 0.3.7 | 334 | 1569 | 2026-01-29 |
+| 6️⃣ | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | 1.2.4 | 310 | 4433 | 2026-01-29 |

 *完整统计请查看 [社区统计报告](./docs/community-stats.zh.md)*
 <!-- STATS_END -->
--- a/docs/badges/downloads.json
+++ b/docs/badges/downloads.json
@@ -1,7 +1,7 @@
 {
  "schemaVersion": 1,
  "label": "downloads",
-  "message": "3.7k",
+  "message": "3.9k",
  "color": "blue",
  "namedLogo": "openwebui"
 }
--- a/docs/badges/followers.json
+++ b/docs/badges/followers.json
@@ -1,6 +1,6 @@
 {
  "schemaVersion": 1,
  "label": "followers",
-  "message": "203",
+  "message": "210",
  "color": "blue"
 }
--- a/docs/badges/points.json
+++ b/docs/badges/points.json
@@ -1,6 +1,6 @@
 {
  "schemaVersion": 1,
  "label": "points",
-  "message": "246",
+  "message": "259",
  "color": "orange"
 }
--- a/docs/badges/upvotes.json
+++ b/docs/badges/upvotes.json
@@ -1,6 +1,6 @@
 {
  "schemaVersion": 1,
  "label": "upvotes",
-  "message": "213",
+  "message": "221",
  "color": "brightgreen"
 }
--- a/docs/community-stats.json
+++ b/docs/community-stats.json
@@ -1,13 +1,14 @@
 {
  "total_posts": 20,
-  "total_downloads": 3675,
-  "total_views": 43326,
-  "total_upvotes": 213,
+  "total_downloads": 3909,
+  "total_views": 45507,
+  "total_upvotes": 221,
  "total_downvotes": 2,
-  "total_saves": 253,
-  "total_comments": 46,
+  "total_saves": 264,
+  "total_comments": 51,
  "by_type": {
-    "action": 15,
+    "pipe": 1,
+    "action": 14,
    "unknown": 4,
    "filter": 1
  },
@@ -19,10 +20,10 @@
      "version": "0.9.2",
      "author": "Fu-Jie",
      "description": "Intelligently analyzes text content and generates interactive mind maps to help users structure and visualize knowledge.",
-      "downloads": 877,
-      "views": 7765,
-      "upvotes": 21,
-      "saves": 47,
+      "downloads": 916,
+      "views": 8070,
+      "upvotes": 22,
+      "saves": 50,
      "comments": 13,
      "created_at": "2025-12-30",
      "updated_at": "2026-01-28",
@@ -35,47 +36,31 @@
      "version": "1.5.0",
      "author": "Fu-Jie",
      "description": "AI-powered infographic generator based on AntV Infographic. Supports professional templates, auto-icon matching, and SVG/PNG downloads.",
-      "downloads": 631,
-      "views": 5825,
-      "upvotes": 23,
-      "saves": 33,
+      "downloads": 665,
+      "views": 6205,
+      "upvotes": 24,
+      "saves": 34,
      "comments": 10,
      "created_at": "2025-12-28",
      "updated_at": "2026-01-30",
      "url": "https://openwebui.com/posts/smart_infographic_ad6f0c7f"
    },
    {
-      "title": "Export to Word (Enhanced)",
+      "title": "Export to Word Enhanced",
      "slug": "export_to_word_enhanced_formatting_fca6a315",
      "type": "action",
-      "version": "0.4.3",
+      "version": "0.4.4",
      "author": "Fu-Jie",
      "description": "Export current conversation from Markdown to Word (.docx) with Mermaid diagrams rendered client-side (Mermaid.js, SVG+PNG), LaTeX math, real hyperlinks, improved tables, syntax highlighting, and blockquote support.",
-      "downloads": 343,
-      "views": 2659,
-      "upvotes": 12,
+      "downloads": 365,
+      "views": 2841,
+      "upvotes": 14,
      "saves": 26,
-      "comments": 2,
+      "comments": 3,
      "created_at": "2026-01-03",
-      "updated_at": "2026-01-28",
+      "updated_at": "2026-02-07",
      "url": "https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315"
    },
-    {
-      "title": "Export to Excel",
-      "slug": "export_mulit_table_to_excel_244b8f9d",
-      "type": "action",
-      "version": "0.3.7",
-      "author": "Fu-Jie",
-      "description": "Extracts tables from chat messages and exports them to Excel (.xlsx) files with smart formatting.",
-      "downloads": 327,
-      "views": 1519,
-      "upvotes": 7,
-      "saves": 6,
-      "comments": 0,
-      "created_at": "2025-05-30",
-      "updated_at": "2026-01-29",
-      "url": "https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d"
-    },
    {
      "title": "Async Context Compression",
      "slug": "async_context_compression_b1655bc8",
@@ -83,15 +68,31 @@
      "version": "1.2.2",
      "author": "Fu-Jie",
      "description": "Reduces token consumption in long conversations while maintaining coherence through intelligent summarization and message compression.",
-      "downloads": 327,
-      "views": 3412,
-      "upvotes": 13,
+      "downloads": 350,
+      "views": 3580,
+      "upvotes": 14,
      "saves": 33,
      "comments": 0,
      "created_at": "2025-11-08",
      "updated_at": "2026-01-28",
      "url": "https://openwebui.com/posts/async_context_compression_b1655bc8"
    },
+    {
+      "title": "Export to Excel",
+      "slug": "export_mulit_table_to_excel_244b8f9d",
+      "type": "action",
+      "version": "0.3.7",
+      "author": "Fu-Jie",
+      "description": "Extracts tables from chat messages and exports them to Excel (.xlsx) files with smart formatting.",
+      "downloads": 334,
+      "views": 1569,
+      "upvotes": 7,
+      "saves": 6,
+      "comments": 0,
+      "created_at": "2025-05-30",
+      "updated_at": "2026-01-29",
+      "url": "https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d"
+    },
    {
      "title": "Markdown Normalizer",
      "slug": "markdown_normalizer_baaa8732",
@@ -99,10 +100,10 @@
      "version": "1.2.4",
      "author": "Fu-Jie",
      "description": "A content normalizer filter that fixes common Markdown formatting issues in LLM outputs, such as broken code blocks, LaTeX formulas, and list formatting.",
-      "downloads": 293,
-      "views": 4243,
+      "downloads": 310,
+      "views": 4433,
      "upvotes": 17,
-      "saves": 27,
+      "saves": 28,
      "comments": 5,
      "created_at": "2026-01-12",
      "updated_at": "2026-01-29",
@@ -115,8 +116,8 @@
      "version": "0.2.4",
      "author": "Fu-Jie",
      "description": "Quickly generates beautiful flashcards from text, extracting key points and categories.",
-      "downloads": 213,
-      "views": 3239,
+      "downloads": 217,
+      "views": 3302,
      "upvotes": 13,
      "saves": 14,
      "comments": 2,
@@ -131,10 +132,10 @@
      "version": "",
      "author": "",
      "description": "",
-      "downloads": 141,
-      "views": 2180,
+      "downloads": 178,
+      "views": 2475,
      "upvotes": 8,
-      "saves": 3,
+      "saves": 4,
      "comments": 0,
      "created_at": "2026-01-28",
      "updated_at": "2026-01-28",
@@ -147,8 +148,8 @@
      "version": "1.0.0",
      "author": "Fu-Jie",
      "description": "A comprehensive thinking lens that dives deep into any content - from context to logic, insights, and action paths.",
-      "downloads": 132,
-      "views": 1155,
+      "downloads": 144,
+      "views": 1218,
      "upvotes": 6,
      "saves": 11,
      "comments": 0,
@@ -157,21 +158,37 @@
      "url": "https://openwebui.com/posts/deep_dive_c0b846e4"
    },
    {
-      "title": "导出为 Word (增强版)",
+      "title": "导出为Word增强版",
      "slug": "导出为_word_支持公式流程图表格和代码块_8a6306c0",
      "type": "action",
-      "version": "0.4.3",
+      "version": "0.4.4",
      "author": "Fu-Jie",
      "description": "将对话导出为 Word (.docx)，支持 Mermaid 图表 (客户端渲染 SVG+PNG)、LaTeX 数学公式、真实超链接、增强表格格式、代码高亮和引用块。",
-      "downloads": 119,
-      "views": 2104,
+      "downloads": 122,
+      "views": 2149,
      "upvotes": 13,
      "saves": 6,
      "comments": 4,
      "created_at": "2026-01-04",
-      "updated_at": "2026-01-28",
+      "updated_at": "2026-02-07",
      "url": "https://openwebui.com/posts/导出为_word_支持公式流程图表格和代码块_8a6306c0"
    },
+    {
+      "title": "GitHub Copilot Official SDK Pipe",
+      "slug": "github_copilot_official_sdk_pipe_ce96f7b4",
+      "type": "pipe",
+      "version": "0.5.1",
+      "author": "Fu-Jie",
+      "description": "Integrate GitHub Copilot SDK. Supports dynamic models, multi-turn conversation, streaming, multimodal input, infinite sessions, and frontend debug logging.",
+      "downloads": 71,
+      "views": 2069,
+      "upvotes": 13,
+      "saves": 6,
+      "comments": 4,
+      "created_at": "2026-01-26",
+      "updated_at": "2026-02-08",
+      "url": "https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4"
+    },
    {
      "title": "智能信息图",
      "slug": "智能信息图_e04a48ff",
@@ -179,8 +196,8 @@
      "version": "1.5.0",
      "author": "Fu-Jie",
      "description": "基于 AntV Infographic 的智能信息图生成插件。支持多种专业模板，自动图标匹配，并提供 SVG/PNG 下载功能。",
-      "downloads": 56,
-      "views": 1003,
+      "downloads": 58,
+      "views": 1031,
      "upvotes": 10,
      "saves": 1,
      "comments": 0,
@@ -195,8 +212,8 @@
      "version": "0.1.0",
      "author": "Fu-Jie",
      "description": "Automatically extracts project rules from conversations and injects them into the folder's system prompt.",
-      "downloads": 55,
-      "views": 1195,
+      "downloads": 58,
+      "views": 1247,
      "upvotes": 6,
      "saves": 8,
      "comments": 0,
@@ -204,22 +221,6 @@
      "updated_at": "2026-01-20",
      "url": "https://openwebui.com/posts/folder_memory_auto_evolving_project_context_4a9875b2"
    },
-    {
-      "title": "GitHub Copilot Official SDK Pipe",
-      "slug": "github_copilot_official_sdk_pipe_ce96f7b4",
-      "type": "action",
-      "version": "0.2.3",
-      "author": "Fu-Jie",
-      "description": "Integrate GitHub Copilot SDK. Supports dynamic models, multi-turn conversation, streaming, multimodal input, infinite sessions, and frontend debug logging.",
-      "downloads": 49,
-      "views": 1835,
-      "upvotes": 11,
-      "saves": 5,
-      "comments": 1,
-      "created_at": "2026-01-26",
-      "updated_at": "2026-01-29",
-      "url": "https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4"
-    },
    {
      "title": "思维导图",
      "slug": "智能生成交互式思维导图帮助用户可视化知识_8d4b097b",
@@ -228,7 +229,7 @@
      "author": "Fu-Jie",
      "description": "智能分析文本内容,生成交互式思维导图,帮助用户结构化和可视化知识。",
      "downloads": 37,
-      "views": 563,
+      "views": 573,
      "upvotes": 6,
      "saves": 2,
      "comments": 0,
@@ -243,10 +244,10 @@
      "version": "1.2.2",
      "author": "Fu-Jie",
      "description": "通过智能摘要和消息压缩，降低长对话的 token 消耗，同时保持对话连贯性。",
-      "downloads": 31,
-      "views": 624,
+      "downloads": 33,
+      "views": 646,
      "upvotes": 7,
-      "saves": 4,
+      "saves": 5,
      "comments": 0,
      "created_at": "2025-11-08",
      "updated_at": "2026-01-28",
@@ -260,7 +261,7 @@
      "author": "Fu-Jie",
      "description": "快速将文本提炼为精美的学习记忆卡片，支持核心要点提取与分类。",
      "downloads": 27,
-      "views": 659,
+      "views": 670,
      "upvotes": 8,
      "saves": 1,
      "comments": 0,
@@ -275,8 +276,8 @@
      "version": "1.0.0",
      "author": "Fu-Jie",
      "description": "全方位的思维透镜 —— 从背景全景到逻辑脉络，从深度洞察到行动路径。",
-      "downloads": 17,
-      "views": 400,
+      "downloads": 24,
+      "views": 419,
      "upvotes": 5,
      "saves": 1,
      "comments": 0,
@@ -292,10 +293,10 @@
      "author": "",
      "description": "",
      "downloads": 0,
-      "views": 1421,
-      "upvotes": 11,
-      "saves": 16,
-      "comments": 7,
+      "views": 1460,
+      "upvotes": 12,
+      "saves": 18,
+      "comments": 8,
      "created_at": "2026-01-25",
      "updated_at": "2026-01-28",
      "url": "https://openwebui.com/posts/open_webui_prompt_plus_ai_powered_prompt_manager_s_15fa060e"
@@ -308,7 +309,7 @@
      "author": "",
      "description": "",
      "downloads": 0,
-      "views": 134,
+      "views": 139,
      "upvotes": 2,
      "saves": 0,
      "comments": 0,
@@ -324,9 +325,9 @@
      "author": "",
      "description": "",
      "downloads": 0,
-      "views": 1391,
+      "views": 1411,
      "upvotes": 14,
-      "saves": 9,
+      "saves": 10,
      "comments": 2,
      "created_at": "2026-01-10",
      "updated_at": "2026-01-10",
@@ -338,11 +339,11 @@
    "name": "Fu-Jie",
    "profile_url": "https://openwebui.com/u/Fu-Jie",
    "profile_image": "https://community.s3.openwebui.com/uploads/users/b15d1348-4347-42b4-b815-e053342d6cb0/profile_d9510745-4bd4-4f8f-a997-4a21847d9300.webp",
-    "followers": 203,
+    "followers": 210,
    "following": 4,
-    "total_points": 246,
-    "post_points": 211,
-    "comment_points": 35,
-    "contributions": 40
+    "total_points": 259,
+    "post_points": 219,
+    "comment_points": 40,
+    "contributions": 43
  }
 }
--- a/docs/community-stats.md
+++ b/docs/community-stats.md
@@ -1,21 +1,22 @@
 # 📊 OpenWebUI Community Stats Report

-> 📅 Updated: 2026-02-07 09:38
+> 📅 Updated: 2026-02-09 11:10

 ## 📈 Overview

 | Metric | Value |
 |------|------|
 | 📝 Total Posts | 20 |
-| ⬇️ Total Downloads | 3675 |
-| 👁️ Total Views | 43326 |
-| 👍 Total Upvotes | 213 |
-| 💾 Total Saves | 253 |
-| 💬 Total Comments | 46 |
+| ⬇️ Total Downloads | 3909 |
+| 👁️ Total Views | 45507 |
+| 👍 Total Upvotes | 221 |
+| 💾 Total Saves | 264 |
+| 💬 Total Comments | 51 |

 ## 📂 By Type

- **action**: 15
+- **pipe**: 1
+- **action**: 14
 - **unknown**: 4
 - **filter**: 1

@@ -23,23 +24,23 @@

 | Rank | Title | Type | Version | Downloads | Views | Upvotes | Saves | Updated |
 |:---:|------|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
-| 1 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | action | 0.9.2 | 877 | 7765 | 21 | 47 | 2026-01-28 |
-| 2 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | action | 1.5.0 | 631 | 5825 | 23 | 33 | 2026-01-30 |
-| 3 | [Export to Word (Enhanced)](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | action | 0.4.3 | 343 | 2659 | 12 | 26 | 2026-01-28 |
-| 4 | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | action | 0.3.7 | 327 | 1519 | 7 | 6 | 2026-01-29 |
-| 5 | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | action | 1.2.2 | 327 | 3412 | 13 | 33 | 2026-01-28 |
-| 6 | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | action | 1.2.4 | 293 | 4243 | 17 | 27 | 2026-01-29 |
-| 7 | [Flash Card](https://openwebui.com/posts/flash_card_65a2ea8f) | action | 0.2.4 | 213 | 3239 | 13 | 14 | 2026-01-28 |
-| 8 | [AI Task Instruction Generator](https://openwebui.com/posts/ai_task_instruction_generator_9bab8b37) | unknown |  | 141 | 2180 | 8 | 3 | 2026-01-28 |
-| 9 | [Deep Dive](https://openwebui.com/posts/deep_dive_c0b846e4) | action | 1.0.0 | 132 | 1155 | 6 | 11 | 2026-01-08 |
-| 10 | [导出为 Word (增强版)](https://openwebui.com/posts/导出为_word_支持公式流程图表格和代码块_8a6306c0) | action | 0.4.3 | 119 | 2104 | 13 | 6 | 2026-01-28 |
-| 11 | [智能信息图](https://openwebui.com/posts/智能信息图_e04a48ff) | action | 1.5.0 | 56 | 1003 | 10 | 1 | 2026-01-29 |
-| 12 | [📂 Folder Memory – Auto-Evolving Project Context](https://openwebui.com/posts/folder_memory_auto_evolving_project_context_4a9875b2) | filter | 0.1.0 | 55 | 1195 | 6 | 8 | 2026-01-20 |
-| 13 | [GitHub Copilot Official SDK Pipe](https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4) | action | 0.2.3 | 49 | 1835 | 11 | 5 | 2026-01-29 |
-| 14 | [思维导图](https://openwebui.com/posts/智能生成交互式思维导图帮助用户可视化知识_8d4b097b) | action | 0.9.2 | 37 | 563 | 6 | 2 | 2026-01-28 |
-| 15 | [异步上下文压缩](https://openwebui.com/posts/异步上下文压缩_5c0617cb) | action | 1.2.2 | 31 | 624 | 7 | 4 | 2026-01-28 |
-| 16 | [闪记卡 (Flash Card)](https://openwebui.com/posts/闪记卡生成插件_4a31eac3) | action | 0.2.4 | 27 | 659 | 8 | 1 | 2026-01-28 |
-| 17 | [精读](https://openwebui.com/posts/精读_99830b0f) | action | 1.0.0 | 17 | 400 | 5 | 1 | 2026-01-08 |
-| 18 | [🚀 Open WebUI Prompt Plus: AI-Powered Prompt Manager](https://openwebui.com/posts/open_webui_prompt_plus_ai_powered_prompt_manager_s_15fa060e) | unknown |  | 0 | 1421 | 11 | 16 | 2026-01-28 |
-| 19 | [Review of Claude Haiku 4.5](https://openwebui.com/posts/review_of_claude_haiku_45_41b0db39) | unknown |  | 0 | 134 | 2 | 0 | 2026-01-14 |
-| 20 | [ 🛠️ Debug Open WebUI Plugins in Your Browser](https://openwebui.com/posts/debug_open_webui_plugins_in_your_browser_81bf7960) | unknown |  | 0 | 1391 | 14 | 9 | 2026-01-10 |
+| 1 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | action | 0.9.2 | 916 | 8070 | 22 | 50 | 2026-01-28 |
+| 2 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | action | 1.5.0 | 665 | 6205 | 24 | 34 | 2026-01-30 |
+| 3 | [Export to Word Enhanced](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | action | 0.4.4 | 365 | 2841 | 14 | 26 | 2026-02-07 |
+| 4 | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | action | 1.2.2 | 350 | 3580 | 14 | 33 | 2026-01-28 |
+| 5 | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | action | 0.3.7 | 334 | 1569 | 7 | 6 | 2026-01-29 |
+| 6 | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | action | 1.2.4 | 310 | 4433 | 17 | 28 | 2026-01-29 |
+| 7 | [Flash Card](https://openwebui.com/posts/flash_card_65a2ea8f) | action | 0.2.4 | 217 | 3302 | 13 | 14 | 2026-01-28 |
+| 8 | [AI Task Instruction Generator](https://openwebui.com/posts/ai_task_instruction_generator_9bab8b37) | unknown |  | 178 | 2475 | 8 | 4 | 2026-01-28 |
+| 9 | [Deep Dive](https://openwebui.com/posts/deep_dive_c0b846e4) | action | 1.0.0 | 144 | 1218 | 6 | 11 | 2026-01-08 |
+| 10 | [导出为Word增强版](https://openwebui.com/posts/导出为_word_支持公式流程图表格和代码块_8a6306c0) | action | 0.4.4 | 122 | 2149 | 13 | 6 | 2026-02-07 |
+| 11 | [GitHub Copilot Official SDK Pipe](https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4) | pipe | 0.5.1 | 71 | 2069 | 13 | 6 | 2026-02-08 |
+| 12 | [智能信息图](https://openwebui.com/posts/智能信息图_e04a48ff) | action | 1.5.0 | 58 | 1031 | 10 | 1 | 2026-01-29 |
+| 13 | [📂 Folder Memory – Auto-Evolving Project Context](https://openwebui.com/posts/folder_memory_auto_evolving_project_context_4a9875b2) | filter | 0.1.0 | 58 | 1247 | 6 | 8 | 2026-01-20 |
+| 14 | [思维导图](https://openwebui.com/posts/智能生成交互式思维导图帮助用户可视化知识_8d4b097b) | action | 0.9.2 | 37 | 573 | 6 | 2 | 2026-01-28 |
+| 15 | [异步上下文压缩](https://openwebui.com/posts/异步上下文压缩_5c0617cb) | action | 1.2.2 | 33 | 646 | 7 | 5 | 2026-01-28 |
+| 16 | [闪记卡 (Flash Card)](https://openwebui.com/posts/闪记卡生成插件_4a31eac3) | action | 0.2.4 | 27 | 670 | 8 | 1 | 2026-01-28 |
+| 17 | [精读](https://openwebui.com/posts/精读_99830b0f) | action | 1.0.0 | 24 | 419 | 5 | 1 | 2026-01-08 |
+| 18 | [🚀 Open WebUI Prompt Plus: AI-Powered Prompt Manager](https://openwebui.com/posts/open_webui_prompt_plus_ai_powered_prompt_manager_s_15fa060e) | unknown |  | 0 | 1460 | 12 | 18 | 2026-01-28 |
+| 19 | [Review of Claude Haiku 4.5](https://openwebui.com/posts/review_of_claude_haiku_45_41b0db39) | unknown |  | 0 | 139 | 2 | 0 | 2026-01-14 |
+| 20 | [ 🛠️ Debug Open WebUI Plugins in Your Browser](https://openwebui.com/posts/debug_open_webui_plugins_in_your_browser_81bf7960) | unknown |  | 0 | 1411 | 14 | 10 | 2026-01-10 |
--- a/docs/community-stats.zh.md
+++ b/docs/community-stats.zh.md
@@ -1,21 +1,22 @@
 # 📊 OpenWebUI 社区统计报告

-> 📅 更新时间: 2026-02-07 09:38
+> 📅 更新时间: 2026-02-09 11:10

 ## 📈 总览

 | 指标 | 数值 |
 |------|------|
 | 📝 发布数量 | 20 |
-| ⬇️ 总下载量 | 3675 |
-| 👁️ 总浏览量 | 43326 |
-| 👍 总点赞数 | 213 |
-| 💾 总收藏数 | 253 |
-| 💬 总评论数 | 46 |
+| ⬇️ 总下载量 | 3909 |
+| 👁️ 总浏览量 | 45507 |
+| 👍 总点赞数 | 221 |
+| 💾 总收藏数 | 264 |
+| 💬 总评论数 | 51 |

 ## 📂 按类型分类

- **action**: 15
+- **pipe**: 1
+- **action**: 14
 - **unknown**: 4
 - **filter**: 1

@@ -23,23 +24,23 @@

 | 排名 | 标题 | 类型 | 版本 | 下载 | 浏览 | 点赞 | 收藏 | 更新日期 |
 |:---:|------|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
-| 1 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | action | 0.9.2 | 877 | 7765 | 21 | 47 | 2026-01-28 |
-| 2 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | action | 1.5.0 | 631 | 5825 | 23 | 33 | 2026-01-30 |
-| 3 | [Export to Word (Enhanced)](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | action | 0.4.3 | 343 | 2659 | 12 | 26 | 2026-01-28 |
-| 4 | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | action | 0.3.7 | 327 | 1519 | 7 | 6 | 2026-01-29 |
-| 5 | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | action | 1.2.2 | 327 | 3412 | 13 | 33 | 2026-01-28 |
-| 6 | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | action | 1.2.4 | 293 | 4243 | 17 | 27 | 2026-01-29 |
-| 7 | [Flash Card](https://openwebui.com/posts/flash_card_65a2ea8f) | action | 0.2.4 | 213 | 3239 | 13 | 14 | 2026-01-28 |
-| 8 | [AI Task Instruction Generator](https://openwebui.com/posts/ai_task_instruction_generator_9bab8b37) | unknown |  | 141 | 2180 | 8 | 3 | 2026-01-28 |
-| 9 | [Deep Dive](https://openwebui.com/posts/deep_dive_c0b846e4) | action | 1.0.0 | 132 | 1155 | 6 | 11 | 2026-01-08 |
-| 10 | [导出为 Word (增强版)](https://openwebui.com/posts/导出为_word_支持公式流程图表格和代码块_8a6306c0) | action | 0.4.3 | 119 | 2104 | 13 | 6 | 2026-01-28 |
-| 11 | [智能信息图](https://openwebui.com/posts/智能信息图_e04a48ff) | action | 1.5.0 | 56 | 1003 | 10 | 1 | 2026-01-29 |
-| 12 | [📂 Folder Memory – Auto-Evolving Project Context](https://openwebui.com/posts/folder_memory_auto_evolving_project_context_4a9875b2) | filter | 0.1.0 | 55 | 1195 | 6 | 8 | 2026-01-20 |
-| 13 | [GitHub Copilot Official SDK Pipe](https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4) | action | 0.2.3 | 49 | 1835 | 11 | 5 | 2026-01-29 |
-| 14 | [思维导图](https://openwebui.com/posts/智能生成交互式思维导图帮助用户可视化知识_8d4b097b) | action | 0.9.2 | 37 | 563 | 6 | 2 | 2026-01-28 |
-| 15 | [异步上下文压缩](https://openwebui.com/posts/异步上下文压缩_5c0617cb) | action | 1.2.2 | 31 | 624 | 7 | 4 | 2026-01-28 |
-| 16 | [闪记卡 (Flash Card)](https://openwebui.com/posts/闪记卡生成插件_4a31eac3) | action | 0.2.4 | 27 | 659 | 8 | 1 | 2026-01-28 |
-| 17 | [精读](https://openwebui.com/posts/精读_99830b0f) | action | 1.0.0 | 17 | 400 | 5 | 1 | 2026-01-08 |
-| 18 | [🚀 Open WebUI Prompt Plus: AI-Powered Prompt Manager](https://openwebui.com/posts/open_webui_prompt_plus_ai_powered_prompt_manager_s_15fa060e) | unknown |  | 0 | 1421 | 11 | 16 | 2026-01-28 |
-| 19 | [Review of Claude Haiku 4.5](https://openwebui.com/posts/review_of_claude_haiku_45_41b0db39) | unknown |  | 0 | 134 | 2 | 0 | 2026-01-14 |
-| 20 | [ 🛠️ Debug Open WebUI Plugins in Your Browser](https://openwebui.com/posts/debug_open_webui_plugins_in_your_browser_81bf7960) | unknown |  | 0 | 1391 | 14 | 9 | 2026-01-10 |
+| 1 | [Smart Mind Map](https://openwebui.com/posts/turn_any_text_into_beautiful_mind_maps_3094c59a) | action | 0.9.2 | 916 | 8070 | 22 | 50 | 2026-01-28 |
+| 2 | [Smart Infographic](https://openwebui.com/posts/smart_infographic_ad6f0c7f) | action | 1.5.0 | 665 | 6205 | 24 | 34 | 2026-01-30 |
+| 3 | [Export to Word Enhanced](https://openwebui.com/posts/export_to_word_enhanced_formatting_fca6a315) | action | 0.4.4 | 365 | 2841 | 14 | 26 | 2026-02-07 |
+| 4 | [Async Context Compression](https://openwebui.com/posts/async_context_compression_b1655bc8) | action | 1.2.2 | 350 | 3580 | 14 | 33 | 2026-01-28 |
+| 5 | [Export to Excel](https://openwebui.com/posts/export_mulit_table_to_excel_244b8f9d) | action | 0.3.7 | 334 | 1569 | 7 | 6 | 2026-01-29 |
+| 6 | [Markdown Normalizer](https://openwebui.com/posts/markdown_normalizer_baaa8732) | action | 1.2.4 | 310 | 4433 | 17 | 28 | 2026-01-29 |
+| 7 | [Flash Card](https://openwebui.com/posts/flash_card_65a2ea8f) | action | 0.2.4 | 217 | 3302 | 13 | 14 | 2026-01-28 |
+| 8 | [AI Task Instruction Generator](https://openwebui.com/posts/ai_task_instruction_generator_9bab8b37) | unknown |  | 178 | 2475 | 8 | 4 | 2026-01-28 |
+| 9 | [Deep Dive](https://openwebui.com/posts/deep_dive_c0b846e4) | action | 1.0.0 | 144 | 1218 | 6 | 11 | 2026-01-08 |
+| 10 | [导出为Word增强版](https://openwebui.com/posts/导出为_word_支持公式流程图表格和代码块_8a6306c0) | action | 0.4.4 | 122 | 2149 | 13 | 6 | 2026-02-07 |
+| 11 | [GitHub Copilot Official SDK Pipe](https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4) | pipe | 0.5.1 | 71 | 2069 | 13 | 6 | 2026-02-08 |
+| 12 | [智能信息图](https://openwebui.com/posts/智能信息图_e04a48ff) | action | 1.5.0 | 58 | 1031 | 10 | 1 | 2026-01-29 |
+| 13 | [📂 Folder Memory – Auto-Evolving Project Context](https://openwebui.com/posts/folder_memory_auto_evolving_project_context_4a9875b2) | filter | 0.1.0 | 58 | 1247 | 6 | 8 | 2026-01-20 |
+| 14 | [思维导图](https://openwebui.com/posts/智能生成交互式思维导图帮助用户可视化知识_8d4b097b) | action | 0.9.2 | 37 | 573 | 6 | 2 | 2026-01-28 |
+| 15 | [异步上下文压缩](https://openwebui.com/posts/异步上下文压缩_5c0617cb) | action | 1.2.2 | 33 | 646 | 7 | 5 | 2026-01-28 |
+| 16 | [闪记卡 (Flash Card)](https://openwebui.com/posts/闪记卡生成插件_4a31eac3) | action | 0.2.4 | 27 | 670 | 8 | 1 | 2026-01-28 |
+| 17 | [精读](https://openwebui.com/posts/精读_99830b0f) | action | 1.0.0 | 24 | 419 | 5 | 1 | 2026-01-08 |
+| 18 | [🚀 Open WebUI Prompt Plus: AI-Powered Prompt Manager](https://openwebui.com/posts/open_webui_prompt_plus_ai_powered_prompt_manager_s_15fa060e) | unknown |  | 0 | 1460 | 12 | 18 | 2026-01-28 |
+| 19 | [Review of Claude Haiku 4.5](https://openwebui.com/posts/review_of_claude_haiku_45_41b0db39) | unknown |  | 0 | 139 | 2 | 0 | 2026-01-14 |
+| 20 | [ 🛠️ Debug Open WebUI Plugins in Your Browser](https://openwebui.com/posts/debug_open_webui_plugins_in_your_browser_81bf7960) | unknown |  | 0 | 1411 | 14 | 10 | 2026-01-10 |
--- a/docs/plugins/actions/export-to-word.md
+++ b/docs/plugins/actions/export-to-word.md
@@ -1,7 +1,7 @@
 # Export to Word

 <span class="category-badge action">Action</span>
-<span class="version-badge">v0.4.3</span>
+<span class="version-badge">v0.4.4</span>

 Export conversation to Word (.docx) with **syntax highlighting**, **native math equations**, **Mermaid diagrams**, **citations**, and **enhanced table formatting**.

@@ -53,11 +53,17 @@ You can configure the following settings via the **Valves** button in the plugin
 | `MATH_ENABLE` | Enable LaTeX math block conversion. | `True` |
 | `MATH_INLINE_DOLLAR_ENABLE` | Enable inline `$ ... $` math conversion. | `True` |

-## 🔥 What's New in v0.4.3
+## 🔥 What's New in v0.4.4
+
+- 🧹 **Content Cleanup**: Enhanced stripping of `<details>` blocks (often used for tool calls/thinking process) to ensure a clean final document.
+- 📄 **Standard Document Formatting**: Applied professional document formatting standards for titles and headings (centered title, bold, optimized font sizes and spacing), including GB/T compliance for Chinese content.
+- 🔠 **Font Rendering Fix**: Fixed an issue where CJK characters would fallback to MS Gothic in Word; now correctly uses the configured Asian font (e.g., SimSun).
+- ⚙️ **Title Alignment**: Added `TITLE_ALIGNMENT` valve to configure document title alignment (left, center, right).

 ### User-Level Configuration (UserValves)

 Users can override the following settings in their personal settings:
+
 - `TITLE_SOURCE`
 - `UI_LANGUAGE`
 - `FONT_LATIN`, `FONT_ASIAN`, `FONT_CODE`
@@ -120,4 +126,4 @@ Users can override the following settings in their personal settings:
 ## Source Code

 [:fontawesome-brands-github: View on GitHub](https://github.com/Fu-Jie/awesome-openwebui/tree/main/plugins/actions/export_to_docx){ .md-button }
-**Author:** [Fu-Jie](https://github.com/Fu-Jie) | **Version:** 0.4.3 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)
+**Author:** [Fu-Jie](https://github.com/Fu-Jie) | **Version:** 0.4.4 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)
--- a/docs/plugins/actions/export-to-word.zh.md
+++ b/docs/plugins/actions/export-to-word.zh.md
@@ -1,7 +1,7 @@
 # Export to Word（导出为 Word）

 <span class="category-badge action">Action</span>
-<span class="version-badge">v0.4.3</span>
+<span class="version-badge">v0.4.4</span>

 将当前对话导出为完美格式的 Word 文档，支持**代码语法高亮**、**原生数学公式**、**Mermaid 图表**、**引用资料**以及**增强表格**渲染。

@@ -53,9 +53,17 @@ Export to Word 插件会把聊天消息从 Markdown 转成精致的 Word 文档
 | `启用数学公式` | 启用 LaTeX 数学公式块转换。 | `True` |
 | `启用行内公式` | 启用行内 `$ ... $` 数学公式转换。 | `True` |

+## 🔥 v0.4.4 更新内容
+
+- 🧹 **内容清理加强**: 增强了对 `<details>` 块（通常包含工具调用或思考过程）的清理，确保最终文档整洁。
+- 📄 **文档格式标准化**: 采用了专业的文档排版标准（兼容中文 GB/T 规范），标题居中加粗，各级标题使用标准字号和间距。
+- 🔠 **字体渲染修复**: 修复了 CJK 字符在 Word 中回退到 MS Gothic 的问题；现在正确使用配置的中文字体（例如宋体）。
+- ⚙️ **标题对齐配置**: 新增 `标题对齐方式` Valve，支持配置文档标题的对齐方式（左对齐、居中、右对齐）。
+
 ### 用户级配置 (UserValves)

 用户可以在个人设置中覆盖以下配置：
+
 - `文档标题来源`
 - `界面语言`
 - `英文字体`, `中文字体`, `代码字体`
@@ -117,4 +125,5 @@ Export to Word 插件会把聊天消息从 Markdown 转成精致的 Word 文档

 ## 源码

-[:fontawes**Author:** [Fu-Jie](https://github.com/Fu-Jie) | **Version:** 0.4.3 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)/tree/main/plugins/actions/export_to_docx){ .md-button }
+[:fontawesome-brands-github: View on GitHub](https://github.com/Fu-Jie/awesome-openwebui/tree/main/plugins/actions/export_to_docx){ .md-button }
+**Author:** [Fu-Jie](https://github.com/Fu-Jie) | **Version:** 0.4.4 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)
--- a/docs/plugins/actions/index.md
+++ b/docs/plugins/actions/index.md
@@ -17,7 +17,7 @@ Actions are interactive plugins that:

 <div class="grid cards" markdown>

-   :material-brain:{ .lg .middle } **Smart Mind Map**
+- :material-brain:{ .lg .middle } **Smart Mind Map**

    ---

@@ -27,7 +27,7 @@ Actions are interactive plugins that:

    [:octicons-arrow-right-24: Documentation](smart-mind-map.md)

-   :material-chart-bar:{ .lg .middle } **Smart Infographic**
+- :material-chart-bar:{ .lg .middle } **Smart Infographic**

    ---

@@ -37,7 +37,7 @@ Actions are interactive plugins that:

    [:octicons-arrow-right-24: Documentation](smart-infographic.md)

-   :material-card-text:{ .lg .middle } **Flash Card**
+- :material-card-text:{ .lg .middle } **Flash Card**

    ---

@@ -47,7 +47,7 @@ Actions are interactive plugins that:

    [:octicons-arrow-right-24: Documentation](flash-card.md)

-   :material-file-excel:{ .lg .middle } **Export to Excel**
+- :material-file-excel:{ .lg .middle } **Export to Excel**

    ---

@@ -57,17 +57,17 @@ Actions are interactive plugins that:

    [:octicons-arrow-right-24: Documentation](export-to-excel.md)

-   :material-file-word-box:{ .lg .middle } **Export to Word (Enhanced Formatting)**
+- :material-file-word-box:{ .lg .middle } **Export to Word (Enhanced Formatting)**

    ---

    Export the current conversation to a formatted Word doc with **syntax highlighting**, **native math equations**, **Mermaid diagrams**, **citations**, and **enhanced table formatting**.
-    
-    **Version:** 0.4.2
+
+    **Version:** 0.4.4

    [:octicons-arrow-right-24: Documentation](export-to-word.md)

-   :material-brain:{ .lg .middle } **Deep Dive**
+- :material-brain:{ .lg .middle } **Deep Dive**

    ---

@@ -77,8 +77,6 @@ Actions are interactive plugins that:

    [:octicons-arrow-right-24: Documentation](deep-dive.md)

-
-
 </div>

 ---
--- a/docs/plugins/actions/index.zh.md
+++ b/docs/plugins/actions/index.zh.md
@@ -17,7 +17,7 @@ Actions 是交互式插件，能够：

 <div class="grid cards" markdown>

-   :material-brain:{ .lg .middle } **Smart Mind Map**
+- :material-brain:{ .lg .middle } **Smart Mind Map**

    ---

@@ -27,7 +27,7 @@ Actions 是交互式插件，能够：

    [:octicons-arrow-right-24: 查看文档](smart-mind-map.md)

-   :material-chart-bar:{ .lg .middle } **Smart Infographic**
+- :material-chart-bar:{ .lg .middle } **Smart Infographic**

    ---

@@ -37,7 +37,7 @@ Actions 是交互式插件，能够：

    [:octicons-arrow-right-24: 查看文档](smart-infographic.md)

-   :material-card-text:{ .lg .middle } **Flash Card（闪记卡）**
+- :material-card-text:{ .lg .middle } **Flash Card（闪记卡）**

    ---

@@ -47,7 +47,7 @@ Actions 是交互式插件，能够：

    [:octicons-arrow-right-24: 查看文档](flash-card.md)

-   :material-file-excel:{ .lg .middle } **Export to Excel**
+- :material-file-excel:{ .lg .middle } **Export to Excel**

    ---

@@ -57,17 +57,17 @@ Actions 是交互式插件，能够：

    [:octicons-arrow-right-24: 查看文档](export-to-excel.md)

-   :material-file-word-box:{ .lg .middle } **Word 导出 (格式增强)**
+- :material-file-word-box:{ .lg .middle } **Word 导出 (格式增强)**

    ---

    将当前对话导出为完美格式的 Word 文档，支持**代码语法高亮**、**原生数学公式**、**Mermaid 图表**、**引用资料**以及**增强表格**渲染。

-    **版本：** 0.4.2
+    **版本：** 0.4.4

    [:octicons-arrow-right-24: 查看文档](export-to-word.md)

-   :material-brain:{ .lg .middle } **精读 (Deep Dive)**
+- :material-brain:{ .lg .middle } **精读 (Deep Dive)**

    ---

@@ -77,8 +77,6 @@ Actions 是交互式插件，能够：

    [:octicons-arrow-right-24: 查看文档](deep-dive.zh.md)

-
-
 </div>

 ---
--- a/docs/plugins/filters/github-copilot-sdk-files-filter.md
+++ b/docs/plugins/filters/github-copilot-sdk-files-filter.md
@@ -0,0 +1,53 @@
+# GitHub Copilot SDK Files Filter (v0.1.2)
+
+This is a dedicated **companion filter plugin** designed specifically for the [GitHub Copilot SDK Pipe](../pipes/github-copilot-sdk.md).
+
+Its core mission is to **protect user-uploaded files from being "pre-processed" by the OpenWebUI core system, ensuring that the Copilot Agent receives the raw files for autonomous analysis.**
+
+## 🎯 Why is this needed?
+
+In OpenWebUI's default workflow, when you upload a file (e.g., PDF, Excel, Python script), OpenWebUI automatically initiates a **RAG (Retrieval-Augmented Generation)** process: parsing the file, vectorizing it, extracting text, and injecting it into the prompt.
+
+While useful for standard models, this is often disruptive for a **Copilot SDK Agent**:
+
+1. **Agent Needs Raw Files**: The Agent may need to run Python code to read an Excel file or analyze a full directory structure, not chopped-up text fragments.
+2. **Context Pollution**: Large amounts of text injected by RAG consume tokens and can confuse the Agent about "where the file is."
+3. **Conflicts**: If you have other multimodal plugins installed (like Gemini Filter), they might compete for file processing rights.
+
+**This plugin acts as a "bodyguard" to solve these issues.**
+
+## 🚀 How it Works
+
+When you select a Copilot model (name containing `copilot_sdk`) in OpenWebUI and send a file:
+
+1. **Intercept**: This plugin runs with high priority (Priority 0), before RAG and other filters.
+2. **Relocate**: Detecting a Copilot model, it moves the `files` list from the request to a secure custom field `copilot_files`.
+3. **Hide**: It clears the original `files` field.
+4. **Status Update**: It emits a status message "Managed X files for Copilot (RAG Bypassed)" to the UI.
+5. **Pass**: The OpenWebUI core sees an empty `files` list and **does not trigger RAG**.
+6. **Deliver**: The subsequent [Copilot SDK Pipe](../pipes/github-copilot-sdk.md) plugin checks `copilot_files`, retrieves file information, and automatically copies them into the Agent's isolated workspace.
+
+## 📦 Installation & Configuration
+
+### 1. Installation
+
+Import this plugin on the OpenWebUI **Functions** page.
+
+### 2. Enable
+
+Ensure this Filter is enabled globally or in chat settings.
+
+### 3. Configuration (Valves)
+
+Default settings work for most users unless you have specific needs:
+
+| Parameter | Description | Default |
+| :--- | :--- | :--- |
+| **priority** | Execution priority of the filter. **Must be lower than OpenWebUI RAG priority**. | `0` |
+| **target_model_keyword** | Keyword to identify Copilot models. Only models containing this keyword will trigger file interception. | `copilot_sdk` |
+
+## ⚠️ Important Notes
+
+* **Must be used with Copilot SDK Pipe**: If you install this plugin without the main Pipe plugin, uploaded files will simply "disappear" (as no subsequent plugin will look for them in `copilot_files`).
+* **Gemini Filter Compatibility**: This plugin is fully compatible with the Gemini Multimodal Filter. As long as the priority is set correctly (This Plugin < Gemini Plugin), they can coexist without interference.
+* **Physical File Path**: Ensure the `OPENWEBUI_UPLOAD_PATH` is correctly set in the Pipe plugin Valves for the actual file transport to work.
--- a/docs/plugins/filters/github-copilot-sdk-files-filter.zh.md
+++ b/docs/plugins/filters/github-copilot-sdk-files-filter.zh.md
@@ -0,0 +1,53 @@
+# GitHub Copilot SDK 文件过滤器 (v0.1.2)
+
+这是一个专门为 [GitHub Copilot SDK Pipe](../pipes/github-copilot-sdk.zh.md) 设计的**伴侣过滤器插件**。
+
+它的核心使命是：**保护用户上传的文件不被 OpenWebUI 核心系统“抢先处理”，确保 Copilot Agent 能够接收到原始文件并进行自主分析。**
+
+## 🎯 为什么需要它？
+
+在 OpenWebUI 的默认流程中，当你上传一个文件（如 PDF、Excel、Python 脚本）时，OpenWebUI 会自动启动 **RAG（检索增强生成）** 流程：解析文件、向量化、提取文本并注入到提示词中。
+
+虽然这对普通模型很有用，但对于 **Copilot SDK Agent** 来说，这往往是干扰：
+
+1. **Agent 需要原始文件**：Agent 可能需要运行 Python 代码读取 Excel，或者分析完整的代码结构，而不是被切碎的文本片段。
+2. **上下文污染**：RAG 注入的大量文本会消耗 Token，且容易让 Agent 混淆“文件在哪里”。
+3. **冲突**：如果你同时安装了其他多模态插件（如 Gemini Filter），它们可能会争夺文件的处理权。
+
+**本插件就是为了解决这个问题而生的“保镖”。**
+
+## 🚀 功能原理
+
+当你在 OpenWebUI 中选择了一个 Copilot 模型（名称包含 `copilot_sdk`）并发送文件时：
+
+1. **拦截 (Intercept)**：本插件会以极高的优先级（Priority 0）运行，先于 RAG 和其他过滤器。
+2. **搬运 (Relocate)**：它检测到模型是 Copilot，便将请求中的 `files`（文件列表）移动到一个安全的自定义字段 `copilot_files` 中。
+3. **隐身 (Hide)**：它清空原始的 `files` 字段。
+4. **状态更新**: 它会在 UI 中发出“已为 Copilot 管理 X 个文件 (已绕过 RAG)”的状态消息。
+5. **放行 (Pass)**：OpenWebUI 核心看到 `files` 为空，便**不会触发 RAG**。
+6. **交付 (Deliver)**：后续的 [Copilot SDK Pipe](../pipes/github-copilot-sdk.zh.md) 插件会检查 `copilot_files`，从中获取文件信息，并自动将其复制到 Agent 的独立工作区中。
+
+## 📦 安装与配置
+
+### 1. 安装
+
+在 OpenWebUI 的 **Functions** 页面导入此插件。
+
+### 2. 启用
+
+确保在全局或对话设置中启用了此 Filter。
+
+### 3. 配置 (Valves)
+
+通常不需要修改默认配置，除非你有特殊需求：
+
+| 参数 | 说明 | 默认值 |
+| :--- | :--- | :--- |
+| **priority** | 过滤器的执行优先级。**必须小于 OpenWebUI RAG 的优先级**。 | `0` |
+| **target_model_keyword** | 用于识别 Copilot 模型的关键词。只有包含此关键词的模型才会触发 file 拦截。 | `copilot_sdk` |
+
+## ⚠️ 注意事项
+
+* **必须配合 Copilot SDK Pipe 使用**：如果你没有安装主 Pipe 插件，本插件将导致上传的文件“凭空消失”（因为没有后续插件去 `copilot_files` 里找它们）。
+* **Gemini Filter 兼容性**：本插件已完美兼容 Gemini 多模态过滤器。只要优先级设置正确（本插件 < Gemini 插件），它们可以共存互不干扰。
+* **物理文件路径**: 确保在 Pipe 插件的 Valves 中正确设置了 `OPENWEBUI_UPLOAD_PATH`，以便文件自动搬运功能正常工作。
--- a/docs/plugins/filters/index.md
+++ b/docs/plugins/filters/index.md
@@ -16,7 +16,7 @@ Filters act as middleware in the message pipeline:

 <div class="grid cards" markdown>

-   :material-arrow-collapse-vertical:{ .lg .middle } **Async Context Compression**
+- :material-arrow-collapse-vertical:{ .lg .middle } **Async Context Compression**

    ---

@@ -26,7 +26,7 @@ Filters act as middleware in the message pipeline:

    [:octicons-arrow-right-24: Documentation](async-context-compression.md)

-   :material-text-box-plus:{ .lg .middle } **Context Enhancement**
+- :material-text-box-plus:{ .lg .middle } **Context Enhancement**

    ---

@@ -36,7 +36,7 @@ Filters act as middleware in the message pipeline:

    [:octicons-arrow-right-24: Documentation](context-enhancement.md)

-   :material-folder-refresh:{ .lg .middle } **Folder Memory**
+- :material-folder-refresh:{ .lg .middle } **Folder Memory**

    ---

@@ -46,7 +46,7 @@ Filters act as middleware in the message pipeline:

    [:octicons-arrow-right-24: Documentation](folder-memory.md)

-   :material-format-paint:{ .lg .middle } **Markdown Normalizer**
+- :material-format-paint:{ .lg .middle } **Markdown Normalizer**

    ---

@@ -56,7 +56,7 @@ Filters act as middleware in the message pipeline:

    [:octicons-arrow-right-24: Documentation](markdown_normalizer.md)

-   :material-merge:{ .lg .middle } **Multi-Model Context Merger**
+- :material-merge:{ .lg .middle } **Multi-Model Context Merger**

    ---

@@ -66,7 +66,7 @@ Filters act as middleware in the message pipeline:

    [:octicons-arrow-right-24: Documentation](multi-model-context-merger.md)

-   :material-file-document-multiple:{ .lg .middle } **Web Gemini Multimodal Filter**
+- :material-file-document-multiple:{ .lg .middle } **Web Gemini Multimodal Filter**

    ---

@@ -76,6 +76,16 @@ Filters act as middleware in the message pipeline:

    [:octicons-arrow-right-24: Documentation](web-gemini-multimodel.md)

+- :material-file-shield:{ .lg .middle } **Copilot SDK Files Filter**
+
+    ---
+
+    A specialized filter to bypass OpenWebUI's default RAG for GitHub Copilot SDK models. It ensures the Agent receives raw files for autonomous analysis.
+
+    **Version:** 0.1.2
+
+    [:octicons-arrow-right-24: Documentation](github-copilot-sdk-files-filter.md)
+
 </div>

 ---
--- a/docs/plugins/filters/index.zh.md
+++ b/docs/plugins/filters/index.zh.md
@@ -16,7 +16,7 @@ Filter 充当消息管线中的中间件：

 <div class="grid cards" markdown>

-   :material-arrow-collapse-vertical:{ .lg .middle } **Async Context Compression**
+- :material-arrow-collapse-vertical:{ .lg .middle } **Async Context Compression**

    ---

@@ -26,7 +26,7 @@ Filter 充当消息管线中的中间件：

    [:octicons-arrow-right-24: 查看文档](async-context-compression.md)

-   :material-text-box-plus:{ .lg .middle } **Context Enhancement**
+- :material-text-box-plus:{ .lg .middle } **Context Enhancement**

    ---

@@ -36,7 +36,7 @@ Filter 充当消息管线中的中间件：

    [:octicons-arrow-right-24: 查看文档](context-enhancement.md)

-   :material-folder-refresh:{ .lg .middle } **Folder Memory**
+- :material-folder-refresh:{ .lg .middle } **Folder Memory**

    ---

@@ -46,7 +46,7 @@ Filter 充当消息管线中的中间件：

    [:octicons-arrow-right-24: 查看文档](folder-memory.zh.md)

-   :material-format-paint:{ .lg .middle } **Markdown Normalizer**
+- :material-format-paint:{ .lg .middle } **Markdown Normalizer**

    ---

@@ -56,6 +56,16 @@ Filter 充当消息管线中的中间件：

    [:octicons-arrow-right-24: 查看文档](markdown_normalizer.zh.md)

+- :material-file-shield:{ .lg .middle } **Copilot SDK 文件过滤器**
+
+    ---
+
+    专门用于绕过 OpenWebUI 默认 RAG 机制的过滤器，针对 GitHub Copilot SDK 模型。确保 Agent 能够接收到原始文件进行自主分析。
+
+    **版本：** 0.1.2
+
+    [:octicons-arrow-right-24: 查看文档](github-copilot-sdk-files-filter.zh.md)
+
 </div>

 ---
--- a/docs/plugins/pipes/github-copilot-sdk-deep-dive.md
+++ b/docs/plugins/pipes/github-copilot-sdk-deep-dive.md
@@ -0,0 +1,80 @@
+# GitHub Copilot SDK Plugin Deep Dive
+
+**Version:** 0.6.0 | **Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Status:** Production-Ready
+
+The GitHub Copilot SDK plugin is far more than a simple API proxy; it is a highly integrated **Intelligent Agent Runtime Environment**. This document explores the core features, implementation details, technical architecture, and security design that define its capabilities.
+
+---
+
+## 🚀 1. Feature Catalog
+
+The plugin implements advanced capabilities that go far beyond standard API calls:
+
+- **✅ Physical Workspace Management**: Automatically creates isolated directories for each chat session to manage uploaded files and Agent-generated scripts.
+- **✅ Real-time TODO Sync**: Mounts to the OpenWebUI database to persist Agent plans directly into a UI progress bar, ensuring transparency for long-running tasks.
+- **✅ Cross-Ecosystem Tool Bridging**: Translates OpenWebUI Search, Python interpreters, and custom MCP tools into native Copilot capabilities.
+- **✅ Intelligent File Transport**: Ensures the Agent can access raw files (Excel, PDF, Code) like a local developer through physical file duplication.
+- **✅ Chain of Thought Streaming**: Full support for rendering the model's reasoning/thinking process in real-time.
+- **✅ BYOK with Plugin Power**: Connect external OpenAI/Anthropic models while retaining all the plugin's workspace and tool management features.
+
+---
+
+## 🎯 2. Use Cases: Beyond Basic Chat
+
+With these features, the plugin excels in complex, real-world scenarios:
+
+### 📁 Autonomous Repository Maintenance (Agentic DevOps)
+>
+> **Action**: Upload a Zip archive containing a codebase with bugs.
+> **Utility**: The Agent extracts the archive, uses `bash` to navigate and search, applies fixes via the `edit` tool, and runs tests to verify the solution—all within its isolated sandbox.
+
+### 📊 Deep Data Analysis (Data Scientist Agent)
+>
+> **Action**: Upload multiple heavy Excel spreadsheets.
+> **Utility**: By bypassing RAG (via the filter), the Agent loads raw files directly into its Python interpreter, performs cross-table calculations, and generates analytical charts presented via Artifacts.
+
+### 📝 Strategic Task Management
+>
+> **Action**: "Develop a full architecture for a new mobile app."
+> **Utility**: The plugin captures the Agent's breakdown of 20+ sub-tasks. The persistent progress bar reflects ongoing progress (e.g., "Designing Schema", "Drafting API"), providing clarity during long marathons.
+
+---
+
+## 🛡️ 3. Technical Architecture
+
+### 3.1 Three-Layer Sandbox Isolation
+
+To prevent data leakage in multi-user environments, the plugin enforces a strict physical directory structure:
+`/app/backend/data/copilot_workspace/{user_id}/{chat_id}/`
+
+- **Constraint**: Code execution and file storage are confined to the session-specific folder.
+- **Persistence**: Data remains valid across container restarts due to volume mounting.
+
+### 3.2 Dynamic Tool Bridging
+
+How does Copilot "learn" to use OpenWebUI tools?
+
+1. **Introspection**: Analyzes docstrings and type hints of OpenWebUI tools.
+2. **Schema Generation**: Dynamically creates JSON descriptions compliant with the GitHub Copilot specification.
+3. **Routing**: Handles parameter validation, identity injection, and result forwarding between systems.
+
+### 3.3 Event-Driven TODO Hub
+
+The plugin captures internal SDK events to power the UI progress bar:
+
+- **Interceptor**: Listens for `tool.execution_complete` events for the `update_todo` tool.
+- **Storage**: Syncs project metrics directly to the `chat_todos` table in the OpenWebUI database (SQLite/PostgreSQL).
+
+---
+
+## ⚡ 4. Runtime Performance
+
+- **Anti-Shake Logic**: Environment checks happen only once every 24 hours per process, preventing redundant system calls.
+- **Tool Caching**: Persists tool definitions across sessions to reduce overhead, improving initial response times by up to 40%.
+
+---
+
+## 🛠️ 5. Development Best Practices
+
+1. **Use the Filter**: Always pair with `github_copilot_sdk_files_filter` to ensure files reach the Agent in their original binary form.
+2. **File-First Execution**: Encourage the Agent to "write code to file and execute" rather than relying on direct interactive shell input for complex logic.
--- a/docs/plugins/pipes/github-copilot-sdk-deep-dive.zh.md
+++ b/docs/plugins/pipes/github-copilot-sdk-deep-dive.zh.md
@@ -0,0 +1,80 @@
+# GitHub Copilot SDK 插件深度解析 (Deep Dive)
+
+**版本:** 0.6.0 | **作者:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **状态:** 生产级
+
+GitHub Copilot SDK 插件不仅仅是一个 API 转发器，它是一个高度集成的 **智能 Agent 运行环境**。本文将从功能实现、应用场景、技术架构及安全设计四个维度，深入解析这一插件的强大之处。
+
+---
+
+## 🚀 1. 核心功能清单 (Feature Catalog)
+
+插件通过深度集成实现了以下超越普通 API 调用的能力：
+
+- **✅ 物理级工作区管理**: 自动为每个对话创建独立的物理目录，管理用户上传和 Agent 生成的所有文件。
+- **✅ 实时 TODO 状态同步**: 通过数据库挂载，实时将 Agent 的计划提取到 UI 进度条中，解决长任务透明度问题。
+- **✅ 跨生态工具桥接**: 自动将 OpenWebUI 的搜索、Python 运行环境和自定义 MCP 工具转化为 Copilot 原生工具。
+- **✅ 智能文件搬运**: 物理级文件副本传输，确保 Agent 可以像本地开发者一样访问 Excel、PDF 和代码仓库。
+- **✅ 思考过程可见性**: 完整模拟 GitHub Copilot 的思维链 (Thinking Process) 流式展示。
+- **✅ BYOK 二次鉴权**: 支持在 Copilot 框架内接入外部 OpenAI/Anthropic 模型，同时享受插件的所有增强功能。
+
+---
+
+## 🎯 2. 这个插件能用来干什么？ (Use Cases)
+
+基于上述功能，该插件可以胜任以下复杂场景：
+
+### 📁 场景 A：全自动代码仓库维护 (Agentic DevOps)
+>
+> **操作**: 上传一个包含 Bug 的 Zip 压缩包。
+> **用途**: Agent 会自动解密、解压，使用 `bash` 定位问题，调用 `edit` 修改代码，最后运行测试脚本。这一切都在隔离沙箱中完成，你只需要审核最终的补丁。
+
+### 📊 场景 B：深度财务数据审计 (Data Analyst Agent)
+>
+> **操作**: 上传一年的 Excel 财务报表。
+> **用途**: 绕过传统 RAG 的文本切片限制，Agent 直接通过 Python 脚本加载原始表格，进行跨表计算和逻辑校验，并生成可视化图表。
+
+### 📝 场景 C：复杂长任务进度追踪 (Project Manager)
+>
+> **操作**: 输入“请基于以下需求文档编写一个完整的 Web 后端方案”。
+> **用途**: 插件捕捉 Agent 拆解的 20+ 个子任务。顶部的实时进度条会告诉你它正在“设计数据库”还是“编写认证逻辑”，确保你对黑盒任务了如指掌。
+
+---
+
+## 🛡️ 3. 技术架构设计 (Technical Architecture)
+
+### 3.1 三层物理安全隔离 (Workspace Isolation)
+
+为了确保多用户环境下的数据安全，插件强制执行以下物理路径：
+`/app/backend/data/copilot_workspace/{user_id}/{chat_id}/`
+
+- **隔离性**: 进程内代码执行被严格约束在 `chat_id` 目录下。
+- **持久性**: 即使容器重启，挂载路径下的工作成果依然保留。
+
+### 3.2 零配置工具桥接 (Dynamic Tool Bridging)
+
+插件如何让 Copilot “学会”使用 OpenWebUI 的工具？
+
+1. **内省 (Introspection)**: 实时读取工具的 `docstring` 和 `type hints`。
+2. **动态转换**: 生成符合 GitHub Copilot 规范的工具描述符。
+3. **双向路由**: 处理参数校验、身份注入（如认证头）以及结果回传。
+
+### 3.3 数据库集成与事件驱动
+
+插件通过监听 `NDJSON` 事件流，实现状态同步：
+
+- **监听器**: 实时过滤 `tool.execution_complete` 事件。
+- **持久层**: 使用 OpenWebUI 核心相同的 `SQLAlchemy` 引擎操作 `chat_todos` 表。
+
+---
+
+## ⚡ 4. 性能优化 (Performance)
+
+- **环境检查防抖**: 全局类变量保护，版本核对周期为 24 小时，避免高并发下的 I/O 争抢。
+- **工具定义缓存**: 仅在变更时刷新工具元数据，首包响应速度（TTFB）提升约 40%。
+
+---
+
+## 🛠️ 5. 开发建议 (Best Practices)
+
+1. **协同工作**: 必须安装 `github_copilot_sdk_files_filter` 以确保文件以“二进制原文”而非“RAG 切片”传递。
+2. **Python 范式**: 鼓励 Agent 采取“写文件 -> 运行文件”的模式，而非交互式 Shell 输出，以获得更好的执行稳定性。
--- a/docs/plugins/pipes/github-copilot-sdk.md
+++ b/docs/plugins/pipes/github-copilot-sdk.md
@@ -1,100 +1,119 @@
 # GitHub Copilot SDK Pipe for OpenWebUI

-**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.3.0 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **License:** MIT
+**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.6.0 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **License:** MIT

-This is an advanced Pipe function for [OpenWebUI](https://github.com/open-webui/open-webui) that allows you to use GitHub Copilot models (such as `gpt-5`, `gpt-5-mini`, `claude-sonnet-4.5`) directly within OpenWebUI. It is built upon the official [GitHub Copilot SDK for Python](https://github.com/github/copilot-sdk), providing a native integration experience.
+This is an advanced Pipe function for [OpenWebUI](https://github.com/open-webui/open-webui) that integrates the official [GitHub Copilot SDK](https://github.com/github/copilot-sdk). It enables you to use **GitHub Copilot models** (e.g., `gpt-5.2-codex`, `claude-sonnet-4.5`,`gemini-3-pro`, `gpt-5-mini`) **AND** your own models via **BYOK** (OpenAI, Anthropic) directly within OpenWebUI, providing a unified agentic experience with **strict User & Chat-level Workspace Isolation**.

-## 🚀 What's New (v0.3.0) - The Power of "Unified Ecosystem"
+> [!IMPORTANT]
+> **Essential Companion**
+> To unlock file handling and data analysis capabilities, you must install the [GitHub Copilot SDK Files Filter](https://openwebui.com/posts/403a62ee-a596-45e7-be65-fab9cc249dd6).

-* **🔌 Zero-Config Tool Bridge**: Automatically transforms your existing OpenWebUI Functions (Tools) into Copilot-compatible tools. **Copilot now has total access to your entire WebUI toolset!**
-* **🔗 Dynamic MCP Discovery**: Seamlessly connects to MCP servers defined in **Admin Settings -> Connections**. No configuration files required—it just works.
-* **⚡ High-Performance Async Engine**: Background CLI updates and optimized event-driven streaming ensure lightning-fast responses without UI lag.
-* **🛡️ Robust Interoperability**: Advanced sanitization and dynamic Pydantic model generation ensure smooth integration even with complex third-party tools.
+> [!TIP]
+> **No Subscription Required for BYOK**
+> If you are using your own API keys (BYOK mode with OpenAI/Anthropic), **you do NOT need a GitHub Copilot subscription.** A subscription is only required to access GitHub's official models.
+
+---
+
+## ✨ v0.6.0 Updates (What's New)
+
+- **👥 User & Chat Management**: Physical management architecture (`user_id/chat_id`) for absolute resource independence.
+- **🤖 Empowering Agent Autonomy**: Automatic synchronization of raw files to the workspace, enabling direct Python-based analysis of Excel/CSV.
+- **🔧 OpenAPI & External Tool Fixes**: Full support for tools mounted via OpenAPI servers.
+- **📊 Cost Control**: Enhanced **Billing Multiplier Limits** (`MAX_MULTIPLIER`, e.g., set to 0 for free models only) and **Model Keyword Filtering** (`EXCLUDE_KEYWORDS`) for precise cost management.
+- **🧠 Persistent TODO Lists**: Database-backed task tracking that persists across sessions.
+
+---

 ## ✨ Key Capabilities

-* **🌉 The Ultimate Bridge**: The first and only plugin that creates a seamless bridge between **OpenWebUI Tools** and **GitHub Copilot SDK**.
-* **🚀 Official & Native**: Built directly on the official Python SDK, providing the most stable and authentic Copilot experience.
-* **🌊 Advanced Streaming (Thought Process)**: Supports full model reasoning/thinking display with typewriter effects.
-* **🖼️ Intelligent Multimodal**: Full support for images and attachments, enabling Copilot to "see" your workspace.
-* **🛠️ Effortless Setup**: Automatic CLI detection, version enforcement, and dependency management.
-* **🔑 Dual-Layer Security**: Supports secure OAuth flow for Chat and standard PAT for extended MCP capabilities.
+- **🔑 Flexible Auth & BYOK**: Official Copilot subscriptions (PAT) or Bring Your Own Key (OpenAI/Anthropic).
+- **🔌 Universal Tool Protocol**: Native support for **MCP (Model Context Protocol)**, OpenAPI, and OpenWebUI built-in tools.
+- **🛡️ Sandbox Workspace Isolation**: Strict per-session sandboxing for data privacy and security.
+- **♾️ Infinite Session Management**: Smart context window management with automatic compaction for indefinite conversation capability.
+- **🧠 Deep Database Integration**: Real-time persistence of TOD·O lists for long-running workflows.
+- **🌊 Advanced Streaming**: Full support for thinking process/Chain of Thought visualization.
+- **🖼️ Intelligent Multimodal**: Vision capabilities and raw file analysis support.
+- **⚡ Interactive Artifacts**: Automatically renders HTML/JS apps generated by the agent directly in the chat interface.

-## 📦 Installation & Usage
+---

-### 1. Import Function
+## ⚙️ Core Configuration (Valves)

-1. Open OpenWebUI.
-2. Go to **Workspace** -> **Functions**.
-3. Click **+** (Create Function).
-4. Paste the content of `github_copilot_sdk.py` (or `github_copilot_sdk_cn.py` for Chinese) completely.
-5. Save.
+### 1. Administrator Settings (Base)

-### 2. Configure Valves (Settings)
+Administrators define the default behavior for all users in the function settings.

-Find "GitHub Copilot" in the function list and click the **⚙️ (Valves)** icon to configure:
-
-| Parameter | Description | Default |
+| Valve | Default | Description |
 | :--- | :--- | :--- |
-| **GH_TOKEN** | **(Required)** GitHub Access Token (PAT or OAuth Token). Access to Chat. | - |
-| **DEBUG** | Whether to enable debug logs (output to browser console). | `False` |
-| **LOG_LEVEL** | Copilot CLI log level: none, error, warning, info, debug, all. | `error` |
-| **SHOW_THINKING** | Show model reasoning/thinking process (requires streaming + model support). | `True` |
-| **COPILOT_CLI_VERSION** | Specific Copilot CLI version to install/enforce. | `0.0.405` |
-| **EXCLUDE_KEYWORDS** | Exclude models containing these keywords (comma separated). | - |
-| **WORKSPACE_DIR** | Restricted workspace directory for file operations. | - |
-| **INFINITE_SESSION** | Enable Infinite Sessions (automatic context compaction). | `True` |
-| **COMPACTION_THRESHOLD** | Background compaction threshold (0.0-1.0). | `0.8` |
-| **BUFFER_THRESHOLD** | Buffer exhaustion threshold (0.0-1.0). | `0.95` |
-| **TIMEOUT** | Timeout for each stream chunk (seconds). | `300` |
-| **CUSTOM_ENV_VARS** | Custom environment variables (JSON format). | - |
-| **REASONING_EFFORT** | Reasoning effort level: low, medium, high. `xhigh` is supported for some models. | `medium` |
-| **ENFORCE_FORMATTING** | Add formatting instructions to system prompt for better readability. | `True` |
-| **ENABLE_MCP_SERVER** | Enable Direct MCP Client connection (Recommended). | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | Enable OpenWebUI Tools (includes defined and server tools). | `True` |
+| `GH_TOKEN` | `""` | Global GitHub Token (Requires 'Copilot Requests' permission). |
+| `ENABLE_OPENWEBUI_TOOLS` | `True` | Enable OpenWebUI Tools (includes defined Tools and Built-in Tools). |
+| `ENABLE_OPENAPI_SERVER` | `True` | Enable OpenAPI Tool Server connection. |
+| `ENABLE_MCP_SERVER` | `True` | Enable Direct MCP Client connection (Recommended). |
+| `REASONING_EFFORT` | `medium` | Reasoning effort level: low, medium, high. |
+| `SHOW_THINKING` | `True` | Show model reasoning/thinking process. |
+| `INFINITE_SESSION` | `True` | Enable Infinite Sessions (automatic context compaction). |
+| `MAX_MULTIPLIER` | `1.0` | Max allowed billing multiplier (0x for free models only). |
+| `EXCLUDE_KEYWORDS` | `""` | Exclude models containing these keywords (comma separated). |
+| `TIMEOUT` | `300` | Timeout for each stream chunk (seconds). |
+| `BYOK_TYPE` | `openai` | BYOK Provider Type: `openai`, `anthropic`. |
+| `BYOK_BASE_URL` | `""` | BYOK Base URL (e.g., <https://api.openai.com/v1>). |
+| `BYOK_MODELS` | `""` | BYOK Model List (comma separated). Leave empty to fetch from API. |
+| `CUSTOM_ENV_VARS` | `""` | Custom environment variables (JSON format). |
+| `DEBUG` | `False` | Enable this to see detailed logs in your browser console. |

-#### User Valves (per-user overrides)
+### 2. User Settings (Individual Overrides)

-These optional settings can be set per user (overrides global Valves):
+Standard users can override these settings in their individual Profile/Function settings.

-| Parameter | Description | Default |
-| :--- | :--- | :--- |
-| **GH_TOKEN** | Personal GitHub Token (overrides global setting). | - |
-| **REASONING_EFFORT** | Reasoning effort level (low/medium/high/xhigh). | - |
-| **DEBUG** | Enable technical debug logs. | `False` |
-| **SHOW_THINKING** | Show model reasoning/thinking process. | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | Enable OpenWebUI Tools (overrides global). | `True` |
-| **ENABLE_MCP_SERVER** | Enable MCP server loading (overrides global). | `True` |
-| **ENFORCE_FORMATTING** | Enforce formatting guidelines (overrides global). | `True` |
+| Valve | Description |
+| :--- | :--- |
+| `GH_TOKEN` | Use your personal GitHub Token. |
+| `REASONING_EFFORT` | Individual reasoning effort preference. |
+| `SHOW_THINKING` | Show model reasoning/thinking process. |
+| `MAX_MULTIPLIER` | Maximum allowed billing multiplier override. |
+| `EXCLUDE_KEYWORDS` | Exclude models containing these keywords. |
+| `BYOK_API_KEY` | Use your personal OpenAI/Anthropic API Key. |

-### 3. Get Token
+---

-To use GitHub Copilot, you need a GitHub Personal Access Token (PAT) with appropriate permissions.
+## 🎯 Use Cases (What can you do?)

-**Steps to generate your token:**
+- **📁 Fully Autonomous DevOps**: Agent analyzes code, runs tests, and applies patches within its isolated sandbox.
+- **📊 Deep Data Auditing**: Directly process raw Excel/CSV data via Python (bypassing RAG) and generate visual reports.
+- **📝 Long-Task Management**: Automatically decomposes complex requests and persists TOD·O progress across sessions.
+
+---
+
+## ⭐ Support
+
+If this plugin has been useful, a **Star** on [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) would be a great motivation for me. Thank you!
+
+---
+
+## 🚀 Installation & Configuration
+
+### 1) Import Function
+
+1. Open OpenWebUI, go to **Workspace** -> **Functions**.
+2. Click **+** (Create Function), paste the content of `github_copilot_sdk.py`.
+3. Save and ensure it is enabled.
+
+### 2) Get Token

 1. Visit [GitHub Token Settings](https://github.com/settings/tokens?type=beta).
-2. Click **Generate new token (fine-grained)**.
-3. **Repository access**: Select **Public Repositories** (simplest) or **All repositories**.
-4. **Permissions**:
-    * If you chose **All repositories**, you must click **Account permissions**.
-    * Find **Copilot Requests**, and select **Access**.
-5. Generate and copy the Token.
+2. Create **Fine-grained token**, granting **Account permissions** -> **Copilot Requests** access.
+3. Paste the generated Token into the `GH_TOKEN` field in Valves.

-## 📋 Dependencies
+---

-This Pipe will automatically attempt to install the following dependencies:
+## 📋 Troubleshooting & Dependencies

-* `github-copilot-sdk` (Python package)
-* `github-copilot-cli` (Binary file, installed via official script)
+- **Agent ignores files?**: Ensure the Files Filter is enabled, otherwise RAG will interfere with raw binaries.
+- **No progress bar?**: The bar only appears when the Agent uses the `update_todo` tool.
+- **Dependencies**: This Pipe automatically installs `github-copilot-sdk` (Python) and `github-copilot-cli` (Binary).

-## Troubleshooting ❓
+---

-* **Images and Multimodal Usage**:
-  * Ensure `MODEL_ID` is a model that supports multimodal input.
-* **Thinking not shown**:
-  * Ensure **streaming is enabled** and the selected model supports reasoning output.
+## Changelog

-## 📄 License
-
-MIT
+See the full history on GitHub: [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)
--- a/docs/plugins/pipes/github-copilot-sdk.zh.md
+++ b/docs/plugins/pipes/github-copilot-sdk.zh.md
@@ -1,100 +1,119 @@
 # GitHub Copilot SDK 官方管道

-**作者:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **版本:** 0.3.0 | **项目:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **许可证:** MIT
+**作者:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **版本:** 0.6.0 | **项目:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **许可证:** MIT

-这是一个用于 [OpenWebUI](https://github.com/open-webui/open-webui) 的高级 Pipe 函数，允许你直接在 OpenWebUI 中使用 GitHub Copilot 模型（如 `gpt-5`, `gpt-5-mini`, `claude-sonnet-4.5`）。它基于官方 [GitHub Copilot SDK for Python](https://github.com/github/copilot-sdk) 构建，提供了原生级的集成体验。
+这是一个用于 [OpenWebUI](https://github.com/open-webui/open-webui) 的高级 Pipe 函数，深度集成了 **GitHub Copilot SDK**。它不仅支持 **GitHub Copilot 官方模型**（如 `gpt-5.2-codex`, `claude-sonnet-4.5`, `gemini-3-pro`, `gpt-5-mini`），还支持 **BYOK (自带 Key)** 模式对接自定义服务商（OpenAI, Anthropic），并具备**严格的用户与会话级工作区隔离**能力，提供统一且安全的 Agent 交互体验。

-## 🚀 最新特性 (v0.3.0) - “统一生态”的力量
+> [!IMPORTANT]
+> **核心伴侣组件**
+> 如需启用文件处理与数据分析能力，请务必安装 [GitHub Copilot SDK Files Filter](https://openwebui.com/posts/403a62ee-a596-45e7-be65-fab9cc249dd6)。

-* **🔌 零配置工具桥接 (Unified Tool Bridge)**: 自动将您现有的 OpenWebUI Functions (工具) 转换为 Copilot 兼容工具。**Copilot 现在可以无缝调用您手头所有的 WebUI 工具！**
-* **🔗 动态 MCP 自动发现**: 直接联动 OpenWebUI **管理面板 -> 连接**。无需编写任何配置文件，即插即用，瞬间扩展 Copilot 能力边界。
-* **⚡ 高性能异步引擎**: 异步 CLI 更新检查与高度优化的事件驱动流式处理，确保对话毫秒级响应。
-* **🛡️ 卓越的兼容性**: 独创的动态 Pydantic 模型生成技术，确保复杂工具参数在 Copilot 端也能得到精准验证。
+> [!TIP]
+> **BYOK 模式无需订阅**
+> 如果您使用自带的 API Key (BYOK 模式对接 OpenAI/Anthropic)，**您不需要 GitHub Copilot 官方订阅**。只有在访问 GitHub 官方模型时才需要订阅。

-## ✨ 核心能力
+---

-* **🌉 强大的生态桥接**: 首个且唯一完美打通 **OpenWebUI Tools** 与 **GitHub Copilot SDK** 的插件。
-* **🚀 官方原生产体验**: 基于官方 Python SDK 构建，提供最稳定、最纯正的 Copilot 交互体验。
-* **🌊 深度推理展示**: 完整支持模型思考过程 (Thinking Process) 的流式渲染。
-* **🖼️ 智能多模态**: 支持图像识别与附件上传，让 Copilot 拥有视觉能力。
-* **🛠️ 极简部署流程**: 自动检测环境、自动下载 CLI、自动管理依赖，全自动化开箱即用。
-* **🔑 安全认证体系**: 完美支持 OAuth 授权与 PAT 模式，兼顾便捷与安全性。
+## ✨ 0.6.0 更新内容 (What's New)

-## 📦 安装与使用
+- **👥 多用户与会话管理**: 采用 `user_id/chat_id` 的物理隔离架构，确保资源独立与稳健管理。
+- **🤖 赋予 Agent 文件自主权**: 自动将上传的文件同步至物理工作区，支持 Python 直接分析 Excel/CSV。
+- **🔧 OpenAPI & 外部工具修复**: 完美支持通过 OpenAPI 服务器挂载的工具调用。
+- **📊 计费与成本控制**: 增强的**计费倍率限制** (`MAX_MULTIPLIER`，例如设为 0 即仅限免费模型) 和**模型关键词过滤** (`EXCLUDE_KEYWORDS`)，实现更精准的成本管控。
+- **🧠 数据库持久化 TODO**: 任务进度跨会话保存，Agent 拥有更持久的任务 memory。

-### 1. 导入函数
+---

-1. 打开 OpenWebUI。
-2. 进入 **Workspace** -> **Functions**。
-3. 点击 **+** (创建函数)。
-4. 将 `github_copilot_sdk_cn.py` 的内容完整粘贴进去。
-5. 保存。
+## ✨ 核心能力 (Key Capabilities)

-### 2. 配置 Valves (设置)
+- **🔑 灵活鉴权与 BYOK**: 支持 GitHub Copilot 官方订阅 (PAT) 或自带 Key (OpenAI/Anthropic)。
+- **🔌 通用工具协议**: 原生支持 **MCP (Model Context Protocol)**、OpenAPI 以及 OpenWebUI 内置工具。
+- **🛡️ 物理级工作区隔离**: 强制执行严格的用户特定沙箱，确保数据隐私与文件安全。
+- **♾️ 无限会话管理**: 智能上下文窗口管理与自动压缩算法，支持无限时长的对话交互。
+- **🧠 深度数据库集成**: 实时持久化 TOD·O 列表到 UI 进度条。
+- **🌊 深度推理展示**: 完整支持模型思考过程 (Thinking Process) 的流式渲染。
+- **🖼️ 智能多模态**: 完整支持图像识别与附件上传分析。
+- **⚡ 交互式伪影 (Artifacts)**: 自动渲染 Agent 生成的 HTML/JS 应用程序，直接在聊天界面交互。

-在函数列表中找到 "GitHub Copilot"，点击 **⚙️ (Valves)** 图标进行配置：
+---

-| 参数 | 说明 | 默认值 |
+## ⚙️ 核心配置参数 (Valves)
+
+### 1. 管理员配置 (基础设置)
+
+管理员可在函数设置中定义全局默认行为。
+
+| 参数 | 默认值 | 说明 |
 | :--- | :--- | :--- |
-| **GH_TOKEN** | **(必填)** GitHub 访问令牌 (PAT 或 OAuth Token)。用于聊天。 | - |
-| **DEBUG** | 是否开启调试日志（输出到浏览器控制台）。 | `False` |
-| **LOG_LEVEL** | Copilot CLI 日志级别: none, error, warning, info, debug, all。 | `error` |
-| **SHOW_THINKING** | 是否显示模型推理/思考过程（需开启流式 + 模型支持）。 | `True` |
-| **COPILOT_CLI_VERSION** | 指定安装/强制使用的 Copilot CLI 版本。 | `0.0.405` |
-| **EXCLUDE_KEYWORDS** | 排除包含这些关键词的模型（逗号分隔）。 | - |
-| **WORKSPACE_DIR** | 文件操作的受限工作区目录。 | - |
-| **INFINITE_SESSION** | 启用无限会话（自动上下文压缩）。 | `True` |
-| **COMPACTION_THRESHOLD** | 后台压缩阈值 (0.0-1.0)。 | `0.8` |
-| **BUFFER_THRESHOLD** | 缓冲区耗尽阈值 (0.0-1.0)。 | `0.95` |
-| **TIMEOUT** | 每个流式分块超时（秒）。 | `300` |
-| **CUSTOM_ENV_VARS** | 自定义环境变量 (JSON 格式)。 | - |
-| **REASONING_EFFORT** | 推理强度级别: low, medium, high. `xhigh` 仅部分模型支持。 | `medium` |
-| **ENFORCE_FORMATTING** | 在系统提示词中添加格式化指导。 | `True` |
-| **ENABLE_MCP_SERVER** | 启用直接 MCP 客户端连接 (建议)。 | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | 启用 OpenWebUI 工具 (包括自定义和服务器工具)。 | `True` |
+| `GH_TOKEN` | `""` | 全局 GitHub Token (需具备 'Copilot Requests' 权限)。 |
+| `ENABLE_OPENWEBUI_TOOLS` | `True` | 启用 OpenWebUI 工具 (包括定义工具和内置工具)。 |
+| `ENABLE_OPENAPI_SERVER` | `True` | 启用 OpenAPI 工具服务器连接。 |
+| `ENABLE_MCP_SERVER` | `True` | 启用直接 MCP 客户端连接 (推荐)。 |
+| `REASONING_EFFORT` | `medium` | 推理强度：low, medium, high。 |
+| `SHOW_THINKING` | `True` | 显示模型推理/思考过程。 |
+| `INFINITE_SESSION` | `True` | 启用无限会话 (自动上下文压缩)。 |
+| `MAX_MULTIPLIER` | `1.0` | 最大允许的模型计费倍率 (0x 为仅限免费模型)。 |
+| `EXCLUDE_KEYWORDS` | `""` | 排除包含这些关键字的模型 (逗号分隔)。 |
+| `TIMEOUT` | `300` | 每个流数据块的超时时间 (秒)。 |
+| `BYOK_TYPE` | `openai` | BYOK 服务商类型：`openai`, `anthropic`。 |
+| `BYOK_BASE_URL` | `""` | BYOK 基础 URL (例如: <https://api.openai.com/v1>)。 |
+| `BYOK_MODELS` | `""` | BYOK 模型列表 (逗号分隔)。留空则从 API 获取。 |
+| `CUSTOM_ENV_VARS` | `""` | 自定义环境变量 (JSON 格式)。 |
+| `DEBUG` | `False` | 开启此项以在前端控制台输出详细调试日志。 |

-#### 用户 Valves（按用户覆盖）
+### 2. 用户配置 (个人覆盖)

-以下设置可按用户单独配置（覆盖全局 Valves）：
+普通用户可在各自的个人设置中根据需要覆盖以下参数。

-| 参数 | 说明 | 默认值 |
-| :--- | :--- | :--- |
-| **GH_TOKEN** | 个人 GitHub Token（覆盖全局设置）。 | - |
-| **REASONING_EFFORT** | 推理强度级别（low/medium/high/xhigh）。 | - |
-| **DEBUG** | 是否启用技术调试日志。 | `False` |
-| **SHOW_THINKING** | 是否显示思考过程。 | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | 启用 OpenWebUI 工具（覆盖全局设置）。 | `True` |
-| **ENABLE_MCP_SERVER** | 启用动态 MCP 服务器加载（覆盖全局设置）。 | `True` |
-| **ENFORCE_FORMATTING** | 强制启用格式化指导（覆盖全局设置）。 | `True` |
+| 参数 | 说明 |
+| :--- | :--- |
+| `GH_TOKEN` | 使用个人的 GitHub Token。 |
+| `REASONING_EFFORT` | 个人偏好的推理强度。 |
+| `SHOW_THINKING` | 显示模型推理/思考过程。 |
+| `MAX_MULTIPLIER` | 最大允许的模型计费倍率覆盖。 |
+| `EXCLUDE_KEYWORDS` | 排除包含这些关键字的模型。 |
+| `BYOK_API_KEY` | 使用个人的 OpenAI/Anthropic API Key。 |

-### 3. 获取 Token
+---

-要使用 GitHub Copilot，您需要一个具有适当权限的 GitHub 个人访问令牌 (PAT)。
+## 🎯 典型应用场景 (Use Cases)

-**获取步骤：**
+- **📁 全自主仓库维护**: Agent 在隔离工作区内自动分析代码、运行测试并应用补丁。
+- **📊 深度财务数据审计**: 直接通过 Python 加载 Excel/CSV 原始数据（绕过 RAG），生成图表并实时预览。
+- **📝 长任务项目管理**: 自动拆解复杂任务并持久化 TOD·O 进度，跨会话跟踪执行状态。

-1. 访问 [GitHub 令牌设置](https://github.com/settings/tokens?type=beta)。
-2. 点击 **Generate new token (fine-grained)**。
-3. **Repository access**: 选择 **Public Repositories** (最简单) 或 **All repositories**。
-4. **Permissions**:
-    * 如果您选择了 **All repositories**，则必须点击 **Account permissions**。
-    * 找到 **Copilot Requests**，选择 **Access**。
-5. 生成并复制令牌。
+---

-## 📋 依赖说明
+## ⭐ 支持与交流 (Support)

-该 Pipe 会自动尝试安装以下依赖（如果环境中缺失）：
+如果这个插件对您有所帮助，请在 [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) 项目上点个 **Star** 💫，这是对我最大的鼓励。

-* `github-copilot-sdk` (Python 包)
-* `github-copilot-cli` (二进制文件，通过官方脚本安装)
+---

-## 故障排除 (Troubleshooting) ❓
+## 🚀 安装与配置 (Installation)

-* **图片及多模态使用说明**：
-  * 确保 `MODEL_ID` 是支持多模态的模型。
-* **看不到思考过程**：
-  * 确认已开启**流式输出**，且所选模型支持推理输出。
+### 1) 导入函数

-## 📄 许可证
+1. 打开 OpenWebUI，前往 **工作区** -> **函数**。
+2. 点击 **+** (创建函数)，完整粘贴 `github_copilot_sdk_cn.py` 的内容。
+3. 点击保存并确保已启用。

-MIT
+### 2) 获取 Token (Get Token)
+
+1. 访问 [GitHub Token 设置](https://github.com/settings/tokens?type=beta)。
+2. 创建 **Fine-grained token**，授予 **Account permissions** -> **Copilot Requests** 访问权限。
+3. 将生成的 Token 填入插件的 `GH_TOKEN` 配置项中。
+
+---
+
+## 📋 常见问题与依赖 (Troubleshooting)
+
+- **Agent 无法识别文件？**: 请确保已安装并启用了 Files Filter 插件，否则原始文件会被 RAG 干扰。
+- **看不到 TODO 进度条？**: 进度条仅在 Agent 使用 `update_todo` 工具（通常是处理复杂任务）时出现。
+- **依赖安装**: 本管道会自动尝试安装 `github-copilot-sdk` (Python 包) 和 `github-copilot-cli` (官方二进制)。
+
+---
+
+## 更新日志 (Changelog)
+
+完整历史记录请见 GitHub: [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)
--- a/docs/plugins/pipes/index.md
+++ b/docs/plugins/pipes/index.md
@@ -15,7 +15,7 @@ Pipes allow you to:

 ## Available Pipe Plugins

- [GitHub Copilot SDK](github-copilot-sdk.md) (v0.3.0) - Official GitHub Copilot SDK integration. Features **zero-config OpenWebUI Tool Bridge** and **dynamic MCP discovery**. Supports streaming, multimodal, and infinite sessions.
+- [GitHub Copilot SDK](github-copilot-sdk.md) (v0.6.0) - Official GitHub Copilot SDK integration. Features **Workspace Isolation**, **Database Persistence**, **Zero-config OpenWebUI Tool Bridge**, **BYOK** support, and **dynamic MCP discovery**. Supports streaming, multimodal, and infinite sessions.

 ---

--- a/docs/plugins/pipes/index.zh.md
+++ b/docs/plugins/pipes/index.zh.md
@@ -15,7 +15,7 @@ Pipes 可以用于：

 ## 可用的 Pipe 插件

- [GitHub Copilot SDK](github-copilot-sdk.zh.md) (v0.3.0) - GitHub Copilot SDK 官方集成。**零配置工具桥接**与**动态 MCP 发现**。支持流式输出、多模态及无限会话。
+- [GitHub Copilot SDK](github-copilot-sdk.zh.md) (v0.6.0) - GitHub Copilot SDK 官方集成。具备**工作区安全隔离**、**数据库持久化**、**零配置工具桥接**与**BYOK (自带 Key) 支持**。支持流式输出、打字机思考过程及无限会话。[查看深度架构解析](github-copilot-sdk-deep-dive.zh.md)。

 ---

--- a/plugins/actions/export_to_docx/README.md
+++ b/plugins/actions/export_to_docx/README.md
@@ -1,14 +1,15 @@
 # 📝 Export to Word (Enhanced)

-**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.4.3 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **License:** MIT
+**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.4.4 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **License:** MIT

 Export conversation to Word (.docx) with **syntax highlighting**, **native math equations**, **Mermaid diagrams**, **citations**, and **enhanced table formatting**.

-## 🔥 What's New in v0.4.3
+## 🔥 What's New in v0.4.4

- ✨ **S3 Object Storage Support**: Direct access to images stored in S3/MinIO via boto3, bypassing API layer for faster exports.
- 🔧 **Multi-level File Fallback**: 6-level fallback mechanism for file retrieval (DB → S3 → Local → URL → API → Attributes).
- 🛡️ **Improved Error Handling**: Better logging and error messages for file retrieval failures.
+- 🧹 **Content Cleanup**: Enhanced stripping of `<details>` blocks (often used for tool calls/thinking process) to ensure a clean final document.
+- 📄 **Standard Document Formatting**: Applied professional document formatting standards for titles and headings (centered title, bold, optimized font sizes and spacing), including GB/T compliance for Chinese content.
+- 🔠 **Font Rendering Fix**: Fixed an issue where CJK characters would fallback to MS Gothic in Word; now correctly uses the configured Asian font (e.g., SimSun).
+- ⚙️ **Title Alignment**: Added `TITLE_ALIGNMENT` valve to configure document title alignment (left, center, right).

 ## ✨ Key Features

--- a/plugins/actions/export_to_docx/README_CN.md
+++ b/plugins/actions/export_to_docx/README_CN.md
@@ -1,14 +1,15 @@
 # 📝 导出为 Word (增强版)

-**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.4.3 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **许可证:** MIT
+**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.4.4 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **许可证:** MIT

 将对话导出为 Word (.docx)，支持**代码语法高亮**、**原生数学公式**、**Mermaid 图表**、**引用参考**和**增强表格格式**。

-## 🔥 v0.4.3 更新内容
+## 🔥 v0.4.4 更新内容

- ✨ **S3 对象存储支持**: 通过 boto3 直连 S3/MinIO，绕过 API 层，导出速度更快。
- 🔧 **多级文件回退**: 6 级文件获取机制（数据库 → S3 → 本地 → URL → API → 属性）。
- 🛡️ **错误处理优化**: 更完善的日志记录和错误提示，便于调试文件访问问题。
+- 🧹 **内容清理加强**: 增强了对 `<details>` 块（通常包含工具调用或思考过程）的清理，确保最终文档整洁。
+- 📄 **文档格式标准化**: 采用了专业的文档排版标准（兼容中文 GB/T 规范），标题居中加粗，各级标题使用标准字号和间距。
+- 🔠 **字体渲染修复**: 修复了 CJK 字符在 Word 中回退到 MS Gothic 的问题；现在正确使用配置的中文字体（例如宋体）。
+- ⚙️ **标题对齐配置**: 新增 `标题对齐方式` Valve，支持配置文档标题的对齐方式（左对齐、居中、右对齐）。

 ## ✨ 核心特性

--- a/plugins/actions/export_to_docx/export_to_word.py
+++ b/plugins/actions/export_to_docx/export_to_word.py
@@ -1,9 +1,9 @@
 """
-title: Export to Word (Enhanced)
+title: Export to Word Enhanced
 author: Fu-Jie
 author_url: https://github.com/Fu-Jie/awesome-openwebui
 funding_url: https://github.com/open-webui
-version: 0.4.3
+version: 0.4.4
 openwebui_id: fca6a315-2a45-42cc-8c96-55cbc85f87f2
 icon_url: data:image/svg+xml;base64,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
 requirements: python-docx, Pygments, latex2mathml, mathml2omml
@@ -101,9 +101,8 @@ _TRANSPARENT_1PX_PNG = base64.b64decode(
 _ASVG_NS = "http://schemas.microsoft.com/office/drawing/2016/SVG/main"
 nsmap.setdefault("asvg", _ASVG_NS)

-_REASONING_DETAILS_RE = re.compile(
-    r"<details\b[^>]*\btype\s*=\s*(?:\"reasoning\"|'reasoning'|reasoning)[^>]*>.*?</details\s*>",
-    re.IGNORECASE | re.DOTALL,
+_ALL_DETAILS_RE = re.compile(
+    r"<details\b[^>]*>.*?</details\s*>", re.IGNORECASE | re.DOTALL
 )
 _THINK_RE = re.compile(r"<think\b[^>]*>.*?</think\s*>", re.IGNORECASE | re.DOTALL)
 _ANALYSIS_RE = re.compile(
@@ -178,6 +177,12 @@ class Action:
            description="Font for code blocks and inline code (e.g., 'Consolas', 'Courier New', 'Monaco')",
        )

+        # Title alignment
+        TITLE_ALIGNMENT: str = Field(
+            default="center",
+            description="Title alignment: 'left', 'center', or 'right'",
+        )
+
        # Table styling
        TABLE_HEADER_COLOR: str = Field(
            default="F2F2F2",
@@ -242,60 +247,60 @@ class Action:
        )

    class UserValves(BaseModel):
-        TITLE_SOURCE: str = Field(
-            default="chat_title",
+        TITLE_SOURCE: Optional[str] = Field(
+            default=None,
            description="Title Source: 'chat_title' (Chat Title), 'ai_generated' (AI Generated), 'markdown_title' (Markdown Title)",
        )
-        UI_LANGUAGE: str = Field(
-            default="en",
+        UI_LANGUAGE: Optional[str] = Field(
+            default=None,
            description="UI language for export messages. Options: 'en' (English), 'zh' (Chinese)",
        )
-        FONT_LATIN: str = Field(
-            default="Times New Roman",
+        FONT_LATIN: Optional[str] = Field(
+            default=None,
            description="Font for Latin characters (e.g., 'Times New Roman', 'Calibri', 'Arial')",
        )
-        FONT_ASIAN: str = Field(
-            default="SimSun",
+        FONT_ASIAN: Optional[str] = Field(
+            default=None,
            description="Font for Asian characters (e.g., 'SimSun', 'Microsoft YaHei', 'PingFang SC')",
        )
-        FONT_CODE: str = Field(
-            default="Consolas",
+        FONT_CODE: Optional[str] = Field(
+            default=None,
            description="Font for code blocks and inline code (e.g., 'Consolas', 'Courier New', 'Monaco')",
        )
-        TABLE_HEADER_COLOR: str = Field(
-            default="F2F2F2",
+        TABLE_HEADER_COLOR: Optional[str] = Field(
+            default=None,
            description="Table header background color (hex, without #)",
        )
-        TABLE_ZEBRA_COLOR: str = Field(
-            default="FBFBFB",
+        TABLE_ZEBRA_COLOR: Optional[str] = Field(
+            default=None,
            description="Table zebra stripe background color for alternate rows (hex, without #)",
        )
-        MERMAID_PNG_SCALE: float = Field(
-            default=3.0,
+        MERMAID_PNG_SCALE: Optional[float] = Field(
+            default=None,
            description="PNG render resolution multiplier (higher = clearer, larger file)",
        )
-        MERMAID_DISPLAY_SCALE: float = Field(
-            default=1.0,
+        MERMAID_DISPLAY_SCALE: Optional[float] = Field(
+            default=None,
            description="Diagram width relative to available page width (<=1 recommended)",
        )
-        MERMAID_OPTIMIZE_LAYOUT: bool = Field(
-            default=False,
+        MERMAID_OPTIMIZE_LAYOUT: Optional[bool] = Field(
+            default=None,
            description="Optimize Mermaid layout: convert LR to TD for graph/flowchart",
        )
-        MERMAID_BACKGROUND: str = Field(
-            default="",
+        MERMAID_BACKGROUND: Optional[str] = Field(
+            default=None,
            description="Mermaid background color. Empty = transparent (recommended for Word dark mode). Used only for optional PNG fill.",
        )
-        MERMAID_CAPTIONS_ENABLE: bool = Field(
-            default=True,
+        MERMAID_CAPTIONS_ENABLE: Optional[bool] = Field(
+            default=None,
            description="Add figure captions under Mermaid images/charts",
        )
-        MATH_ENABLE: bool = Field(
-            default=True,
+        MATH_ENABLE: Optional[bool] = Field(
+            default=None,
            description="Enable LaTeX math block conversion (\\\\[...\\\\] and $$...$$) into Word equations",
        )
-        MATH_INLINE_DOLLAR_ENABLE: bool = Field(
-            default=True,
+        MATH_INLINE_DOLLAR_ENABLE: Optional[bool] = Field(
+            default=None,
            description="Enable inline $...$ math conversion into Word equations (conservative parsing to reduce false positives)",
        )

@@ -449,13 +454,21 @@ class Action:
            user_id = __user__.get("id", "unknown_user")

        # Apply UserValves if present
-        if __user__ and "valves" in __user__:
-            # Update self.valves with user-specific values
-            # Note: This assumes per-request instantiation or that we are okay with modifying the singleton.
-            # Given the plugin architecture, we'll update it for this execution.
-            for key, value in __user__["valves"].model_dump().items():
-                if hasattr(self.valves, key):
-                    setattr(self.valves, key, value)
+        if __user__:
+            # Robustly parse UserValves whether it's a dict or Pydantic model
+            raw_valves = __user__.get("valves", {})
+            if isinstance(raw_valves, self.UserValves):
+                user_valves = raw_valves
+            elif isinstance(raw_valves, dict):
+                user_valves = self.UserValves(**raw_valves)
+            else:
+                user_valves = None
+
+            if user_valves:
+                for key, value in user_valves.model_dump(exclude_unset=True).items():
+                    # Only override if the value is not None (and explicitly set)
+                    if hasattr(self.valves, key) and value is not None:
+                        setattr(self.valves, key, value)

        # Get user language from Valves configuration
        self._user_lang = self._get_lang_key(self.valves.UI_LANGUAGE)
@@ -492,6 +505,37 @@ class Action:
            try:
                message_content = last_assistant_message["content"]
                if isinstance(message_content, str):
+                    if __event_emitter__ and self.valves.SHOW_DEBUG_LOG:
+                        debug_data = {}
+                        for name, regex in [
+                            ("Details Block", _ALL_DETAILS_RE),
+                            ("Think Block", _THINK_RE),
+                            ("Analysis Block", _ANALYSIS_RE),
+                        ]:
+                            matches = regex.findall(message_content)
+                            if matches:
+                                debug_data[name] = [
+                                    (m[:200] + "...") if len(m) > 200 else m
+                                    for m in matches
+                                ]
+                        if debug_data:
+                            await self._emit_debug_log(
+                                __event_emitter__,
+                                "Context Stripping Analysis",
+                                debug_data,
+                            )
+
+                        # Log font configuration
+                        await self._emit_debug_log(
+                            __event_emitter__,
+                            "Font Configuration",
+                            {
+                                "Latin Font": self.valves.FONT_LATIN,
+                                "Asian Font": self.valves.FONT_ASIAN,
+                                "Code Font": self.valves.FONT_CODE,
+                            },
+                        )
+
                    message_content = self._strip_reasoning_blocks(message_content)

                if not message_content or not message_content.strip():
@@ -1107,30 +1151,7 @@ class Action:
        if not isinstance(name, str):
            return ""

-        def _is_emoji_codepoint(codepoint: int) -> bool:
-            # Common emoji ranges + flag regional indicators.
-            return (
-                0x1F000 <= codepoint <= 0x1FAFF
-                or 0x1F1E6 <= codepoint <= 0x1F1FF
-                or 0x2600 <= codepoint <= 0x26FF
-                or 0x2700 <= codepoint <= 0x27BF
-                or 0x2300 <= codepoint <= 0x23FF
-                or 0x2B00 <= codepoint <= 0x2BFF
-            )
-
-        def _is_emoji_modifier(codepoint: int) -> bool:
-            # VS15/VS16, ZWJ, keycap, skin tones, and tag characters used in some emoji sequences.
-            return (
-                codepoint in (0x200D, 0xFE0E, 0xFE0F, 0x20E3)
-                or 0x1F3FB <= codepoint <= 0x1F3FF
-                or 0xE0020 <= codepoint <= 0xE007F
-            )
-
-        without_emoji = "".join(
-            ch
-            for ch in name
-            if not (_is_emoji_codepoint(ord(ch)) or _is_emoji_modifier(ord(ch)))
-        )
+        without_emoji = self._remove_emojis(name)
        cleaned = re.sub(r'[\\/*?:"<>|]', "", without_emoji)
        cleaned = re.sub(r"\s+", " ", cleaned).strip().strip(".")
        return cleaned[:50].strip()
@@ -1498,7 +1519,10 @@ class Action:

            # If there is no h1 in content, prepend chat title as h1 when provided
            if top_heading and not has_h1:
-                self.add_heading(doc, top_heading, 1)
+                # Remove emojis from title for a professional look
+                clean_title = self._remove_emojis(top_heading)
+                # Use Title style (level 0) for the main document title
+                self.add_heading(doc, clean_title, 0)

            lines = markdown_text.split("\n")
            i = 0
@@ -1758,7 +1782,7 @@ class Action:
        cur = text
        for _ in range(10):
            prev = cur
-            cur = _REASONING_DETAILS_RE.sub("", cur)
+            cur = _ALL_DETAILS_RE.sub("", cur)
            cur = _THINK_RE.sub("", cur)
            cur = _ANALYSIS_RE.sub("", cur)
            if cur == prev:
@@ -2242,14 +2266,155 @@ class Action:
        font = style.font
        font.name = self.valves.FONT_LATIN
        font.size = Pt(11)
-        # Set Asian font
-        style._element.rPr.rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
+
+        # Ensure rPr element exists
+        rPr = style._element.get_or_add_rPr()
+        rFonts = rPr.get_or_add_rFonts()
+
+        # Set Latin and Asian fonts explicitly
+        rFonts.set(qn("w:ascii"), self.valves.FONT_LATIN)
+        rFonts.set(qn("w:hAnsi"), self.valves.FONT_LATIN)
+        rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
+
+        # Set language to zh-CN to prevent MS Gothic fallback (Japanese font)
+        # Even for English interface, we want to prioritize Chinese glyphs over Japanese for CJK
+        lang = rPr.find(qn("w:lang"))
+        if lang is None:
+            lang = OxmlElement("w:lang")
+            rPr.append(lang)
+        lang.set(qn("w:val"), "en-US")
+        lang.set(qn("w:eastAsia"), "zh-CN")
+
+        logger.info(
+            f"[Font Config] Latin: {self.valves.FONT_LATIN}, Asian: {self.valves.FONT_ASIAN}"
+        )

        # Set paragraph format
        paragraph_format = style.paragraph_format
        paragraph_format.line_spacing_rule = WD_LINE_SPACING.ONE_POINT_FIVE
        paragraph_format.space_after = Pt(6)

+        # Configure Title style (used for document title)
+        # Standard format: 22pt (二号), bold, centered, 24pt space after
+        if "Title" in doc.styles:
+            title_style = doc.styles["Title"]
+            title_font = title_style.font
+            title_font.name = self.valves.FONT_LATIN
+            title_font.size = Pt(22)  # 二号字体
+            title_font.bold = True
+            title_font.color.rgb = RGBColor(0, 0, 0)
+
+            # Set paragraph format: alignment based on configuration
+            title_pf = title_style.paragraph_format
+            alignment_map = {
+                "left": WD_ALIGN_PARAGRAPH.LEFT,
+                "center": WD_ALIGN_PARAGRAPH.CENTER,
+                "right": WD_ALIGN_PARAGRAPH.RIGHT,
+            }
+            title_pf.alignment = alignment_map.get(
+                self.valves.TITLE_ALIGNMENT.lower(), WD_ALIGN_PARAGRAPH.CENTER
+            )
+            title_pf.space_before = Pt(0)
+            title_pf.space_after = Pt(24)
+
+            t_rPr = title_style._element.get_or_add_rPr()
+            t_rFonts = t_rPr.get_or_add_rFonts()
+            t_rFonts.set(qn("w:ascii"), self.valves.FONT_LATIN)
+            t_rFonts.set(qn("w:hAnsi"), self.valves.FONT_LATIN)
+            t_rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
+
+            # Set language to zh-CN
+            t_lang = t_rPr.find(qn("w:lang"))
+            if t_lang is None:
+                t_lang = OxmlElement("w:lang")
+                t_rPr.append(t_lang)
+            t_lang.set(qn("w:val"), "en-US")
+            t_lang.set(qn("w:eastAsia"), "zh-CN")
+
+        # Standard heading sizes based on Chinese document standards:
+        # Heading 1: 16pt (三号), bold, space before 24pt, space after 12pt
+        # Heading 2: 15pt (小三), bold, space before 18pt, space after 6pt
+        # Heading 3: 14pt (四号), bold, space before 12pt, space after 6pt
+        # Heading 4-9: 12pt (小四), bold, gradually reduced spacing
+        heading_formats = {
+            1: {"size": 16, "space_before": 24, "space_after": 12},
+            2: {"size": 15, "space_before": 18, "space_after": 6},
+            3: {"size": 14, "space_before": 12, "space_after": 6},
+            4: {"size": 12, "space_before": 12, "space_after": 6},
+            5: {"size": 12, "space_before": 6, "space_after": 6},
+            6: {"size": 11, "space_before": 6, "space_after": 3},
+            7: {"size": 11, "space_before": 6, "space_after": 3},
+            8: {"size": 10.5, "space_before": 6, "space_after": 3},
+            9: {"size": 10.5, "space_before": 6, "space_after": 3},
+        }
+
+        # Apply font settings to Heading 1-9
+        for i in range(1, 10):
+            style_id = f"Heading {i}"
+            if style_id in doc.styles:
+                heading_style = doc.styles[style_id]
+                heading_font = heading_style.font
+                heading_font.name = self.valves.FONT_LATIN
+                heading_font.color.rgb = RGBColor(0, 0, 0)
+
+                # Apply standard formatting
+                fmt = heading_formats.get(
+                    i, {"size": 11, "space_before": 6, "space_after": 3}
+                )
+                heading_font.size = Pt(fmt["size"])
+                heading_font.bold = True
+
+                heading_pf = heading_style.paragraph_format
+                heading_pf.space_before = Pt(fmt["space_before"])
+                heading_pf.space_after = Pt(fmt["space_after"])
+
+                # Ensure rPr exists
+                h_rPr = heading_style._element.get_or_add_rPr()
+                h_rFonts = h_rPr.get_or_add_rFonts()
+
+                # Set fonts
+                h_rFonts.set(qn("w:ascii"), self.valves.FONT_LATIN)
+                h_rFonts.set(qn("w:hAnsi"), self.valves.FONT_LATIN)
+                h_rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
+
+                # Set language to zh-CN
+                h_lang = h_rPr.find(qn("w:lang"))
+                if h_lang is None:
+                    h_lang = OxmlElement("w:lang")
+                    h_rPr.append(h_lang)
+                h_lang.set(qn("w:val"), "en-US")
+                h_lang.set(qn("w:eastAsia"), "zh-CN")
+
+    def _remove_emojis(self, text: str) -> str:
+        """Remove emojis from text based on unicode ranges."""
+        if not isinstance(text, str):
+            return ""
+
+        def _is_emoji_codepoint(codepoint: int) -> bool:
+            # Common emoji ranges + flag regional indicators.
+            return (
+                0x1F000 <= codepoint <= 0x1FAFF
+                or 0x1F1E6 <= codepoint <= 0x1F1FF
+                or 0x2600 <= codepoint <= 0x26FF
+                or 0x2700 <= codepoint <= 0x27BF
+                or 0x2300 <= codepoint <= 0x23FF
+                or 0x2B00 <= codepoint <= 0x2BFF
+            )
+
+        def _is_emoji_modifier(codepoint: int) -> bool:
+            # VS15/VS16, ZWJ, keycap, skin tones, and tag characters used in some emoji sequences.
+            return (
+                codepoint in (0x200D, 0xFE0E, 0xFE0F, 0x20E3)
+                or 0x1F3FB <= codepoint <= 0x1F3FF
+                or 0xE0020 <= codepoint <= 0xE007F
+            )
+
+        return "".join(
+            ch
+            for ch in text
+            if not (_is_emoji_codepoint(ord(ch)) or _is_emoji_modifier(ord(ch)))
+        )
+
    def add_heading(self, doc: Document, text: str, level: int):
        """Add heading"""
        # Word heading levels start from 0, Markdown from 1
@@ -2285,6 +2450,12 @@ class Action:
        if strike:
            run.font.strike = True

+        # Explicitly set East Asian font to prevent MS Gothic fallback
+        # Word may not inherit w:eastAsia from style, causing Japanese font fallback for CJK
+        rPr = run._element.get_or_add_rPr()
+        rFonts = rPr.get_or_add_rFonts()
+        rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
+
    def _add_inline_code(self, paragraph, s: str):
        if s == "":
            return
@@ -2684,7 +2855,11 @@ class Action:
    ):
        u = self._normalize_url(url)
        if not u:
-            paragraph.add_run(display_text or text)
+            run = paragraph.add_run(display_text or text)
+            # Set East Asian font to prevent MS Gothic fallback
+            rPr = run._element.get_or_add_rPr()
+            rFonts = rPr.get_or_add_rFonts()
+            rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
            return

        part = getattr(paragraph, "part", None)
@@ -2693,6 +2868,10 @@ class Action:
            run = paragraph.add_run(display_text or text)
            run.font.color.rgb = RGBColor(0, 0, 255)
            run.font.underline = True
+            # Set East Asian font to prevent MS Gothic fallback
+            rPr = run._element.get_or_add_rPr()
+            rFonts = rPr.get_or_add_rFonts()
+            rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
            return

        r_id = part.relate_to(u, RT.HYPERLINK, is_external=True)
@@ -2706,6 +2885,11 @@ class Action:
        rStyle.set(qn("w:val"), "Hyperlink")
        rPr.append(rStyle)

+        # Set East Asian font to prevent MS Gothic fallback
+        rFonts = OxmlElement("w:rFonts")
+        rFonts.set(qn("w:eastAsia"), self.valves.FONT_ASIAN)
+        rPr.append(rFonts)
+
        color = OxmlElement("w:color")
        color.set(qn("w:val"), "0000FF")
        rPr.append(color)
--- a/plugins/actions/export_to_docx/export_to_word_cn.py
+++ b/plugins/actions/export_to_docx/export_to_word_cn.py
@@ -1,9 +1,9 @@
 """
-title: 导出为 Word (增强版)
+title: 导出为Word增强版
 author: Fu-Jie
 author_url: https://github.com/Fu-Jie/awesome-openwebui
 funding_url: https://github.com/open-webui
-version: 0.4.3
+version: 0.4.4
 openwebui_id: 8a6306c0-d005-4e46-aaae-8db3532c9ed5
 icon_url: data:image/svg+xml;base64,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
 requirements: python-docx, Pygments, latex2mathml, mathml2omml
@@ -101,9 +101,8 @@ _TRANSPARENT_1PX_PNG = base64.b64decode(
 _ASVG_NS = "http://schemas.microsoft.com/office/drawing/2016/SVG/main"
 nsmap.setdefault("asvg", _ASVG_NS)

-_REASONING_DETAILS_RE = re.compile(
-    r"<details\b[^>]*\btype\s*=\s*(?:\"reasoning\"|'reasoning'|reasoning)[^>]*>.*?</details\s*>",
-    re.IGNORECASE | re.DOTALL,
+_ALL_DETAILS_RE = re.compile(
+    r"<details\b[^>]*>.*?</details\s*>", re.IGNORECASE | re.DOTALL
 )
 _THINK_RE = re.compile(r"<think\b[^>]*>.*?</think\s*>", re.IGNORECASE | re.DOTALL)
 _ANALYSIS_RE = re.compile(
@@ -178,6 +177,12 @@ class Action:
            description="Font for code blocks and inline code (e.g., 'Consolas', 'Courier New', 'Monaco')",
        )

+        # Title alignment
+        标题对齐方式: str = Field(
+            default="center",
+            description="标题对齐方式: 'left' (左对齐), 'center' (居中), 或 'right' (右对齐)",
+        )
+
        # Table styling
        表头背景色: str = Field(
            default="F2F2F2",
@@ -242,60 +247,60 @@ class Action:
        )

    class UserValves(BaseModel):
-        文档标题来源: str = Field(
-            default="chat_title",
+        文档标题来源: Optional[str] = Field(
+            default=None,
            description="Title Source: 'chat_title' (Chat Title), 'ai_generated' (AI Generated), 'markdown_title' (Markdown Title)",
        )
-        界面语言: str = Field(
-            default="zh",
+        界面语言: Optional[str] = Field(
+            default=None,
            description="UI language for export messages. Options: 'en' (English), 'zh' (Chinese)",
        )
-        英文字体: str = Field(
-            default="Calibri",
+        英文字体: Optional[str] = Field(
+            default=None,
            description="Font for Latin characters (e.g., 'Times New Roman', '', 'Arial')",
        )
-        中文字体: str = Field(
-            default="SimSun",
+        中文字体: Optional[str] = Field(
+            default=None,
            description="Font for Asian characters (e.g., 'SimSun', 'Microsoft YaHei', 'PingFang SC')",
        )
-        代码字体: str = Field(
-            default="Consolas",
+        代码字体: Optional[str] = Field(
+            default=None,
            description="Font for code blocks and inline code (e.g., 'Consolas', 'Courier New', 'Monaco')",
        )
-        表头背景色: str = Field(
-            default="F2F2F2",
+        表头背景色: Optional[str] = Field(
+            default=None,
            description="Table header background color (hex, without #)",
        )
-        表格隔行背景色: str = Field(
-            default="FBFBFB",
+        表格隔行背景色: Optional[str] = Field(
+            default=None,
            description="Table zebra stripe background color for alternate rows (hex, without #)",
        )
-        Mermaid_PNG缩放比例: float = Field(
-            default=3.0,
+        Mermaid_PNG缩放比例: Optional[float] = Field(
+            default=None,
            description="PNG render resolution multiplier (higher = clearer, larger file)",
        )
-        Mermaid显示比例: float = Field(
-            default=1.0,
+        Mermaid显示比例: Optional[float] = Field(
+            default=None,
            description="Diagram width relative to available page width (<=1 recommended)",
        )
-        Mermaid布局优化: bool = Field(
-            default=False,
+        Mermaid布局优化: Optional[bool] = Field(
+            default=None,
            description="Optimize Mermaid layout: convert LR to TD for graph/flowchart",
        )
-        Mermaid背景色: str = Field(
-            default="",
+        Mermaid背景色: Optional[str] = Field(
+            default=None,
            description="Mermaid background color. Empty = transparent (recommended for Word dark mode). Used only for optional PNG fill.",
        )
-        启用Mermaid图注: bool = Field(
-            default=True,
+        启用Mermaid图注: Optional[bool] = Field(
+            default=None,
            description="Add figure captions under Mermaid images/charts",
        )
-        启用数学公式: bool = Field(
-            default=True,
+        启用数学公式: Optional[bool] = Field(
+            default=None,
            description="Enable LaTeX math block conversion (\\\\[...\\\\] and $$...$$) into Word equations",
        )
-        启用行内公式: bool = Field(
-            default=True,
+        启用行内公式: Optional[bool] = Field(
+            default=None,
            description="Enable inline $...$ math conversion into Word equations (conservative parsing to reduce false positives)",
        )

@@ -449,11 +454,21 @@ class Action:
            user_id = __user__.get("id", "unknown_user")

        # Apply UserValves if present
-        if __user__ and "valves" in __user__:
-            # Update self.valves with user-specific values
-            for key, value in __user__["valves"].model_dump().items():
-                if hasattr(self.valves, key):
-                    setattr(self.valves, key, value)
+        if __user__:
+            # Robustly parse UserValves whether it's a dict or Pydantic model
+            raw_valves = __user__.get("valves", {})
+            if isinstance(raw_valves, self.UserValves):
+                user_valves = raw_valves
+            elif isinstance(raw_valves, dict):
+                user_valves = self.UserValves(**raw_valves)
+            else:
+                user_valves = None
+
+            if user_valves:
+                for key, value in user_valves.model_dump(exclude_unset=True).items():
+                    # Only override if the value is not None (and explicitly set)
+                    if hasattr(self.valves, key) and value is not None:
+                        setattr(self.valves, key, value)

        # Get user language from Valves configuration
        self._user_lang = self._get_lang_key(self.valves.界面语言)
@@ -490,6 +505,37 @@ class Action:
            try:
                message_content = last_assistant_message["content"]
                if isinstance(message_content, str):
+                    if __event_emitter__ and self.valves.SHOW_DEBUG_LOG:
+                        debug_data = {}
+                        for name, regex in [
+                            ("Details Block (详情块)", _ALL_DETAILS_RE),
+                            ("Think Block (思考块)", _THINK_RE),
+                            ("Analysis Block (分析块)", _ANALYSIS_RE),
+                        ]:
+                            matches = regex.findall(message_content)
+                            if matches:
+                                debug_data[name] = [
+                                    (m[:200] + "...") if len(m) > 200 else m
+                                    for m in matches
+                                ]
+                        if debug_data:
+                            await self._emit_debug_log(
+                                __event_emitter__,
+                                "上下文内容清理分析 (Context Stripping Analysis)",
+                                debug_data,
+                            )
+
+                        # Log font configuration
+                        await self._emit_debug_log(
+                            __event_emitter__,
+                            "字体配置 (Font Configuration)",
+                            {
+                                "英文字体 (Latin Font)": self.valves.英文字体,
+                                "中文字体 (Asian Font)": self.valves.中文字体,
+                                "代码字体 (Code Font)": self.valves.代码字体,
+                            },
+                        )
+
                    message_content = self._strip_reasoning_blocks(message_content)

                if not message_content or not message_content.strip():
@@ -1101,34 +1147,11 @@ class Action:
        return title.strip() if isinstance(title, str) else ""

    def clean_filename(self, name: str) -> str:
-        """Clean illegal characters from filename and strip emoji."""
+        """清理文件名中的非法字符并移除 Emoji"""
        if not isinstance(name, str):
            return ""

-        def _is_emoji_codepoint(codepoint: int) -> bool:
-            # Common emoji ranges + flag regional indicators.
-            return (
-                0x1F000 <= codepoint <= 0x1FAFF
-                or 0x1F1E6 <= codepoint <= 0x1F1FF
-                or 0x2600 <= codepoint <= 0x26FF
-                or 0x2700 <= codepoint <= 0x27BF
-                or 0x2300 <= codepoint <= 0x23FF
-                or 0x2B00 <= codepoint <= 0x2BFF
-            )
-
-        def _is_emoji_modifier(codepoint: int) -> bool:
-            # VS15/VS16, ZWJ, keycap, skin tones, and tag characters used in some emoji sequences.
-            return (
-                codepoint in (0x200D, 0xFE0E, 0xFE0F, 0x20E3)
-                or 0x1F3FB <= codepoint <= 0x1F3FF
-                or 0xE0020 <= codepoint <= 0xE007F
-            )
-
-        without_emoji = "".join(
-            ch
-            for ch in name
-            if not (_is_emoji_codepoint(ord(ch)) or _is_emoji_modifier(ord(ch)))
-        )
+        without_emoji = self._remove_emojis(name)
        cleaned = re.sub(r'[\\/*?:"<>|]', "", without_emoji)
        cleaned = re.sub(r"\s+", " ", cleaned).strip().strip(".")
        return cleaned[:50].strip()
@@ -1496,7 +1519,10 @@ class Action:

            # If there is no h1 in content, prepend chat title as h1 when provided
            if top_heading and not has_h1:
-                self.add_heading(doc, top_heading, 1)
+                # Remove emojis from title for a professional look
+                clean_title = self._remove_emojis(top_heading)
+                # Use Title style (level 0) for the main document title
+                self.add_heading(doc, clean_title, 0)

            lines = markdown_text.split("\n")
            i = 0
@@ -1756,7 +1782,7 @@ class Action:
        cur = text
        for _ in range(10):
            prev = cur
-            cur = _REASONING_DETAILS_RE.sub("", cur)
+            cur = _ALL_DETAILS_RE.sub("", cur)
            cur = _THINK_RE.sub("", cur)
            cur = _ANALYSIS_RE.sub("", cur)
            if cur == prev:
@@ -2240,14 +2266,154 @@ class Action:
        font = style.font
        font.name = self.valves.英文字体
        font.size = Pt(11)
-        # Set Asian font
-        style._element.rPr.rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
+
+        # Ensure rPr element exists
+        rPr = style._element.get_or_add_rPr()
+        rFonts = rPr.get_or_add_rFonts()
+
+        # Set Latin and Asian fonts explicitly
+        rFonts.set(qn("w:ascii"), self.valves.英文字体)
+        rFonts.set(qn("w:hAnsi"), self.valves.英文字体)
+        rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
+
+        # Set language to zh-CN to prevent MS Gothic fallback (Japanese font)
+        lang = rPr.find(qn("w:lang"))
+        if lang is None:
+            lang = OxmlElement("w:lang")
+            rPr.append(lang)
+        lang.set(qn("w:val"), "en-US")
+        lang.set(qn("w:eastAsia"), "zh-CN")
+
+        logger.info(
+            f"[Font Config] Latin: {self.valves.英文字体}, Asian: {self.valves.中文字体}"
+        )

        # Set paragraph format
        paragraph_format = style.paragraph_format
        paragraph_format.line_spacing_rule = WD_LINE_SPACING.ONE_POINT_FIVE
        paragraph_format.space_after = Pt(6)

+        # 配置 Title 样式 (用于文档标题)
+        # 标准格式: 22pt (二号), 加粗, 居中, 段后 24pt
+        if "Title" in doc.styles:
+            title_style = doc.styles["Title"]
+            title_font = title_style.font
+            title_font.name = self.valves.英文字体
+            title_font.size = Pt(22)  # 二号字体
+            title_font.bold = True
+            title_font.color.rgb = RGBColor(0, 0, 0)
+
+            # 段落格式: 根据配置设置对齐方式和间距
+            title_pf = title_style.paragraph_format
+            alignment_map = {
+                "left": WD_ALIGN_PARAGRAPH.LEFT,
+                "center": WD_ALIGN_PARAGRAPH.CENTER,
+                "right": WD_ALIGN_PARAGRAPH.RIGHT,
+            }
+            title_pf.alignment = alignment_map.get(
+                self.valves.标题对齐方式.lower(), WD_ALIGN_PARAGRAPH.CENTER
+            )
+            title_pf.space_before = Pt(0)
+            title_pf.space_after = Pt(24)
+
+            t_rPr = title_style._element.get_or_add_rPr()
+            t_rFonts = t_rPr.get_or_add_rFonts()
+            t_rFonts.set(qn("w:ascii"), self.valves.英文字体)
+            t_rFonts.set(qn("w:hAnsi"), self.valves.英文字体)
+            t_rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
+
+            # Set language for Title
+            t_lang = t_rPr.find(qn("w:lang"))
+            if t_lang is None:
+                t_lang = OxmlElement("w:lang")
+                t_rPr.append(t_lang)
+            t_lang.set(qn("w:val"), "en-US")
+            t_lang.set(qn("w:eastAsia"), "zh-CN")
+
+        # 标准标题字号 (基于中文文档规范):
+        # Heading 1: 16pt (三号), 加粗, 段前 24pt, 段后 12pt
+        # Heading 2: 15pt (小三), 加粗, 段前 18pt, 段后 6pt
+        # Heading 3: 14pt (四号), 加粗, 段前 12pt, 段后 6pt
+        # Heading 4-9: 12pt (小四), 加粗, 逐级减小间距
+        heading_formats = {
+            1: {"size": 16, "space_before": 24, "space_after": 12},
+            2: {"size": 15, "space_before": 18, "space_after": 6},
+            3: {"size": 14, "space_before": 12, "space_after": 6},
+            4: {"size": 12, "space_before": 12, "space_after": 6},
+            5: {"size": 12, "space_before": 6, "space_after": 6},
+            6: {"size": 11, "space_before": 6, "space_after": 3},
+            7: {"size": 11, "space_before": 6, "space_after": 3},
+            8: {"size": 10.5, "space_before": 6, "space_after": 3},
+            9: {"size": 10.5, "space_before": 6, "space_after": 3},
+        }
+
+        # Apply font settings to Heading 1-9
+        for i in range(1, 10):
+            style_id = f"Heading {i}"
+            if style_id in doc.styles:
+                heading_style = doc.styles[style_id]
+                heading_font = heading_style.font
+                heading_font.name = self.valves.英文字体
+                heading_font.color.rgb = RGBColor(0, 0, 0)
+
+                # 应用标准格式
+                fmt = heading_formats.get(
+                    i, {"size": 11, "space_before": 6, "space_after": 3}
+                )
+                heading_font.size = Pt(fmt["size"])
+                heading_font.bold = True
+
+                heading_pf = heading_style.paragraph_format
+                heading_pf.space_before = Pt(fmt["space_before"])
+                heading_pf.space_after = Pt(fmt["space_after"])
+
+                # Ensure rPr exists
+                h_rPr = heading_style._element.get_or_add_rPr()
+                h_rFonts = h_rPr.get_or_add_rFonts()
+
+                # Set fonts
+                h_rFonts.set(qn("w:ascii"), self.valves.英文字体)
+                h_rFonts.set(qn("w:hAnsi"), self.valves.英文字体)
+                h_rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
+
+                # Set language for Heading
+                h_lang = h_rPr.find(qn("w:lang"))
+                if h_lang is None:
+                    h_lang = OxmlElement("w:lang")
+                    h_rPr.append(h_lang)
+                h_lang.set(qn("w:val"), "en-US")
+                h_lang.set(qn("w:eastAsia"), "zh-CN")
+
+    def _remove_emojis(self, text: str) -> str:
+        """从文本中移除 Emoji (基于 Unicode 范围)"""
+        if not isinstance(text, str):
+            return ""
+
+        def _is_emoji_codepoint(codepoint: int) -> bool:
+            # Common emoji ranges + flag regional indicators.
+            return (
+                0x1F000 <= codepoint <= 0x1FAFF
+                or 0x1F1E6 <= codepoint <= 0x1F1FF
+                or 0x2600 <= codepoint <= 0x26FF
+                or 0x2700 <= codepoint <= 0x27BF
+                or 0x2300 <= codepoint <= 0x23FF
+                or 0x2B00 <= codepoint <= 0x2BFF
+            )
+
+        def _is_emoji_modifier(codepoint: int) -> bool:
+            # VS15/VS16, ZWJ, keycap, skin tones, and tag characters used in some emoji sequences.
+            return (
+                codepoint in (0x200D, 0xFE0E, 0xFE0F, 0x20E3)
+                or 0x1F3FB <= codepoint <= 0x1F3FF
+                or 0xE0020 <= codepoint <= 0xE007F
+            )
+
+        return "".join(
+            ch
+            for ch in text
+            if not (_is_emoji_codepoint(ord(ch)) or _is_emoji_modifier(ord(ch)))
+        )
+
    def add_heading(self, doc: Document, text: str, level: int):
        """Add heading"""
        # Word heading levels start from 0, Markdown from 1
@@ -2283,6 +2449,12 @@ class Action:
        if strike:
            run.font.strike = True

+        # Explicitly set East Asian font to prevent MS Gothic fallback
+        # Word may not inherit w:eastAsia from style, causing Japanese font fallback for CJK
+        rPr = run._element.get_or_add_rPr()
+        rFonts = rPr.get_or_add_rFonts()
+        rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
+
    def _add_inline_code(self, paragraph, s: str):
        if s == "":
            return
@@ -2678,7 +2850,11 @@ class Action:
    ):
        u = self._normalize_url(url)
        if not u:
-            paragraph.add_run(display_text or text)
+            run = paragraph.add_run(display_text or text)
+            # Set East Asian font to prevent MS Gothic fallback
+            rPr = run._element.get_or_add_rPr()
+            rFonts = rPr.get_or_add_rFonts()
+            rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
            return

        part = getattr(paragraph, "part", None)
@@ -2687,6 +2863,10 @@ class Action:
            run = paragraph.add_run(display_text or text)
            run.font.color.rgb = RGBColor(0, 0, 255)
            run.font.underline = True
+            # Set East Asian font to prevent MS Gothic fallback
+            rPr = run._element.get_or_add_rPr()
+            rFonts = rPr.get_or_add_rFonts()
+            rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
            return

        r_id = part.relate_to(u, RT.HYPERLINK, is_external=True)
@@ -2700,6 +2880,11 @@ class Action:
        rStyle.set(qn("w:val"), "Hyperlink")
        rPr.append(rStyle)

+        # Set East Asian font to prevent MS Gothic fallback
+        rFonts = OxmlElement("w:rFonts")
+        rFonts.set(qn("w:eastAsia"), self.valves.中文字体)
+        rPr.append(rFonts)
+
        color = OxmlElement("w:color")
        color.set(qn("w:val"), "0000FF")
        rPr.append(color)
--- a/plugins/debug/copilot_sdk_research/GUIDE_ERRATA.md
+++ b/plugins/debug/copilot_sdk_research/GUIDE_ERRATA.md
@@ -0,0 +1,98 @@
+# 开发指南勘误与更新
+
+## 权限控制章节修正（第 2.2 节）
+
+### ⚠️ 关键勘误
+
+在实际测试中发现，Shell 权限请求使用的是 **`fullCommandText`** 字段，而非文档中提到的 `command` 字段。
+
+### 需要修改的代码行
+
+**第 89 行（错误）：**
+
+```python
+command = request.get("command", "")
+```
+
+**应改为（正确）：**
+
+```python
+command = request.get("fullCommandText", "") or request.get("command", "")
+```
+
+### 完整的正确实现
+
+```python
+async def on_user_permission_request(request, context):
+    """
+    统一权限审批网关
+    """
+    kind = request.get("kind")  # shell, write, mcp, read, url
+    # ✅ 正确：使用 fullCommandText（shell）或 command（其他）
+    command = request.get("fullCommandText", "") or request.get("command", "")
+
+    # 1. 超级模式：全部允许
+    if self.valves.PERMISSIONS_ALLOW_ALL:
+        return {"kind": "approved"}
+
+    # 2. 默认安全：始终允许 "读" 和 "Web浏览"
+    if kind in ["read", "url"]:
+        return {"kind": "approved"}
+
+    # 3. 细粒度控制
+    if kind == "shell":
+        if self.valves.PERMISSIONS_ALLOW_SHELL:
+            return {"kind": "approved"}
+        
+        pattern = self.valves.PERMISSIONS_SHELL_ALLOW_PATTERN
+        if pattern and command:
+            try:
+                if re.match(pattern, command):
+                    return {"kind": "approved"}
+            except re.error:
+                print(f"[Config Error] Invalid Regex: {pattern}")
+
+    if kind == "write" and self.valves.PERMISSIONS_ALLOW_WRITE:
+        return {"kind": "approved"}
+        
+    if kind == "mcp" and self.valves.PERMISSIONS_ALLOW_MCP:
+        return {"kind": "approved"}
+
+    # 4. 默认拒绝
+    print(f"[Permission Denied] Blocked: {kind} {command}")
+    return {
+        "kind": "denied-by-rules", 
+        "rules": [{"kind": "check-openwebui-valves"}]
+    }
+```
+
+### Shell 权限请求的完整结构
+
+```json
+{
+  "kind": "shell",
+  "toolCallId": "call_xxx",
+  "fullCommandText": "ls -la",                    // ← 关键字段
+  "intention": "List all files and directories",
+  "commands": [
+    {
+      "identifier": "ls -la",
+      "readOnly": false
+    }
+  ],
+  "possiblePaths": [],
+  "possibleUrls": [],
+  "hasWriteFileRedirection": false,
+  "canOfferSessionApproval": false
+}
+```
+
+## 测试验证
+
+已通过完整测试套件验证（8/8 通过），详见 [PERMISSION_TEST_REPORT.md](./PERMISSION_TEST_REPORT.md)。
+
+---
+
+**更新日期**: 2026-01-30  
+**验证状态**: ✅ 已测试  
+**影响范围**: 2.2 权限与确认章节
--- a/plugins/debug/copilot_sdk_research/PERMISSION_TEST_REPORT.md
+++ b/plugins/debug/copilot_sdk_research/PERMISSION_TEST_REPORT.md
@@ -0,0 +1,173 @@
+# GitHub Copilot SDK 权限控制测试报告
+
+## 测试日期
+
+2026-01-30
+
+## 测试环境
+
+- **Model**: gpt-4.1
+- **Python**: 3.12
+- **Copilot SDK**: Latest
+
+## 关键发现
+
+### 1. Shell 权限请求结构
+
+Shell 类型的权限请求使用 **`fullCommandText`** 字段，而非 `command` 字段。
+
+**完整请求示例：**
+
+```json
+{
+  "kind": "shell",
+  "toolCallId": "call_JKLi7tz3uSDQWE3LgzCpvSVy",
+  "fullCommandText": "ls -la",
+  "intention": "List all files and directories with details in the current directory",
+  "commands": [
+    {
+      "identifier": "ls -la",
+      "readOnly": false
+    }
+  ],
+  "possiblePaths": [],
+  "possibleUrls": [],
+  "hasWriteFileRedirection": false,
+  "canOfferSessionApproval": false
+}
+```
+
+### 2. 正则匹配模式验证
+
+正则白名单模式已验证有效，必须使用 `fullCommandText` 字段：
+
+```python
+command = request.get("fullCommandText", "") or request.get("command", "")
+pattern = self.valves.PERMISSIONS_SHELL_ALLOW_PATTERN
+if pattern and command:
+    if re.match(pattern, command):
+        return {"kind": "approved"}
+```
+
+## 测试结果
+
+### 完整测试套件（8/8 通过 ✅）
+
+| # | 测试用例 | 配置 | 提示词 | 预期 | 结果 | 状态 |
+|---|---------|------|--------|------|------|------|
+| 1 | Default Deny Shell | 默认 | 请执行: ls -la | ❌ Denied | ❌ Denied | ✅ |
+| 2 | Allow All | allow_all=True | 请执行: ls -la | ✅ Approved | ✅ Approved | ✅ |
+| 3 | Allow Shell | allow_shell=True | 请执行: pwd | ✅ Approved | ✅ Approved | ✅ |
+| 4 | Regex Match: ^ls | pattern='^ls' | 请执行: ls -la | ✅ Approved | ✅ Approved | ✅ |
+| 5 | Regex No Match | pattern='^ls' | 请执行: pwd | ❌ Denied | ❌ Denied | ✅ |
+| 6 | Regex Complex | pattern='^(ls\|pwd\|echo)' | 请执行: pwd | ✅ Approved | ✅ Approved | ✅ |
+| 7 | Regex No Match: git | pattern='^(ls\|pwd\|echo)' | 请执行: git status | ❌ Denied | ❌ Denied | ✅ |
+| 8 | Read Permission | 默认 | Read: README.md | ✅ Approved | ✅ Approved | ✅ |
+
+**总体通过率: 100%** 🎉
+
+## 推荐配置示例
+
+### 1. 安全模式（推荐生产环境）
+
+```python
+PERMISSIONS_ALLOW_ALL: bool = False
+PERMISSIONS_ALLOW_SHELL: bool = False
+PERMISSIONS_SHELL_ALLOW_PATTERN: str = "^(ls|pwd|echo|cat).*"
+PERMISSIONS_ALLOW_WRITE: bool = False
+PERMISSIONS_ALLOW_MCP: bool = True
+```
+
+### 2. 开发模式
+
+```python
+PERMISSIONS_ALLOW_ALL: bool = False
+PERMISSIONS_ALLOW_SHELL: bool = False
+PERMISSIONS_SHELL_ALLOW_PATTERN: str = "^(ls|pwd|echo|cat|grep|git status|npm test).*"
+PERMISSIONS_ALLOW_WRITE: bool = False
+PERMISSIONS_ALLOW_MCP: bool = True
+```
+
+### 3. 完全信任模式（仅限受控环境）
+
+```python
+PERMISSIONS_ALLOW_ALL: bool = True
+```
+
+## 实现建议
+
+### 正确的权限处理代码
+
+```python
+import re
+from typing import Any, Dict
+
+async def on_user_permission_request(request: Dict[str, Any], context: Dict[str, str]):
+    """
+    统一权限审批网关
+    """
+    kind = request.get("kind")
+    # 关键：使用 fullCommandText 而非 command
+    command = request.get("fullCommandText", "") or request.get("command", "")
+
+    # 1. 超级模式
+    if self.valves.PERMISSIONS_ALLOW_ALL:
+        return {"kind": "approved"}
+
+    # 2. 默认安全（read、url）
+    if kind in ["read", "url"]:
+        return {"kind": "approved"}
+
+    # 3. Shell 细粒度控制
+    if kind == "shell":
+        if self.valves.PERMISSIONS_ALLOW_SHELL:
+            return {"kind": "approved"}
+        
+        pattern = self.valves.PERMISSIONS_SHELL_ALLOW_PATTERN
+        if pattern and command:
+            try:
+                if re.match(pattern, command):
+                    return {"kind": "approved"}
+            except re.error as e:
+                logger.error(f"Invalid regex: {pattern} - {e}")
+
+    # 4. Write 权限
+    if kind == "write" and self.valves.PERMISSIONS_ALLOW_WRITE:
+        return {"kind": "approved"}
+        
+    # 5. MCP 权限
+    if kind == "mcp" and self.valves.PERMISSIONS_ALLOW_MCP:
+        return {"kind": "approved"}
+
+    # 6. 默认拒绝
+    logger.warning(f"Permission Denied: {kind} {command}")
+    return {
+        "kind": "denied-by-rules", 
+        "rules": [{"kind": "security-policy"}]
+    }
+```
+
+## 常见正则模式示例
+
+| 用途 | 正则表达式 | 说明 |
+|------|-----------|------|
+| 只读命令 | `^(ls|pwd|cat|echo|grep).*` | 允许常见只读命令 |
+| Git 只读 | `^git (status\|log\|diff\|show).*` | 允许 Git 只读操作 |
+| npm/yarn 测试 | `^(npm\|yarn) (test\|run).*` | 允许测试脚本 |
+| 完全 shell | `.*` | ⚠️ 危险：允许所有命令 |
+
+## 测试脚本位置
+
+- 基础测试: [test_shell_permission_pattern.py](./test_shell_permission_pattern.py)
+- 完整测试套件: [test_permission_comprehensive.py](./test_permission_comprehensive.py)
+
+## 结论
+
+✅ **GitHub Copilot SDK 的权限控制机制完全有效**
+✅ **正则白名单模式已验证可用**  
+⚠️ **必须使用 `fullCommandText` 字段获取命令内容**
+
+---
+
+**测试执行者**: GitHub Copilot  
+**审核状态**: ✅ 已验证
--- a/plugins/debug/copilot_sdk_research/copilot_pipe_enhancement_guide.md
+++ b/plugins/debug/copilot_sdk_research/copilot_pipe_enhancement_guide.md
@@ -0,0 +1,238 @@
+# OpenWebUI GitHub Copilot Pipe Enhancement Guide
+
+基于 Copilot SDK 源码级研究的深度技术总结，旨在指导 OpenWebUI Pipe 的功能增强开发。
+
+## 1. 认证机制 (Authentication)
+
+官方支持通过环境变量传递 Token。在 Pipe 中，只要确保 `GH_TOKEN` 或 `GITHUB_TOKEN` 存在于环境变量中，Copilot CLI 即可自动识别，无需在 `CopilotClient` 构造函数中重复注入。
+
+### 核心实现
+
+Pipe 应确保将 Token（来自 Valve 或 Env）正确设置到当前进程的环境变量中。
+
+```python
+import os
+from copilot import CopilotClient
+
+# 1. 设置环境变量 (如果从 Valve 获取)
+if self.valves.GH_TOKEN:
+    os.environ["GH_TOKEN"] = self.valves.GH_TOKEN
+
+# 2. 初始化客户端
+# CopilotClient 启动的 CLI 子进程会自动继承当前环境中的 GH_TOKEN
+client = CopilotClient({
+    # "cli_path": ...,
+    # 注意：无需在此处重复传入 github_token，CLI 会自动读取环境变量
+})
+
+# 3. 启动前检查 (建议)
+# status = await client.get_auth_status()
+# if not status.isAuthenticated: ...
+```
+
+## 2. 权限与确认 (Permissions & Tools) - 核心控制点
+
+这是用户最关心的部分：如何知道有哪些工具，以及如何控制它们的执行。
+
+### 2.1 内置工具 (Built-in Tools)
+
+Copilot CLI 内部管理了一组标准工具，**Python SDK 目前没有直接的 API (`client.list_tools()`) 来列出这些工具**。
+
+但是，根据 SDK 的 `PermissionRequest` 类型定义 (`copilot/types.py`)，我们可以反推其能力类别：
+
+* **`shell`**: 执行终端命令 (对应 `run_terminal_command` 等)
+* **`filesystem`** (对应 `read/write`): 文件读写 (对应 `read_file`, `edit_file`, `delete_file` 等)
+* **`url`**: 网络访问 (对应 `fetch_url` 等)
+* **`mcp`**: 连接的 MCP 服务器工具
+
+> **提示**: `available_tools` 参数可以用来“隐藏”工具，让 Agent 根本不知道它有一把锤子。而 `on_permission_request` 是用来拦截 Agent 挥舞锤子的动作。通常我们建议**能力全开 (不设置 available_tools 限制)**，而在**权限层 (on_permission_request) 做拦截**。
+
+### 2.2 实现“全部允许”与“按需允许”
+
+建议在 Valves 中增加权限控制字段，并在 `on_permission_request` 中实现逻辑。
+
+```python
+import re
+
+class Valves(BaseModel):
+    # ... 其他 Valve ...
+    # 权限控制开关
+    PERMISSIONS_ALLOW_ALL: bool = Field(default=False, description="DANGER: Auto-approve ALL actions (shell, write, etc).")
+    PERMISSIONS_ALLOW_SHELL: bool = Field(default=False, description="Auto-approve shell commands.")
+    PERMISSIONS_SHELL_ALLOW_PATTERN: str = Field(default="", description="Regex for approved shell commands (e.g., '^ls|^grep').")
+    PERMISSIONS_ALLOW_WRITE: bool = Field(default=False, description="Auto-approve file write/edit/delete.")
+    PERMISSIONS_ALLOW_MCP: bool = Field(default=True, description="Auto-approve MCP tool execution.")
+
+# 权限处理 Hook 实现
+async def on_user_permission_request(request, context):
+    """
+    统一权限审批网关
+    request keys: kind, toolCallId, ... (shell requests have 'command')
+    """
+    kind = request.get("kind") # shell, write, mcp, read, url
+
+    # 1. 超级模式：全部允许
+    if self.valves.PERMISSIONS_ALLOW_ALL:
+        return {"kind": "approved"}
+
+    # 2. 默认安全：始终允许 "读" 和 "Web浏览" (根据需求调整)
+    if kind in ["read", "url"]:
+        return {"kind": "approved"}
+
+    # 3. 细粒度控制
+    if kind == "shell":
+        # 3.1 完全允许 Shell
+        if self.valves.PERMISSIONS_ALLOW_SHELL:
+            return {"kind": "approved"}
+        
+        # 3.2 基于正则允许特定命令
+        command = request.get("command", "")
+        pattern = self.valves.PERMISSIONS_SHELL_ALLOW_PATTERN
+        if pattern and command:
+            try:
+                if re.match(pattern, command):
+                    return {"kind": "approved"}
+            except re.error:
+                print(f"[Config Error] Invalid Regex: {pattern}")
+
+    if kind == "write" and self.valves.PERMISSIONS_ALLOW_WRITE:
+        return {"kind": "approved"}
+        
+    if kind == "mcp" and self.valves.PERMISSIONS_ALLOW_MCP:
+        return {"kind": "approved"}
+
+    # 4. 默认拒绝
+    print(f"[Permission Denied] Blocked request for: {kind} {request.get('command', '')}")
+    return {
+        "kind": "denied-by-rules", 
+        "rules": [{"kind": "check-openwebui-valves"}]
+    }
+
+# 注册 Hook
+session = await client.create_session({
+    # ... 
+    "on_permission_request": on_user_permission_request
+})
+```
+
+## 3. Agent 与 MCP 集成 (Agents & MCP)
+
+SDK 中的 Agent 和 MCP 并非独立文件，而是会话配置 (`SessionConfig`) 的一部分。Pipe 可以通过 Valves 动态构建这些配置。
+
+### 关键映射关系
+
+| SDK 概念 | OpenWebUI 对应 | 实现位置 | 关键参数 |
+| :--- | :--- | :--- | :--- |
+| **Custom Agent** | 自定义模型 / Persona | `create_session(custom_agents=[...])` | `name`, `prompt`, `tools` (仅名称) |
+| **Agent Tools** | Valve 开关 / 预置工具 | `create_session(tools=[func1, func2])` | 必须先在 `tools` 注册函数，Agent 才能引用 |
+| **MCP Server** | Valve 配置 (JSON) | `create_session(mcp_servers={...})` | `command`, `args`, `env` (本地) |
+
+### 代码范式：动态构建 Agent
+
+```python
+async def create_agent_session(client, user_prompt, model_name):
+    # 1. 定义工具 (必须是函数引用)
+    # 假设已从 OpenWebUI Tools 转换或内置
+    available_tools = [tool_web_search, tool_run_script] 
+    
+    # 2. 构建 Agent Manifest (针对当前请求的虚拟 Agent)
+    agent_manifest = {
+        "name": "openwebui_agent",
+        "description": "Dynamic agent from OpenWebUI",
+        "prompt": "You are a helpful assistant...", # 这里注入 System Prompt
+        "tools": ["web_search", "run_script"],    # 引用上方工具的 name
+        "mcp_servers": {
+            # 可以在这里为特定 Agent 绑定 MCP
+        }
+    }
+
+    # 3. 创建会话
+    session = await client.create_session({
+        "model": "gpt-4", # 底层模型
+        "custom_agents": [agent_manifest],
+        "tools": available_tools,          # 注册实际代码
+        "available_tools": ["web_search"], # 白名单控制当前可用工具
+        # ... 权限配置
+    })
+```
+
+## 4. MCP 服务器配置 (Native MCP Support)
+
+Pipe 可以直接支持标准 MCP 协议（Stdio）。不需要额外的 MCP 客户端代理，SDK 原生支持。
+
+### Valve 配置结构建议
+
+建议在 Pipe 的 Valves 中增加一个 `MCP_CONFIG` 字段（JSON 字符串），解析后直接传给 SDK。
+
+```python
+# Valve 输入示例 (JSON)
+# {
+#   "brave_search": {
+#     "type": "local",
+#     "command": "npx",
+#     "args": ["-y", "@modelcontextprotocol/server-brave-search"],
+#     "env": {"BRAVE_API_KEY": "..."}
+#   }
+# }
+
+# 代码实现
+mcp_config = json.loads(self.valves.MCP_CONFIG)
+session = await client.create_session({
+    # ...
+    "mcp_servers": mcp_config,
+    # 注意：必须配合权限自动审批，否则 MCP 工具无法调用
+    "on_permission_request": auto_approve_policy 
+})
+```
+
+## 5. 会话管理：持久化 vs 重放 (Persistence)
+
+OpenWebUI 是无状态的，但 Copilot SDK 是有状态的（保留上下文窗口优化）。
+
+### 最佳实践：以 `chat_id` 为锚点
+
+利用 OpenWebUI 提供的 `chat_id` 来决定是 `resume` 还是 `start`。
+
+1. **Map**: 维护 `Dict[chat_id, session_id]` (内存或数据库)。
+2. **Flow**:
+    * 请求进来 -> 检查 `chat_id` 是否有对应的 `session_id`。
+    * **有**: 尝试 `client.resume_session(session_id)`。
+        * *注意*：Resume 时必须重新传入 `tools`, `hooks`, `on_permission_request`，因为这些 Python 对象不会被序列化保存。
+    * **无/失败**: 调用 `client.create_session()`，并将新 `session_id` 存入 Map。
+3. **Fallback**: 如果 Resume 失败（例如后端重启 SDK 进程丢失），回退到 Create 新会话，并可选地将 OpenWebUI 传来的 `messages` 历史以 System Message 或历史插入的方式“重放”进去（虽然 SDK 不直接支持 insert history，但可以通过连续的 `send` 模拟，但这很慢）。
+    * *简易方案*：Resume 失败就作为新对话开始，只带入 System Prompt。
+
+## 6. 高级 Hook：提示词增强
+
+利用 `on_user_prompt_submitted` 钩子，可以在不修改用户可见内容的情况下，向 Copilot 注入隐式上下文（例如当前文件内容、Pipe 的元指令）。
+
+```python
+async def inject_context_hook(input_data, ctx):
+    user_prompt = input_data["prompt"]
+    
+    # 比如：检测到用户在问代码，自动附加上下文
+    additional_context = "Current Language: Python. Framework: OpenWebUI."
+    
+    return {
+        "modifiedPrompt": user_prompt, # 可以在这里改写提示词
+        "additionalContext": additional_context # 注入隐藏上下文
+    }
+
+session = await client.create_session({
+    # ...
+    "hooks": {
+        "on_user_prompt_submitted": inject_context_hook
+    }
+})
+```
+
+---
+
+**总结开发清单：**
+
+1. [ ] **Env Auth**: 读取环境变量 -> `CopilotClient`。
+2. [ ] **Permission Valve**: 实现 `PERMISSIONS_ALLOW_ALL/SHELL` 等 Valves。
+3. [ ] **Auto-Approve Hook**: 实现 `on_permission_request` 逻辑。
+4. [ ] **MCP Valve**: 添加 JSON Valve -> `session.mcp_servers`。
+5. [ ] **Session Map**: 实现 `chat_id` <-> `session_id` 的简单的内存映射。
+6. [ ] **Resume Logic**: 优先 `resume_session`，并记得并在 resume 时重传 Hook 和 Tools。
--- a/plugins/debug/copilot_sdk_research/copilot_sdk_guide.py
+++ b/plugins/debug/copilot_sdk_research/copilot_sdk_guide.py
@@ -0,0 +1,620 @@
+#!/usr/bin/env python3
+import argparse
+import asyncio
+import datetime as dt
+import json
+import logging
+import os
+import sys
+import textwrap
+from typing import Iterable, List, Optional
+
+from copilot import CopilotClient
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
+)
+logger = logging.getLogger("copilot_sdk_guide")
+
+DEFAULT_CONTEXT_URLS = [
+    "https://raw.githubusercontent.com/github/copilot-sdk/main/README.md",
+    "https://raw.githubusercontent.com/github/copilot-sdk/main/python/README.md",
+    "https://raw.githubusercontent.com/github/copilot-sdk/main/docs/getting-started.md",
+    "https://raw.githubusercontent.com/github/copilot-cli/main/README.md",
+    "https://raw.githubusercontent.com/github/copilot-cli/main/changelog.md",
+    "https://docs.github.com/en/copilot/concepts/agents/about-copilot-cli",
+    "https://docs.github.com/en/copilot/concepts/agents/about-agent-skills",
+    "https://raw.githubusercontent.com/github/awesome-copilot/main/README.md",
+    "https://raw.githubusercontent.com/github/awesome-copilot/main/skills/copilot-sdk/SKILL.md",
+    "https://raw.githubusercontent.com/github/awesome-copilot/main/instructions/agent-skills.instructions.md",
+]
+
+AWESOME_COPILOT_REPO = "github/awesome-copilot"
+AWESOME_COPILOT_BRANCH = "main"
+AWESOME_COPILOT_DOC_DIRS = ["docs/", "instructions/"]
+
+TOPICS = [
+    "MCP Server Integration: JSON-RPC config and SDK hooks",
+    "Agent Manifests: Defining capabilities and permissions programmatically",
+    "Headless Auth: Device Code Flow and credential persistence",
+    "Session Replay vs Resume: Handling stateless frontend history",
+    "Advanced Session Hooks: Intercepting and modifying user prompts",
+    "Workspace Virtualization: Handling CWD for remote/virtual files",
+    "Error Recovery: Handling session disconnects and re-auth",
+    "Confirmation Events: programmatic handling of 'confirmation_required'",
+    "Skills: Conflict resolution and precedence defaults",
+    "Debugging: Tracing JSON-RPC traffic in the SDK",
+    "Billing & Policies: How seat management affects SDK features",
+]
+
+QUESTION_TEMPLATES = [
+    "Give a concise overview of {topic}.",
+    "Provide best practices and common pitfalls for {topic}.",
+    "Show a minimal example snippet for {topic}.",
+    "List recommended configuration defaults for {topic}.",
+    "How does {topic} relate to building a custom Agent?",
+]
+
+CLI_FOCUS_QUESTIONS = [
+    "How to configure MCP servers in ~/.copilot/config.json for SDK usage?",
+    "What CLI environment variables force 'Agent' mode vs 'Generic' mode?",
+    "Explain the 'confirmation' flow in CLI and how it maps to SDK events.",
+    "Does the CLI support 'dry-run' permission checks for tools?",
+    "What are the undocumented requirements for 'workspace' context updates?",
+    "How does the CLI handle 'device code' re-authentication automatically?",
+]
+
+
+def build_questions(max_questions: int) -> List[str]:
+    questions: List[str] = []
+
+    for topic in TOPICS:
+        for template in QUESTION_TEMPLATES:
+            questions.append(template.format(topic=topic))
+
+    questions.extend(CLI_FOCUS_QUESTIONS)
+
+    # De-duplicate while preserving order
+    seen = set()
+    uniq: List[str] = []
+    for q in questions:
+        if q in seen:
+            continue
+        seen.add(q)
+        uniq.append(q)
+
+    return uniq[:max_questions]
+
+
+def build_deep_dive_prompts() -> List[str]:
+    return [
+        "Provide a python code example for configuring `CopilotClient` to connect to a local MCP server (e.g. Brave Search) via `CopilotClient` config.",
+        "Explain how to programmatically handle `tool.confirmation_required` events in a non-interactive stream using `session.on()`.",
+        "Show how to implement a 'Device Flow' login helper using SDK primitives (if available) or raw HTTP showing how to persist credentials.",
+        "Compare the pros and cons of 'Session Replay' (fast-forwarding history) vs 'Session Resume' (stateful ID) for a stateless web backend like OpenWebUI.",
+        "Detail the exact protocol for 'Virtual Workspace': how to implement a file system provider that feeds content to Copilot without physical files.",
+        "Create an 'Agent Manifest' example: how to define an Agent capable of specific high-privileged tools via SDK.",
+        "List all 'hidden' `SessionConfig` parameters relevant to Agent behavior and personality.",
+    ]
+
+
+def load_questions(path: str) -> List[str]:
+    if path.lower().endswith(".json"):
+        with open(path, "r", encoding="utf-8") as f:
+            data = json.load(f)
+        if isinstance(data, list):
+            return [str(x).strip() for x in data if str(x).strip()]
+        raise ValueError("JSON must be an array of strings")
+
+    with open(path, "r", encoding="utf-8") as f:
+        lines = [line.strip() for line in f.readlines()]
+    return [line for line in lines if line]
+
+
+def fetch_url(url: str, headers: Optional[dict] = None) -> str:
+    import urllib.request
+    import time
+
+    retries = 3
+    if headers is None:
+        headers = {}
+
+    req = urllib.request.Request(url, headers=headers)
+
+    for i in range(retries):
+        try:
+            with urllib.request.urlopen(req, timeout=20) as response:
+                return response.read().decode("utf-8", errors="replace")
+        except Exception as exc:
+            if i == retries - 1:
+                logger.warning(
+                    "Failed to fetch %s after %d attempts: %s", url, retries, exc
+                )
+                return ""
+            time.sleep(1 * (i + 1))
+    return ""
+
+
+def list_repo_markdown_urls(
+    repo: str,
+    branch: str,
+    dir_prefixes: List[str],
+) -> List[str]:
+    api_url = f"https://api.github.com/repos/{repo}/git/trees/{branch}?recursive=1"
+    headers = {}
+    if os.environ.get("GITHUB_TOKEN"):
+        headers["Authorization"] = f"token {os.environ.get('GITHUB_TOKEN')}"
+
+    try:
+        content = fetch_url(api_url, headers=headers)
+        if not content:
+            return []
+        data = json.loads(content)
+    except Exception as exc:
+        logger.warning("Failed to list repo tree: %s", exc)
+        return []
+
+    tree = data.get("tree", []) if isinstance(data, dict) else []
+    urls: List[str] = []
+    for item in tree:
+        if not isinstance(item, dict):
+            continue
+        path = item.get("path", "")
+        if not path or not path.endswith(".md"):
+            continue
+        if any(path.startswith(prefix) for prefix in dir_prefixes):
+            raw = f"https://raw.githubusercontent.com/{repo}/{branch}/{path}"
+            urls.append(raw)
+    return urls
+
+
+def read_local_sdk_source(max_chars: int = 300000) -> str:
+    """
+    Locates the installed 'copilot' package and reads its source code.
+    This ensures analysis is based on the actual installed version, not just docs.
+    """
+    try:
+        import copilot
+    except ImportError:
+        logger.error("Could not import 'copilot' SDK. Is it installed?")
+        return ""
+
+    package_dir = os.path.dirname(copilot.__file__)
+    logger.info(f"Reading SDK source from: {package_dir}")
+
+    source_chunks = []
+    total_chars = 0
+
+    # Prioritize key files that define core logic
+    priority_files = ["client.py", "session.py", "types.py", "events.py", "__init__.py"]
+
+    # First pass: Recursively find all .py files
+    all_py_files = []
+    for root, dirs, files in os.walk(package_dir):
+        if "__pycache__" in root:
+            continue
+        for file in files:
+            if file.endswith(".py"):
+                all_py_files.append(os.path.join(root, file))
+
+    # Sort files: priority files first, then alphabetical
+    def sort_key(path):
+        fname = os.path.basename(path)
+        if fname in priority_files:
+            return (0, priority_files.index(fname))
+        return (1, path)
+
+    all_py_files.sort(key=sort_key)
+
+    for path in all_py_files:
+        rel_path = os.path.relpath(path, os.path.dirname(package_dir))
+        try:
+            with open(path, "r", encoding="utf-8") as f:
+                content = f.read()
+
+            # Add file delimiter for the model
+            header = f"\n\n# ==================================================\n# SOURCE CODE FILE: {rel_path}\n# ==================================================\n"
+            chunk = header + content
+
+            if total_chars + len(chunk) > max_chars:
+                remaining = max_chars - total_chars
+                if remaining > len(header) + 100:
+                    source_chunks.append(
+                        chunk[:remaining] + "\n# [TRUNCATED DUE TO LENGTH LIMIT]"
+                    )
+                logger.warning(f"Context limit reached. Stopping at {rel_path}")
+                break
+
+            source_chunks.append(chunk)
+            total_chars += len(chunk)
+            logger.info(f"Loaded source file: {rel_path} ({len(content)} chars)")
+
+        except Exception as e:
+            logger.warning(f"Failed to read source file {path}: {e}")
+
+    return "".join(source_chunks)
+
+
+def build_context(urls: Iterable[str], max_chars: int) -> str:
+    chunks: List[str] = []
+    remaining = max_chars
+
+    for url in urls:
+        if remaining <= 0:
+            break
+        try:
+            content = fetch_url(url)
+            header = f"[Source: {url}]\n"
+            if len(header) >= remaining:
+                break
+            remaining -= len(header)
+
+            if len(content) > remaining:
+                content = content[:remaining] + "\n[TRUNCATED]\n"
+                remaining = 0
+            else:
+                remaining -= len(content)
+
+            chunks.append(header + content)
+            logger.info("Fetched context: %s", url)
+        except Exception as exc:
+            logger.warning("Failed to fetch %s: %s", url, exc)
+
+    return "\n\n".join(chunks)
+
+
+def write_jsonl(path: str, item: dict) -> None:
+    with open(path, "a", encoding="utf-8") as f:
+        f.write(json.dumps(item, ensure_ascii=False) + "\n")
+
+
+def write_markdown_header(path: str, title: str, meta: dict) -> None:
+    with open(path, "w", encoding="utf-8") as f:
+        f.write(f"# {title}\n\n")
+        for k, v in meta.items():
+            f.write(f"- **{k}**: {v}\n")
+        f.write("\n---\n\n")
+
+
+def append_markdown_qa(path: str, question: str, answer: str) -> None:
+    with open(path, "a", encoding="utf-8") as f:
+        f.write(f"## Q: {question}\n\n")
+        f.write(f"{answer}\n\n")
+
+
+def clamp_questions(questions: List[str], max_questions: int) -> List[str]:
+    return questions[: max(1, min(max_questions, 400))]
+
+
+def print_progress_bar(
+    iteration,
+    total,
+    prefix="",
+    suffix="",
+    decimals=1,
+    length=50,
+    fill="█",
+    printEnd="\r",
+):
+    """
+    Call in a loop to create terminal progress bar
+    """
+    percent = ("{0:." + str(decimals) + "f}").format(100 * (iteration / float(total)))
+    filledLength = int(length * iteration // total)
+    bar = fill * filledLength + "-" * (length - filledLength)
+    # Clear line extension to handle shrinking suffixes
+    print(f"\r{prefix} |{bar}| {percent}% {suffix}\033[K", end=printEnd)
+    # Print New Line on Complete
+    if iteration == total:
+        print()
+
+
+async def run_session(
+    model: str,
+    questions: List[str],
+    output_dir: str,
+    context: str,
+    session_id: Optional[str],
+    delay: float,
+    output_lang: str,
+    enable_infinite_sessions: bool,
+    timeout: int,
+) -> None:
+    client = CopilotClient()
+    await client.start()
+
+    session_config = {"model": model}
+    if session_id:
+        session_config["session_id"] = session_id
+    if enable_infinite_sessions:
+        session_config["infinite_sessions"] = {
+            "enabled": True,
+            "background_compaction_threshold": 0.8,
+            "buffer_exhaustion_threshold": 0.95,
+        }
+
+    session = await client.create_session(session_config)
+
+    timestamp = dt.datetime.now().strftime("%Y%m%d_%H%M%S")
+    jsonl_path = os.path.join(output_dir, f"copilot_sdk_guide_{timestamp}.jsonl")
+    md_path = os.path.join(output_dir, f"copilot_sdk_guide_{timestamp}.md")
+
+    write_markdown_header(
+        md_path,
+        "GitHub Copilot SDK & CLI 研究报告",
+        {
+            "model": model,
+            "questions": len(questions),
+            "timestamp": timestamp,
+            "language": output_lang,
+        },
+    )
+
+    lang_instruction = "Chinese" if "zh" in output_lang.lower() else "English"
+
+    system_prompt = textwrap.dedent(
+        f"""
+        You are an expert assistant. Focus on GitHub Copilot SDK and GitHub Copilot CLI.
+        
+        CRITICAL INSTRUCTION: SOURCE CODE FIRST.
+        You have been provided with the ACTUAL PYTHON SOURCE CODE of the `copilot` SDK in the context.
+        When answering questions:
+        1. FIRST, analyze the provided source code (look for class definitions, type hints, methods).
+        2. THEN, refer to documentation if source code is ambiguous.
+        3. Do NOT hallucinate methods that do not exist in the source code.
+        4. If a feature (like MCP) is not explicitly in the code, explain how to implement it using the available primitives (low-level hooks/events).
+
+        Provide accurate, concise answers in {lang_instruction}. When relevant, include command names,
+        configuration keys, and pitfalls. Use bullet points where useful.
+
+        Output requirements:
+        - Write in {lang_instruction}.
+        - Provide practical code snippets (Python/TypeScript/CLI) when helpful.
+        - Include a short "建议/落地" section for integration into a pipe.
+        - If citing facts from provided context, briefly mention the source URL.
+        """
+    ).strip()
+
+    if context:
+        system_prompt += "\n\nAdditional context:\n" + context
+
+    await session.send_and_wait({"prompt": system_prompt}, timeout=timeout)
+
+    total_q = len(questions)
+    print_progress_bar(0, total_q, prefix="Progress:", suffix="Starting...", length=30)
+
+    for idx, question in enumerate(questions, start=1):
+        # Update progress bar (Asking...)
+        q_short = (question[:40] + "...") if len(question) > 40 else question.ljust(43)
+        print_progress_bar(
+            idx - 1, total_q, prefix="Progress:", suffix=f"Asking: {q_short}", length=30
+        )
+
+        # Log to file/debug only
+        logger.debug("[%s/%s] Asking: %s", idx, total_q, question)
+
+        answer = ""
+        max_retries = 3
+        for attempt in range(max_retries):
+            try:
+                response = await session.send_and_wait(
+                    {"prompt": question}, timeout=timeout
+                )
+                answer = response.data.content if response and response.data else ""
+                break
+            except Exception as e:
+                logger.error(
+                    f"Error asking question (Attempt {attempt+1}/{max_retries}): {e}"
+                )
+                if attempt < max_retries - 1:
+                    await asyncio.sleep(2)
+                else:
+                    answer = f"Error retrieving answer: {e}"
+
+        write_jsonl(
+            jsonl_path,
+            {
+                "index": idx,
+                "question": question,
+                "answer": answer,
+                "model": model,
+            },
+        )
+        append_markdown_qa(md_path, question, answer)
+
+        # Update progress bar (Done...)
+        print_progress_bar(
+            idx, total_q, prefix="Progress:", suffix=f"Done: {q_short}", length=30
+        )
+
+        if delay > 0:
+            await asyncio.sleep(delay)
+
+    await session.destroy()
+    await client.stop()
+
+    logger.info("Saved output to %s and %s", jsonl_path, md_path)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description="Ask up to 100 Copilot SDK questions via GitHub Copilot SDK",
+    )
+    parser.add_argument("--model", default="gpt-5.2-codex", help="Model to use")
+    parser.add_argument(
+        "--max-questions",
+        type=int,
+        default=100,
+        help="Max number of questions (1-400)",
+    )
+    parser.add_argument(
+        "--questions-file",
+        default="",
+        help="Path to .txt or .json list of questions",
+    )
+    parser.add_argument(
+        "--context-url",
+        action="append",
+        default=[],
+        help="Additional context URL (repeatable)",
+    )
+    parser.add_argument(
+        "--no-default-context",
+        action="store_true",
+        help="Disable default Copilot SDK context URLs",
+    )
+    parser.add_argument(
+        "--include-awesome-copilot-docs",
+        action="store_true",
+        help="Include all markdown files from awesome-copilot/docs",
+    )
+    parser.add_argument(
+        "--include-awesome-copilot-instructions",
+        action="store_true",
+        help="Include all markdown files from awesome-copilot/instructions",
+    )
+    parser.add_argument(
+        "--no-sdk-source",
+        action="store_true",
+        help="Do NOT read local SDK source code (default: reads source)",
+    )
+    parser.add_argument(
+        "--session-id",
+        default="",
+        help="Optional custom session ID",
+    )
+    parser.add_argument(
+        "--output-dir",
+        default="",
+        help="Directory to save outputs",
+    )
+    parser.add_argument(
+        "--delay",
+        type=float,
+        default=0.5,
+        help="Delay between questions (seconds)",
+    )
+    parser.add_argument(
+        "--max-context-chars",
+        type=int,
+        default=400000,
+        help="Max characters of aggregated context (default: 400000)",
+    )
+    parser.add_argument(
+        "--disable-infinite-sessions",
+        action="store_true",
+        help="Disable infinite sessions (default: enabled)",
+    )
+    parser.add_argument(
+        "--output-lang",
+        default="zh-CN",
+        help="Output language (default: zh-CN)",
+    )
+    parser.add_argument(
+        "--deep-dive",
+        action="store_true",
+        help="Append deep-dive prompts for more detailed research",
+    )
+    parser.add_argument(
+        "--timeout",
+        type=int,
+        default=3600,
+        help="Session request timeout in seconds (default: 3600)",
+    )
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+
+    if args.questions_file:
+        questions = load_questions(args.questions_file)
+    else:
+        # Generate enough questions to cover everything
+        questions = build_questions(9999)
+
+    if args.deep_dive:
+        # Prepend deep dive questions to ensure they are prioritized
+        questions = build_deep_dive_prompts() + questions
+
+    questions = clamp_questions(questions, args.max_questions)
+    if not questions:
+        logger.error("No questions to ask")
+        sys.exit(1)
+
+    context_urls = [] if args.no_default_context else list(DEFAULT_CONTEXT_URLS)
+
+    if args.include_awesome_copilot_docs:
+        context_urls.extend(
+            list_repo_markdown_urls(
+                AWESOME_COPILOT_REPO,
+                AWESOME_COPILOT_BRANCH,
+                ["docs/"],
+            )
+        )
+
+    if args.include_awesome_copilot_instructions:
+        context_urls.extend(
+            list_repo_markdown_urls(
+                AWESOME_COPILOT_REPO,
+                AWESOME_COPILOT_BRANCH,
+                ["instructions/"],
+            )
+        )
+
+    context_urls.extend(args.context_url or [])
+
+    # 1. Read local source code first (Priority: High)
+    # We allocate up to max_context_chars to source code initially.
+    # The actual usage will likely be less for a typical SDK.
+    source_context = ""
+    source_chars_count = 0
+    if not args.no_sdk_source:
+        source_context = read_local_sdk_source(args.max_context_chars)
+        source_chars_count = len(source_context)
+        logger.info(f"Source context usage: {source_chars_count} chars")
+
+    # 2. Calculate remaining budget for Web Docs (Priority: Secondary)
+    # We ensure we don't exceed the global limit.
+    remaining_chars = max(10000, args.max_context_chars - source_chars_count)
+    logger.info(f"Remaining budget for web docs: {remaining_chars} chars")
+
+    # 3. Fetch remote docs
+    web_context = build_context(context_urls, remaining_chars)
+
+    combined_context = ""
+    # Assemble context in order of authority (Source > Docs)
+    if source_context:
+        combined_context += (
+            "# PRIMARY SOURCE: LOCAL SDK CODE (AUTHORITATIVE)\n"
+            + source_context
+            + "\n\n"
+        )
+    if web_context:
+        combined_context += (
+            "# SECONDARY SOURCE: WEB DOCUMENTATION & AWESOME-COPILOT\n" + web_context
+        )
+
+    output_dir = args.output_dir or os.path.join(
+        os.getcwd(), "plugins", "debug", "copilot_sdk_research", "outputs"
+    )
+    os.makedirs(output_dir, exist_ok=True)
+
+    asyncio.run(
+        run_session(
+            model=args.model,
+            questions=questions,
+            output_dir=output_dir,
+            context=combined_context,
+            session_id=args.session_id or None,
+            delay=args.delay,
+            output_lang=args.output_lang,
+            enable_infinite_sessions=not args.disable_infinite_sessions,
+            timeout=args.timeout,
+        )
+    )
+
+
+if __name__ == "__main__":
+    main()
--- a/plugins/debug/copilot_sdk_research/dummy_test_file.txt
+++ b/plugins/debug/copilot_sdk_research/dummy_test_file.txt
@@ -0,0 +1,2 @@
+This is a dummy test file for verifying permission controls.
+You can delete this file safely.
--- a/plugins/debug/copilot_sdk_research/inspect_permissions.py
+++ b/plugins/debug/copilot_sdk_research/inspect_permissions.py
@@ -0,0 +1,26 @@
+import inspect
+import json
+
+try:
+    import copilot
+    from copilot import types
+
+    print(f"Copilot location: {copilot.__file__}")
+
+    print("\n=== Inspecting PermissionRequest types ===")
+    # Look for definition of PermissionRequest or similar
+
+    # In the absence of direct access to the CLI output structure documentation,
+    # we can check if there are type hints or typed dicts in copilot.types
+
+    for name, obj in inspect.getmembers(types):
+        if "Permission" in name or "Request" in name:
+            print(f"\nType: {name}")
+            try:
+                if hasattr(obj, "__annotations__"):
+                    print(obj.__annotations__)
+            except:
+                pass
+
+except ImportError:
+    print("copilot module not installed")
--- a/plugins/debug/copilot_sdk_research/outputs/copilot_sdk_guide_20260130_114642.jsonl
+++ b/plugins/debug/copilot_sdk_research/outputs/copilot_sdk_guide_20260130_114642.jsonl
@@ -0,0 +1,68 @@
+{"index": 1, "question": "Provide a python code example for configuring `CopilotClient` to connect to a local MCP server (e.g. Brave Search) via `CopilotClient` config.", "answer": "SDK 源码显示 MCP 服务器通过 `create_session`/`resume_session` 的 `mcp_servers` 配置传入并映射为 `mcpServers`，不是 CopilotClient 构造参数（源：本地源码 copilot/types.py / copilot/client.py）。下面是本地 stdio MCP（以 Brave Search 为例，命令/密钥请替换）的 Python 示例。  \n```python\nimport asyncio\nfrom copilot import CopilotClient\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"mcp_servers\": {\n            \"brave_search\": {\n                \"type\": \"local\",  # 或 \"stdio\"\n                \"command\": \"/path/to/brave-search-mcp\",  # 替换为你的 MCP server 启动命令\n                \"args\": [],\n                \"env\": {\"BRAVE_API_KEY\": \"YOUR_KEY\"},\n                \"tools\": [\"*\"],\n                \"timeout\": 30000,  # 毫秒\n            }\n        },\n    })\n\n    await session.send_and_wait({\"prompt\": \"用 Brave Search 查询 OpenAI 最新公告\"})\n    await client.stop()\n\nasyncio.run(main())\n```\n\n要点/坑：\n- `tools`: `[\"*\"]` 表示全量，`[]` 表示不启用\n- `timeout`: 毫秒\n- `command/args/env`: 按你的 MCP server 实际启动方式替换\n\n建议/落地：\n- pipe 内复用同一 session，结束时统一 `client.stop()`", "model": "gpt-5.2-codex"}
+{"index": 2, "question": "Explain how to programmatically handle `tool.confirmation_required` events in a non-interactive stream using `session.on()`.", "answer": "SDK 源码未定义 `tool.confirmation_required` 事件，权限确认由 JSON‑RPC `permission.request` 触发并交给 `on_permission_request` 回调处理，`session.on()` 只能订阅 `tool.user_requested`/`tool.execution_start` 等事件做记录关联（来源：本地源码 copilot/client.py、copilot/session.py、copilot/types.py）。  \n在非交互流里必须提供 `on_permission_request` 自动审批/拒绝，否则默认拒绝且回调异常也会被拒绝。\n\n要点/坑  \n- 配置键: `on_permission_request`（SessionConfig）  \n- 权限类型: `shell` / `write` / `mcp` / `read` / `url`  \n- 回调异常或缺失 -> `denied-no-approval-rule-and-could-not-request-from-user`\n\n```python\nimport asyncio\nfrom copilot import CopilotClient\nfrom copilot.generated.session_events import SessionEventType\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    def on_event(ev):\n        if ev.type == SessionEventType.TOOL_USER_REQUESTED:\n            print(f\"tool requested: {ev.data.tool_name} {ev.data.tool_call_id}\")\n        elif ev.type == SessionEventType.TOOL_EXECUTION_START:\n            print(f\"tool start: {ev.data.tool_name}\")\n\n    async def on_permission_request(req, ctx):\n        kind = req.get(\"kind\")\n        if kind in (\"read\", \"mcp\"):\n            return {\"kind\": \"approved\"}\n        return {\"kind\": \"denied-by-rules\", \"rules\": [{\"kind\": \"deny-non-read\"}]}\n\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"on_permission_request\": on_permission_request,\n    })\n    session.on(on_event)\n\n    await session.send_and_wait({\"prompt\": \"请调用需要权限的工具\"})\n    await client.stop()\n\nasyncio.run(main())\n```\n\n建议/落地  \n- pipe 启动时注册审批策略 + `session.on` 记录 `tool_call_id` 做审计", "model": "gpt-5.2-codex"}
+{"index": 3, "question": "Show how to implement a 'Device Flow' login helper using SDK primitives (if available) or raw HTTP showing how to persist credentials.", "answer": "SDK 本身没有 device‑flow 登录 API；可用鉴权原语只有 CopilotClient 的 `github_token`/`use_logged_in_user` 与 `get_auth_status`（源：本地源码 copilot/client.py、copilot/types.py）。  \n下面用 GitHub OAuth Device Flow 原始 HTTP 获取 token、持久化本地并注入 `github_token`（端点说明：https://docs.github.com/en/apps/oauth-apps/building-oauth-apps/authorizing-oauth-apps#device-flow）。\n\n```python\nimport asyncio\nimport json\nimport os\nimport time\nimport urllib.parse\nimport urllib.request\nfrom pathlib import Path\n\nfrom copilot import CopilotClient\n\nCLIENT_ID = \"YOUR_OAUTH_APP_CLIENT_ID\"\nSCOPE = \"read:user\"  # 按你的 OAuth App 需求调整\nTOKEN_PATH = Path.home() / \".config\" / \"myapp\" / \"copilot_token.json\"\n\ndef http_post(url, data):\n    body = urllib.parse.urlencode(data).encode()\n    req = urllib.request.Request(url, data=body, headers={\"Accept\": \"application/json\"})\n    with urllib.request.urlopen(req) as resp:\n        return json.loads(resp.read().decode())\n\ndef load_token():\n    if TOKEN_PATH.exists():\n        return json.loads(TOKEN_PATH.read_text()).get(\"access_token\")\n    return None\n\ndef save_token(token):\n    TOKEN_PATH.parent.mkdir(parents=True, exist_ok=True)\n    TOKEN_PATH.write_text(json.dumps(token))\n    os.chmod(TOKEN_PATH, 0o600)\n\ndef device_flow():\n    code = http_post(\n        \"https://github.com/login/device/code\",\n        {\"client_id\": CLIENT_ID, \"scope\": SCOPE},\n    )\n    print(f\"Open {code['verification_uri']} and enter {code['user_code']}\")\n    interval = int(code.get(\"interval\", 5))\n\n    while True:\n        token = http_post(\n            \"https://github.com/login/oauth/access_token\",\n            {\n                \"client_id\": CLIENT_ID,\n                \"device_code\": code[\"device_code\"],\n                \"grant_type\": \"urn:ietf:params:oauth:grant-type:device_code\",\n            },\n        )\n        if \"access_token\" in token:\n            return token\n        if token.get(\"error\") == \"slow_down\":\n            interval += 5\n        elif token.get(\"error\") not in (\"authorization_pending\",):\n            raise RuntimeError(token)\n        time.sleep(interval)\n\nasync def main():\n    access_token = load_token()\n    if not access_token:\n        token = device_flow()\n        save_token(token)\n        access_token = token[\"access_token\"]\n\n    client = CopilotClient({\"github_token\": access_token})\n    await client.start()\n    # ... create_session / send ...\n    await client.stop()\n\nif __name__ == \"__main__\":\n    asyncio.run(main())\n```\n\n要点/坑  \n- `cli_url` 与 `github_token` 互斥  \n- SDK 不保存凭据 需自行持久化  \n- 轮询需处理 `authorization_pending` / `slow_down`\n\n建议/落地  \n- pipe 启动时读取 token 文件 不存在则走 device flow  \n- 用 `github_token` 注入 保持非交互运行", "model": "gpt-5.2-codex"}
+{"index": 4, "question": "Compare the pros and cons of 'Session Replay' (fast-forwarding history) vs 'Session Resume' (stateful ID) for a stateless web backend like OpenWebUI.", "answer": "对无状态后端(OpenWebUI)而言，SDK 只支持通过 `resume_session(session_id)` 的状态续接；所谓“Session Replay”需要你自行重放历史 prompt（来源：本地 SDK 源码 copilot/client.py、copilot/session.py）。  \n对比要点（短语）：  \n- Replay：无状态/易扩展；缺点是耗时耗 tokens、易截断/丢工具状态、可能重复副作用  \n- Resume：省 tokens、保留工具/compaction/`workspace_path` 状态；缺点是需持久化 session_id、依赖会话存活/跨实例一致性。  \n建议/落地：优先持久化 chat_id→session_id 并 resume_session，失效时回退为重放最近 N 轮并重建工具上下文。", "model": "gpt-5.2-codex"}
+{"index": 5, "question": "Detail the exact protocol for 'Virtual Workspace': how to implement a file system provider that feeds content to Copilot without physical files.", "answer": "SDK 源码里没有“Virtual Workspace”协议或文件系统 provider 接口；会话只识别真实路径（`SessionConfig.working_directory`、`MessageOptions.attachments`）。要在无物理文件情况下供给内容，只能用自定义工具或 MCP 服务器把“读写文件”变成 tool 调用（来源：本地 SDK 源码 copilot/types.py / copilot/client.py）。\n\n可用原语（源码）\n- `SessionConfig.tools` / `available_tools`：自定义工具与白名单\n- `MessageOptions.attachments`：仅支持真实路径文件/目录\n\n方案 A：自定义工具实现 VFS（推荐）\n```python\nimport asyncio\nfrom pydantic import BaseModel, Field\nfrom copilot import CopilotClient, define_tool\n\nVFS = {\"README.md\": \"# Virtual README\\n内容来自内存\\n\"}\n\nclass VfsReadParams(BaseModel):\n    path: str = Field(description=\"virtual path\")\n\n@define_tool(name=\"vfs_read\", description=\"Read a virtual file\")\ndef vfs_read(params: VfsReadParams) -> str:\n    return VFS.get(params.path, \"\")\n\n@define_tool(name=\"vfs_list\", description=\"List virtual files\")\ndef vfs_list() -> list[str]:\n    return sorted(VFS.keys())\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"tools\": [vfs_read, vfs_list],\n        \"available_tools\": [\"vfs_read\", \"vfs_list\"],\n        \"system_message\": {\"content\": \"Use vfs_* tools for all file access.\"},\n    })\n    await session.send_and_wait({\"prompt\": \"打开 README.md\"})\n    await client.stop()\n\nasyncio.run(main())\n```\n\n方案 B：MCP 服务器（本地 stdio，VFS 由你实现）\n```python\nsession = await client.create_session({\n    \"model\": \"gpt-5\",\n    \"mcp_servers\": {\n        \"vfs\": {\n            \"type\": \"local\",\n            \"command\": \"python\",\n            \"args\": [\"-m\", \"my_vfs_mcp_server\"],\n            \"tools\": [\"*\"],\n        }\n    },\n})\n```\n\n注意/坑  \n- `available_tools` 只保留你的 VFS 工具以强制使用  \n- 内置文件工具只读真实文件；要兼容只能落地临时文件  \n- 大文件需分块返回，避免上下文爆炸  \n\n建议/落地  \n- pipe 内维护 VFS 存储（DB/Redis/内存）  \n- 启动 session 时注入 `system_message` + `available_tools` 强制 VFS 入口", "model": "gpt-5.2-codex"}
+{"index": 6, "question": "Create an 'Agent Manifest' example: how to define an Agent capable of specific high-privileged tools via SDK.", "answer": "SDK 没有独立“Agent Manifest”文件格式，等价于 `create_session` 的 `custom_agents`（CustomAgentConfig），并用 `tools`/`available_tools` 约束高权限工具，权限由 `on_permission_request` 处理（来源：本地 SDK 源码 copilot/types.py、copilot/client.py）。  \n示例如下（manifest 为 Python dict，可序列化为 JSON 配置）。\n\n```python\nimport asyncio\nfrom pydantic import BaseModel, Field\nfrom copilot import CopilotClient, define_tool\n\nclass AdminShellParams(BaseModel):\n    cmd: str = Field(description=\"approved shell command\")\n\n@define_tool(name=\"admin_shell\", description=\"Run an approved shell command\")\ndef admin_shell(params: AdminShellParams) -> str:\n    # TODO: 执行受控命令（白名单/沙箱）\n    return f\"approved: {params.cmd}\"\n\nclass AdminWriteParams(BaseModel):\n    path: str = Field(description=\"target path\")\n    content: str = Field(description=\"content\")\n\n@define_tool(name=\"admin_write\", description=\"Write to privileged store\")\ndef admin_write(params: AdminWriteParams) -> str:\n    # TODO: 受控写入\n    return \"ok\"\n\nasync def on_permission_request(req, _ctx):\n    if req.get(\"kind\") in (\"shell\", \"write\"):\n        return {\"kind\": \"approved\"}\n    return {\"kind\": \"denied-by-rules\", \"rules\": [{\"kind\": \"deny\"}]}\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    agent_manifest = {\n        \"name\": \"ops-admin\",\n        \"display_name\": \"Ops Admin\",\n        \"description\": \"Privileged ops agent\",\n        \"prompt\": \"Use only admin_* tools for privileged ops.\",\n        \"tools\": [\"admin_shell\", \"admin_write\"],\n        \"infer\": True,\n    }\n\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"tools\": [admin_shell, admin_write],\n        \"custom_agents\": [agent_manifest],\n        \"available_tools\": [\"admin_shell\", \"admin_write\"],\n        \"on_permission_request\": on_permission_request,\n    })\n\n    await session.send_and_wait({\"prompt\": \"使用 ops-admin 执行受控操作\"})\n    await client.stop()\n\nasyncio.run(main())\n```\n\n要点/坑  \n- session.tools 未注册 -> agent.tools 不可用  \n- custom_agents.tools 仅写名称  \n- 非交互缺少 on_permission_request -> 默认拒绝  \n- SDK 无显式 agent 选择参数 -> 需在 prompt 指定  \n\n建议/落地  \n- pipe 启动时加载 manifest 注入 custom_agents + available_tools + on_permission_request", "model": "gpt-5.2-codex"}
+{"index": 7, "question": "List all 'hidden' `SessionConfig` parameters relevant to Agent behavior and personality.", "answer": "基于本地 SDK 源码 `copilot/types.py` 的 SessionConfig，影响 Agent 行为/人格的相关参数如下（来源：本地 SDK 源码）。  \n- `system_message`: `append`/`replace`；`replace` 会移除 SDK guardrails（风险）  \n- `custom_agents`: `prompt`/`description`/`display_name`/`tools`/`mcp_servers`/`infer`；`tools` 仅名称且必须在 `tools` 注册  \n- `tools` / `available_tools` / `excluded_tools`: 能力与行为边界；`available_tools` 优先于 `excluded_tools`  \n- `hooks`: `on_user_prompt_submitted`/`on_pre_tool_use`/`on_post_tool_use`/`on_session_start`/`on_session_end`/`on_error_occurred`，可改写 prompt/工具参数/结果  \n- `on_permission_request`: 非交互审批策略；缺失或异常 => 拒绝  \n- `on_user_input_request`: 启用/处理 `ask_user`  \n- `mcp_servers`: 注入外部工具集  \n- `skill_directories` / `disabled_skills`: 技能启用/禁用  \n- `model` / `provider`: 模型与 BYOK 影响输出风格与能力  \n- `working_directory`: 上下文范围  \n- `infinite_sessions`: compaction 阈值影响记忆保留  \n- `config_dir`: CLI 配置/状态目录，行为取决于 CLI 侧配置  \n\n建议/落地：在 pipe 启动时固定 `system_message` + `available_tools/excluded_tools` + `hooks`，并持久化 session_id 以便 `resume_session`。", "model": "gpt-5.2-codex"}
+{"index": 8, "question": "Give a concise overview of MCP Server Integration: JSON-RPC config and SDK hooks.", "answer": "SDK 中 MCP 通过 create_session/resume_session 的 mcp_servers 传入（映射到 JSON‑RPC 的 mcpServers），支持 local/stdio 与 http/sse，两类配置字段分别为 type/command/args/env/cwd/timeout/tools 或 type/url/headers/timeout/tools（来源：本地 SDK 源码）。  \nSDK hooks 使用 hooks(on_pre_tool_use/on_post_tool_use/on_user_prompt_submitted/on_session_start/on_session_end/on_error_occurred) 进行拦截与附加上下文，权限需 on_permission_request 对 kind='mcp' 明确批准，且 session.on 可审计 tool.execution_* 事件中的 mcp_server_name/mcp_tool_name。  \n示例见下，建议/落地：pipe 中启用 on_permission_request、限制 tools 白名单并记录 MCP 事件。\n\n```python\nimport asyncio\nfrom copilot import CopilotClient\n\nasync def on_permission_request(req, _):\n    return {\"kind\": \"approved\"} if req.get(\"kind\") == \"mcp\" else {\n        \"kind\": \"denied-by-rules\",\n        \"rules\": [{\"kind\": \"deny\"}],\n    }\n\nasync def on_pre_tool_use(input, _):\n    if input.get(\"toolName\", \"\").startswith(\"mcp.\"):\n        return {\"additionalContext\": \"MCP tool call observed\"}\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"mcp_servers\": {\n            \"brave\": {\n                \"type\": \"local\",\n                \"command\": \"/path/to/brave-mcp\",\n                \"args\": [],\n                \"env\": {\"BRAVE_API_KEY\": \"YOUR_KEY\"},\n                \"tools\": [\"*\"],\n                \"timeout\": 30000,\n            }\n        },\n        \"hooks\": {\"on_pre_tool_use\": on_pre_tool_use},\n        \"on_permission_request\": on_permission_request,\n    })\n\n    await session.send_and_wait({\"prompt\": \"用 Brave Search 查询 Copilot SDK 新闻\"})\n    await client.stop()\n\nasyncio.run(main())\n```", "model": "gpt-5.2-codex"}
+{"index": 9, "question": "Provide best practices and common pitfalls for MCP Server Integration: JSON-RPC config and SDK hooks.", "answer": "基于本地 SDK 源码，MCP 配置通过 `create_session`/`resume_session` 的 `mcp_servers` 进入 JSON‑RPC `mcpServers`，并通过 `on_permission_request` 与 `hooks` 控制工具审批与行为（来源：本地 SDK 源码）  \n- 最佳实践：local/stdio 用 `type/command/args/env/cwd/timeout/tools`；remote http/sse 用 `type/url/headers/timeout/tools`；`tools` 尽量白名单、必要时 `[\"*\"]`；`timeout` 毫秒；`hooks.on_pre_tool_use` 做改参/拒绝，`session.on` 记录 `tool.execution_*` 与 `mcp_server_name`/`mcp_tool_name`  \n- 常见坑：把 MCP 配置放在 CopilotClient；缺少 `on_permission_request`（kind=\"mcp\"）导致默认拒绝；`tools: []` 等于禁用全部；设置 `available_tools` 却未包含 MCP 工具名导致被过滤  \n\n建议/落地：pipe 中固定审批策略 + 统一 MCP 配置模板 + 审计日志", "model": "gpt-5.2-codex"}
+{"index": 10, "question": "Show a minimal example snippet for MCP Server Integration: JSON-RPC config and SDK hooks.", "answer": "基于本地 SDK 源码，MCP 通过 `mcp_servers` 传入并用 `hooks` 处理工具调用（来源：本地 SDK 源码）。最小示例：  \n```python\nimport asyncio\nfrom copilot import CopilotClient\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"mcp_servers\": {\n            \"brave\": {\n                \"type\": \"local\",\n                \"command\": \"/path/to/brave-mcp\",\n                \"args\": [],\n                \"env\": {\"BRAVE_API_KEY\": \"YOUR_KEY\"},\n                \"tools\": [\"*\"],\n            }\n        },\n        \"hooks\": {\n            \"on_pre_tool_use\": lambda input, _: {\"additionalContext\": \"MCP tool call\"},\n        },\n        \"on_permission_request\": lambda req, _: {\"kind\": \"approved\"},\n    })\n\n    await session.send_and_wait({\"prompt\": \"用 Brave Search 搜索 Copilot SDK\"})\n    await client.stop()\n\nasyncio.run(main())\n```\n\n建议/落地：固定 `on_permission_request`（kind=\"mcp\"）并限制 `tools` 白名单。", "model": "gpt-5.2-codex"}
+{"index": 11, "question": "List recommended configuration defaults for MCP Server Integration: JSON-RPC config and SDK hooks.", "answer": "推荐默认值（MCP JSON‑RPC 配置，来源：本地 SDK 源码）：local/stdio 服务器用 `type=\"local\"/\"stdio\"`（远程用 `http`/`sse`）、`tools` 白名单（无则 `\"*\"`）、`timeout=30000` 毫秒、`env` 仅放密钥、`cwd` 指向受控目录。  \nHooks 默认：`on_pre_tool_use` 追加上下文并可设 `permissionDecision`（ask/allow）、必要时改参，`on_post_tool_use` 记录结果，`on_error_occurred` 设置 retry/skip。  \n建议/落地：非交互场景配置 `on_permission_request` 仅放行 `kind=\"mcp\"`，否则拒绝；工具名不确定时先不设 `available_tools`，避免误过滤。", "model": "gpt-5.2-codex"}
+{"index": 12, "question": "How does MCP Server Integration: JSON-RPC config and SDK hooks relate to building a custom Agent?", "answer": "SDK 源码中，自定义 Agent 通过 `SessionConfig.custom_agents` 定义，`CustomAgentConfig` 支持 `mcp_servers`，因此“Agent 绑定 MCP”就是在 agent config 里声明 `mcp_servers`，底层在 `create_session` 时转成 JSON‑RPC 的 `mcpServers`（来源：本地 SDK 源码 copilot/types.py、copilot/client.py）。  \nHooks 是会话级拦截器（`on_pre_tool_use`/`on_post_tool_use` 等），对 MCP 工具同样生效；权限确认需 `on_permission_request` 审批 `kind=\"mcp\"`。  \n建议/落地：pipe 中为目标 Agent 绑定 MCP + 审批钩子，示例：  \n```python\nsession = await client.create_session({\n    \"model\": \"gpt-5\",\n    \"custom_agents\": [{\n        \"name\": \"search\",\n        \"prompt\": \"Use MCP for search.\",\n        \"mcp_servers\": {\n            \"brave\": {\n                \"type\": \"local\",\n                \"command\": \"/path/to/brave-mcp\",\n                \"args\": [],\n                \"env\": {\"BRAVE_API_KEY\": \"YOUR_KEY\"},\n                \"tools\": [\"*\"],\n            }\n        },\n    }],\n    \"hooks\": {\"on_pre_tool_use\": lambda i, _: {\"additionalContext\": \"mcp\"}},\n    \"on_permission_request\": lambda r, _: {\"kind\": \"approved\"},\n})\n```", "model": "gpt-5.2-codex"}
+{"index": 13, "question": "Give a concise overview of Agent Manifests: Defining capabilities and permissions programmatically.", "answer": "SDK 没有独立的 Agent Manifest 文件，等价做法是用 `SessionConfig.custom_agents` 描述 agent，并用 `tools`/`available_tools`/`excluded_tools` 限定能力。  \n权限通过 `on_permission_request` 与 `hooks`（如 `on_pre_tool_use`）在会话级控制，且 `custom_agents.tools` 只能写工具名并需在 `SessionConfig.tools` 注册。  \n建议/落地：pipe 启动时加载 manifest→注入 `custom_agents` + `available_tools` + `on_permission_request` 固化权限边界。", "model": "gpt-5.2-codex"}
+{"index": 14, "question": "Provide best practices and common pitfalls for Agent Manifests: Defining capabilities and permissions programmatically.", "answer": "SDK 没有独立 Manifest 文件，等价于 `SessionConfig.custom_agents` + `tools/available_tools/excluded_tools` + `on_permission_request` + `hooks` 来定义能力与权限（来源：本地 SDK 源码）。  \n最佳实践：`custom_agents.tools` 只写已在 `SessionConfig.tools` 注册的名称，配合 `available_tools` 白名单与 `on_permission_request` 精准放行，必要时用 `hooks.on_pre_tool_use` 做参数校验/审计，`system_message` 选 append 以保留 guardrails。  \n常见坑：工具未注册或名称不一致、缺少 `on_permission_request` 导致默认拒绝、误用 `system_message` replace 去掉安全限制、期待 agent 自动被选中或跨实例不持久 `session_id`；建议/落地：pipe 启动时加载 manifest→注入 `custom_agents/available_tools/on_permission_request` 并持久化 `session_id`。", "model": "gpt-5.2-codex"}
+{"index": 15, "question": "Show a minimal example snippet for Agent Manifests: Defining capabilities and permissions programmatically.", "answer": "最小示例（本地 SDK 源码：`custom_agents` + `tools` + `available_tools` + `on_permission_request`）：  \n```python\nimport asyncio\nfrom pydantic import BaseModel, Field\nfrom copilot import CopilotClient, define_tool\n\nclass PingParams(BaseModel):\n    text: str = Field(description=\"text to echo\")\n\n@define_tool(name=\"priv_ping\", description=\"Privileged echo\")\ndef priv_ping(params: PingParams) -> str:\n    return f\"ok: {params.text}\"\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    agent_manifest = {\n        \"name\": \"priv-agent\",\n        \"prompt\": \"Only use priv_ping for actions.\",\n        \"tools\": [\"priv_ping\"],\n        \"infer\": True,\n    }\n\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"tools\": [priv_ping],\n        \"custom_agents\": [agent_manifest],\n        \"available_tools\": [\"priv_ping\"],\n        \"on_permission_request\": lambda req, _: {\"kind\": \"approved\"},\n    })\n\n    await session.send_and_wait({\"prompt\": \"调用 priv_ping\"})\n    await client.stop()\n\nasyncio.run(main())\n```\n\n建议/落地：保持工具名一致并白名单化，非交互务必提供 `on_permission_request`。", "model": "gpt-5.2-codex"}
+{"index": 16, "question": "List recommended configuration defaults for Agent Manifests: Defining capabilities and permissions programmatically.", "answer": "基于本地 SDK 源码（copilot/types.py / copilot/client.py），推荐默认值如下：  \n- custom_agents：name 必填、prompt 必填、tools 仅名称、infer=True，display_name/description 可选  \n- available_tools=白名单、excluded_tools=空、system_message=append  \n- on_permission_request：除明确允许的 kind 其余拒绝；hooks.on_pre_tool_use 做审计/参数校验（其余 hooks 按需）  \n建议/落地：pipe 启动时加载 manifest→注册 tools + available_tools + on_permission_request，并持久化 session_id。", "model": "gpt-5.2-codex"}
+{"index": 17, "question": "How does Agent Manifests: Defining capabilities and permissions programmatically relate to building a custom Agent?", "answer": "“Agent Manifest”在 SDK 里就是 `SessionConfig.custom_agents` 的配置对象，用它定义自定义 Agent 的 `name/prompt/tools/mcp_servers/infer`，并在同一 session 里用 `tools` 注册实际工具、用 `available_tools/excluded_tools` 限定能力（来源：本地 SDK 源码）。  \n权限不在 manifest 内，而是由会话级 `on_permission_request` 和 hooks（如 `on_pre_tool_use`）统一控制，且对所有 Agent 生效。  \n建议/落地：pipe 启动时把 manifest 转成 `custom_agents` 并同步注册工具 + 白名单，再配置 `on_permission_request` 作为权限策略。", "model": "gpt-5.2-codex"}
+{"index": 18, "question": "Give a concise overview of Headless Auth: Device Code Flow and credential persistence.", "answer": "SDK 仅提供 `CopilotClient` 的 `github_token/use_logged_in_user` 与 `get_auth_status()`，没有内置 device‑flow API（源码：copilot/client.py、copilot/types.py）。  \nHeadless Device Flow 需走 GitHub OAuth Device Flow HTTP 端点获取 token（https://docs.github.com/en/apps/oauth-apps/building-oauth-apps/authorizing-oauth-apps#device-flow），将 token 持久化（权限 600）并通过 `CopilotClient({\"github_token\": token})` 注入，且与 `cli_url` 互斥。  \n建议/落地：启动时读 token 文件，不存在则执行 device flow 并保存，再启动客户端。\n\n```python\ntoken = load_token() or device_flow()  # 调用 device code + access_token 端点\nsave_token(token)  # chmod 600\nclient = CopilotClient({\"github_token\": token[\"access_token\"]})\nawait client.start()\n```", "model": "gpt-5.2-codex"}
+{"index": 19, "question": "Provide best practices and common pitfalls for Headless Auth: Device Code Flow and credential persistence.", "answer": "SDK 不内置 device flow，只能用 GitHub OAuth Device Flow HTTP 换取 token 并通过 CopilotClient({github_token}) 注入，且与 cli_url 互斥（源码：copilot/client.py、copilot/types.py）。  \n最佳实践/坑：- token 文件 chmod 600 或 OS keychain；- 记录到期与刷新策略，启动时用 get_auth_status 验证；- 轮询需处理 authorization_pending/slow_down；- 未持久化、use_logged_in_user 与 github_token 同时使用或与 cli_url 混用会失败（端点说明：https://docs.github.com/en/apps/oauth-apps/building-oauth-apps/authorizing-oauth-apps#device-flow）。  \n建议/落地：pipe 启动时读 token 文件，不存在则走 device flow 并保存，失败时回退重新授权。", "model": "gpt-5.2-codex"}
+{"index": 20, "question": "Show a minimal example snippet for Headless Auth: Device Code Flow and credential persistence.", "answer": "最小示例：用 GitHub Device Flow（端点说明见 https://docs.github.com/en/apps/oauth-apps/building-oauth-apps/authorizing-oauth-apps#device-flow）拿 token，持久化后注入 CopilotClient({github_token}).  \n```python\nimport asyncio\nimport json\nimport os\nimport time\nimport urllib.parse\nimport urllib.request\nfrom pathlib import Path\nfrom copilot import CopilotClient\n\nCLIENT_ID = \"YOUR_OAUTH_APP_CLIENT_ID\"\nTOKEN_PATH = Path.home() / \".config\" / \"myapp\" / \"copilot_token.json\"\n\ndef post(url, data):\n    body = urllib.parse.urlencode(data).encode()\n    req = urllib.request.Request(url, data=body, headers={\"Accept\": \"application/json\"})\n    with urllib.request.urlopen(req) as resp:\n        return json.loads(resp.read().decode())\n\ndef device_flow():\n    code = post(\"https://github.com/login/device/code\", {\"client_id\": CLIENT_ID, \"scope\": \"read:user\"})\n    print(f\"Open {code['verification_uri']} and enter {code['user_code']}\")\n    interval = int(code.get(\"interval\", 5))\n    while True:\n        tok = post(\n            \"https://github.com/login/oauth/access_token\",\n            {\n                \"client_id\": CLIENT_ID,\n                \"device_code\": code[\"device_code\"],\n                \"grant_type\": \"urn:ietf:params:oauth:grant-type:device_code\",\n            },\n        )\n        if \"access_token\" in tok:\n            return tok\n        if tok.get(\"error\") == \"slow_down\":\n            interval += 5\n        elif tok.get(\"error\") != \"authorization_pending\":\n            raise RuntimeError(tok)\n        time.sleep(interval)\n\ndef load_token():\n    return json.loads(TOKEN_PATH.read_text()).get(\"access_token\") if TOKEN_PATH.exists() else None\n\ndef save_token(tok):\n    TOKEN_PATH.parent.mkdir(parents=True, exist_ok=True)\n    TOKEN_PATH.write_text(json.dumps(tok))\n    os.chmod(TOKEN_PATH, 0o600)\n\nasync def main():\n    access_token = load_token()\n    if not access_token:\n        tok = device_flow()\n        save_token(tok)\n        access_token = tok[\"access_token\"]\n\n    client = CopilotClient({\"github_token\": access_token})\n    await client.start()\n    # ... create_session / send ...\n    await client.stop()\n\nasyncio.run(main())\n```\n\n建议/落地  \n- 启动先读 token 文件；缺失再 device flow  \n- token 文件权限 600  \n- 失效则重新授权", "model": "gpt-5.2-codex"}
+{"index": 21, "question": "List recommended configuration defaults for Headless Auth: Device Code Flow and credential persistence.", "answer": "基于本地 SDK 源码（`github_token` 与 `cli_url` 互斥）与 GitHub Device Flow 文档（https://docs.github.com/en/apps/oauth-apps/building-oauth-apps/authorizing-oauth-apps#device-flow），推荐默认值：  \n- Token 持久化：`~/.config/<app>/copilot_token.json` 或系统密钥链，文件权限 600  \n- Client 选项：仅 `github_token`，`use_logged_in_user` 留空/False，禁用 `cli_url`  \n- 轮询：使用 device code 返回的 `interval`（缺省 5s），处理 `authorization_pending`/`slow_down`  \n- 启动校验：`get_auth_status()` 失败即重新授权  \n\n建议/落地：启动先读 token 文件 → 无则 device flow → 保存 → `client.start()`", "model": "gpt-5.2-codex"}
+{"index": 22, "question": "How does Headless Auth: Device Code Flow and credential persistence relate to building a custom Agent?", "answer": "Headless Auth 只负责拿到并持久化 `github_token`，让 `CopilotClient` 能在无交互环境启动会话并加载 `custom_agents`；因此它是自定义 Agent 能稳定运行的前置条件，而不是 Agent 定义本身（来源：本地 SDK 源码）。  \n建议/落地：在无状态后端启动时读取 token 文件（不存在就走 device flow）→ `CopilotClient({\"github_token\": token})` → `create_session(custom_agents=...)`。", "model": "gpt-5.2-codex"}
+{"index": 23, "question": "Give a concise overview of Session Replay vs Resume: Handling stateless frontend history.", "answer": "基于本地 SDK 源码（copilot/client.py、copilot/session.py），Resume 是 `resume_session(session_id)` 保留会话状态，Replay 只能由应用重放历史 prompt。优缺点：Resume 省 tokens/保工具与 compaction 状态但依赖持久化 session_id；Replay 无状态易扩展但成本高且易丢副作用控制。建议/落地：无状态前端请保存 session_id 并优先 Resume，失效时回放最近 N 轮。", "model": "gpt-5.2-codex"}
+{"index": 24, "question": "Provide best practices and common pitfalls for Session Replay vs Resume: Handling stateless frontend history.", "answer": "最佳实践：在无状态前端每次请求携带 session_id 并优先 `resume_session`，同时恢复 tools/hooks/on_permission_request/system_message 以保持行为一致（来源：本地 SDK 源码）。  \n常见坑：Replay 会重复执行工具且耗 tokens；Resume 未重新注册工具或权限处理器会出现 tool not supported/默认拒绝；会话失效未回退。  \n建议/落地：持久化 chat_id→session_id，resume 失败时仅回放最近 N 轮并禁用/幂等化有副作用的工具。", "model": "gpt-5.2-codex"}
+{"index": 25, "question": "Show a minimal example snippet for Session Replay vs Resume: Handling stateless frontend history.", "answer": "来源：本地 SDK 源码（copilot/client.py, copilot/session.py），最小示例如下（优先 Resume，失败则 Replay）。  \n```python\nimport asyncio\nfrom copilot import CopilotClient\n\nasync def handle_request(session_id: str | None, history: list[str], latest: str):\n    client = CopilotClient()\n    await client.start()\n\n    try:\n        # Resume: 复用会话状态\n        if not session_id:\n            raise RuntimeError(\"no session_id\")\n        session = await client.resume_session(session_id)\n    except Exception:\n        # Replay: 逐条重放历史\n        session = await client.create_session({\"model\": \"gpt-5\"})\n        session_id = session.session_id\n        for prompt in history:\n            await session.send_and_wait({\"prompt\": prompt})\n\n    response = await session.send_and_wait({\"prompt\": latest})\n    await client.stop()\n    return response.data.content, session_id\n\n# asyncio.run(handle_request(session_id, history, latest_prompt))\n```\n建议/落地：持久化 session_id，resume 失败时仅回放最近 N 轮。", "model": "gpt-5.2-codex"}
+{"index": 26, "question": "List recommended configuration defaults for Session Replay vs Resume: Handling stateless frontend history.", "answer": "基于本地 SDK 源码，推荐默认配置如下（Resume / Replay）：  \n- Resume：持久化 session_id；`resume_session({...})` 传回 tools、hooks、on_permission_request、working_directory、streaming、provider、mcp_servers、custom_agents、skill_directories、disabled_skills；`disable_resume` 保持默认 False  \n- Replay：`create_session({...})` 传回 model、system_message、tools、available_tools/excluded_tools、hooks、on_permission_request、working_directory、provider、mcp_servers、custom_agents、infinite_sessions(默认启用)；仅回放最近 N 轮；默认拒绝 write/shell 或用只读白名单", "model": "gpt-5.2-codex"}
+{"index": 27, "question": "How does Session Replay vs Resume: Handling stateless frontend history relate to building a custom Agent?", "answer": "Resume 会保留 session_id 关联的 custom_agents/工具状态，Replay 需要每次 create_session 时重新注入 custom_agents + tools，否则 persona/能力会丢失（来源：本地 SDK 源码）。对于无状态前端，Resume 更可靠且省 token，但要持久化 session_id 并在 resume 时重新注册 tools/hooks；Replay 则需回放历史并保证相同 custom_agents 配置。建议/落地：pipe 优先 resume_session，失败再 replay 最近 N 轮并重建 custom_agents/tools/权限策略。", "model": "gpt-5.2-codex"}
+{"index": 28, "question": "Give a concise overview of Advanced Session Hooks: Intercepting and modifying user prompts.", "answer": "拦截/修改用户提示要用 `SessionHooks.on_user_prompt_submitted`（不是 `session.on` 事件订阅），`input.prompt` 可改为 `output.modifiedPrompt` 并附加 `additionalContext`/`suppressOutput`，且需在 `create_session`/`resume_session` 传入 `hooks` 才会触发 `hooks.invoke`（来源：本地 SDK 源码 copilot/types.py、copilot/session.py、copilot/client.py）。  \n```python\nfrom copilot import CopilotClient\n\nasync def on_user_prompt_submitted(input, _ctx):\n    return {\"modifiedPrompt\": input[\"prompt\"].strip(), \"additionalContext\": \"normalize whitespace\"}\n\nsession = await client.create_session({\n    \"model\": \"gpt-5\",\n    \"hooks\": {\"on_user_prompt_submitted\": on_user_prompt_submitted},\n})\n```\n建议/落地：在 pipe 入口统一做 prompt 清洗/拼接上下文，并记录修改前后用于审计。", "model": "gpt-5.2-codex"}
+{"index": 29, "question": "Provide best practices and common pitfalls for Advanced Session Hooks: Intercepting and modifying user prompts.", "answer": "最佳实践：在 create_session/resume_session 传入 hooks.on_user_prompt_submitted，返回 {modifiedPrompt, additionalContext, suppressOutput}，修改保持幂等、最小化、可审计（来源：本地 SDK 源码 copilot/types.py、copilot/client.py、copilot/session.py）。  \n常见坑：忘记传 hooks 或 resume 时未重新传、handler 抛异常会被吞并返回 None、返回非 dict 或过度改写提示导致无效/偏航。  \n建议/落地：在 pipe 入口统一注册该 hook 并记录修改前后，保持 system_message 为 append。", "model": "gpt-5.2-codex"}
+{"index": 30, "question": "Show a minimal example snippet for Advanced Session Hooks: Intercepting and modifying user prompts.", "answer": "最小示例（来源：本地 SDK 源码）：  \n```python\nimport asyncio\nfrom copilot import CopilotClient\n\nasync def on_user_prompt_submitted(hook_input, _ctx):\n    return {\n        \"modifiedPrompt\": hook_input[\"prompt\"].strip(),\n        \"additionalContext\": \"normalized\",\n    }\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n    session = await client.create_session({\n        \"model\": \"gpt-5\",\n        \"hooks\": {\"on_user_prompt_submitted\": on_user_prompt_submitted},\n    })\n    await session.send_and_wait({\"prompt\": \"  hello  \"})\n    await client.stop()\n\nasyncio.run(main())\n```\n建议/落地：在 pipe 入口统一注册该 hook，并记录修改前后以便审计。", "model": "gpt-5.2-codex"}
+{"index": 31, "question": "List recommended configuration defaults for Advanced Session Hooks: Intercepting and modifying user prompts.", "answer": "- 推荐默认值（来源：本地 SDK 源码 copilot/types.py）：`hooks={\"on_user_prompt_submitted\": handler}`，handler 默认返回 `{\"modifiedPrompt\": input[\"prompt\"], \"additionalContext\": \"\", \"suppressOutput\": False}` 或 `None`（不改写）。  \n- `create_session`/`resume_session` 都要传同一 hooks；`system_message` 保持 append 以保留 guardrails。  \n- 建议/落地：pipe 入口统一注册该 hook 并记录原/改提示用于审计。", "model": "gpt-5.2-codex"}
+{"index": 32, "question": "How does Advanced Session Hooks: Intercepting and modifying user prompts relate to building a custom Agent?", "answer": "Hooks 是会话级拦截器，对所有自定义 Agent 都生效；因此它是定义 Agent 行为边界和输入规范化的“外层控制”，而不是 Agent Manifest 本身（来源：本地 SDK 源码 copilot/types.py、copilot/session.py）。  \n在自定义 Agent 场景下，`on_user_prompt_submitted` 可统一清洗/增强用户指令，再由 custom_agents 的 prompt 决定具体风格。  \n建议/落地：pipe 启动时同时注入 custom_agents + hooks，确保 resume 时同样注册 hooks。", "model": "gpt-5.2-codex"}
+{"index": 33, "question": "Give a concise overview of Workspace Virtualization: Handling CWD for remote/virtual files.", "answer": "SDK 只有 `SessionConfig.working_directory` 来约束 CWD，`attachments` 也要求真实路径，没有“虚拟工作区/FS provider”协议（来源：本地 SDK 源码 copilot/types.py、copilot/client.py）。要处理远程/虚拟文件，需用自定义工具或 MCP 服务器实现 VFS 读写，或把内容落地到 working_directory 下的临时文件再交给模型。建议/落地：pipe 固定 working_directory 映射到会话工作区，统一走 vfs_* 工具/MCP，且在 on_permission_request 限制 write/shell。", "model": "gpt-5.2-codex"}
+{"index": 34, "question": "Provide best practices and common pitfalls for Workspace Virtualization: Handling CWD for remote/virtual files.", "answer": "SDK 没有虚拟工作区/FS provider，仅支持 `SessionConfig.working_directory` 与真实路径 `attachments`（来源：本地 SDK 源码），因此远程/虚拟文件需落地或用自定义工具/MCP 提供内容。  \n最佳实践：固定 working_directory 到可控目录/会话 workspace，VFS 工具分块返回，配合 `available_tools`/`on_permission_request` 限权；常见坑：直接传虚拟路径/URL、working_directory 不存在、resume 未重注册工具/权限、一次性返回过大内容导致上下文截断。  \n建议/落地：pipe 入口做 remote→临时文件或 vfs_* 映射，并持久化路径索引。", "model": "gpt-5.2-codex"}
+{"index": 35, "question": "Show a minimal example snippet for Workspace Virtualization: Handling CWD for remote/virtual files.", "answer": "SDK 仅提供 `SessionConfig.working_directory` 与真实路径附件，没有虚拟 FS/provider（来源：本地 SDK 源码 copilot/types.py）。最小做法是把远程内容落地到 CWD 下再作为附件传入：  \n```python\nimport asyncio\nimport pathlib\nfrom copilot import CopilotClient\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    workdir = \"/tmp/owui\"\n    pathlib.Path(workdir).mkdir(parents=True, exist_ok=True)\n    remote_content = \"...\"\n    path = f\"{workdir}/remote.txt\"\n    pathlib.Path(path).write_text(remote_content)\n\n    session = await client.create_session({\"model\": \"gpt-5\", \"working_directory\": workdir})\n    await session.send_and_wait({\n        \"prompt\": \"Summarize remote.txt\",\n        \"attachments\": [{\"type\": \"file\", \"path\": path}],\n    })\n    await client.stop()\n\nasyncio.run(main())\n```\n建议/落地：pipe 入口固定 working_directory，落地临时文件后立即处理，并用 `on_permission_request` 限制 write/shell。", "model": "gpt-5.2-codex"}
+{"index": 36, "question": "List recommended configuration defaults for Workspace Virtualization: Handling CWD for remote/virtual files.", "answer": "推荐默认值（来源：本地 SDK 源码 copilot/types.py）：  \n- `working_directory`: 指向可控的会话目录（如 session.workspace_path 或 /tmp/<app>/<session_id>）  \n- `available_tools`: 只允许 vfs_* 或只读工具，避免任意 shell/write  \n- `on_permission_request`: 默认拒绝 write/shell，仅放行 read/mcp  \n- `attachments`: 仅真实路径文件；远程内容先落地  \n\n建议/落地：pipe 入口统一映射 remote→临时文件或 VFS 工具，并持久化映射表。", "model": "gpt-5.2-codex"}
+{"index": 37, "question": "How does Workspace Virtualization: Handling CWD for remote/virtual files relate to building a custom Agent?", "answer": "SDK 没有虚拟文件系统协议，custom agent 只能依赖 `working_directory` 和真实路径附件，因此处理远程/虚拟文件需通过自定义工具/MCP 并在 agent prompt/available_tools 中强制使用（来源：本地 SDK 源码 copilot/types.py、copilot/client.py）。  \n这使“工作区虚拟化”成为 agent 能力与权限设计的一部分，需要 `available_tools`/`on_permission_request`/hooks 来限定访问范围。  \n建议/落地：pipe 中固定 working_directory 到会话目录，注入 vfs_* 工具白名单并在 prompt 明确使用 VFS。", "model": "gpt-5.2-codex"}
+{"index": 38, "question": "Give a concise overview of Error Recovery: Handling session disconnects and re-auth.", "answer": "本地 SDK 源码中没有显式“重连”API，断线通常表现为请求异常或 `SessionEventType.SESSION_ERROR`，需要 `stop()/force_stop()` 后重新 `start()` 并 `resume_session`（失败则 replay）（来源：本地 SDK 源码）。  \n认证可用 `get_auth_status()` 检测，未认证时重新创建 `CopilotClient({\"github_token\": token})`；若 `cli_url` 连接外部服务器则认证由外部处理（来源：本地 SDK 源码）。\n\n```python\nimport asyncio\nfrom copilot import CopilotClient\nfrom copilot.generated.session_events import SessionEventType\n\nasync def safe_send(session_id, prompt, token):\n    client = CopilotClient({\"github_token\": token})\n    await client.start()\n    try:\n        session = await (client.resume_session(session_id) if session_id else client.create_session({\"model\": \"gpt-5\"}))\n        session.on(lambda ev: ev.type == SessionEventType.SESSION_ERROR and print(ev.data.message))\n        resp = await session.send_and_wait({\"prompt\": prompt})\n        return resp, session.session_id\n    except Exception:\n        await client.stop()\n        client = CopilotClient({\"github_token\": token})\n        await client.start()\n        session = await client.create_session({\"model\": \"gpt-5\"})\n        resp = await session.send_and_wait({\"prompt\": prompt})\n        return resp, session.session_id\n    finally:\n        await client.stop()\n```\n\n建议/落地：持久化 session_id 与 token，优先 resume，失败再回放最近 N 轮。", "model": "gpt-5.2-codex"}
+{"index": 39, "question": "Provide best practices and common pitfalls for Error Recovery: Handling session disconnects and re-auth.", "answer": "最佳实践（基于本地 SDK 源码）：监听 `SESSION_ERROR`/捕获请求异常后 `stop()/force_stop()`→重新 `start()`→`resume_session`（失败再 replay），启动时用 `get_auth_status()` 校验并在失效时重建 `CopilotClient({\"github_token\": ...})`，同时在 resume 时重新注册 tools/hooks/权限处理器。  \n常见坑：把 `cli_url` 与 `github_token` 混用、忘记重注册工具导致 `tool not supported`、缺少 `on_permission_request` 导致默认拒绝、以及不回退到 replay 造成会话中断。  \n建议/落地：无状态后端持久化 `session_id+token`、实现幂等的重试/回放策略并记录错误上下文用于告警。", "model": "gpt-5.2-codex"}
+{"index": 40, "question": "Show a minimal example snippet for Error Recovery: Handling session disconnects and re-auth.", "answer": "最小示例（本地 SDK 源码：copilot/client.py、copilot/session.py）：  \n```python\nimport asyncio\nfrom copilot import CopilotClient\n\nasync def send_with_recovery(session_id, prompt, token):\n    client = CopilotClient({\"github_token\": token})\n    await client.start()\n    try:\n        session = await (client.resume_session(session_id) if session_id else client.create_session({\"model\": \"gpt-5\"}))\n        resp = await session.send_and_wait({\"prompt\": prompt})\n        return resp.data.content, session.session_id\n    except Exception:\n        await client.stop()\n        client = CopilotClient({\"github_token\": token})\n        await client.start()\n        session = await client.create_session({\"model\": \"gpt-5\"})\n        resp = await session.send_and_wait({\"prompt\": prompt})\n        return resp.data.content, session.session_id\n    finally:\n        await client.stop()\n```\n建议/落地：持久化 token+session_id，resume 失败再重建并回放必要历史。", "model": "gpt-5.2-codex"}
+{"index": 41, "question": "List recommended configuration defaults for Error Recovery: Handling session disconnects and re-auth.", "answer": "推荐默认值（字段来源：本地 SDK 源码）：CopilotClientOptions—auto_start=True、log_level='info'、use_stdio=True（非 cli_url）、github_token=持久化 token；SessionConfig/ResumeSessionConfig—hooks.on_error_occurred=handler、on_permission_request=非交互审批、working_directory 固定、mcp_servers/custom_agents/streaming 与首次一致、可固定 config_dir 并持久化 session_id。  \n常见坑：cli_url 与 github_token 互斥、resume 时未重传 tools/hooks 导致 tool not supported 或默认拒绝。  \n建议/落地：pipe 保存 token+session_id，断线后先 resume_session，失败再回放最近 N 轮。", "model": "gpt-5.2-codex"}
+{"index": 42, "question": "How does Error Recovery: Handling session disconnects and re-auth relate to building a custom Agent?", "answer": "错误恢复与自定义 Agent 强相关：Agent 定义与工具/权限处理器都是 session 级配置，断线或重认证后必须重建客户端并重新注入 `custom_agents`、`tools`、`hooks`、`on_permission_request`，否则能力/人格会丢失（来源：本地 SDK 源码）。  \n建议/落地：持久化 session_id+token，重连优先 `resume_session` 并重新注册上述配置，失败再回放最近 N 轮。", "model": "gpt-5.2-codex"}
+{"index": 43, "question": "Give a concise overview of Confirmation Events: programmatic handling of 'confirmation_required'.", "answer": "SDK 源码里没有独立的 `confirmation_required` 事件；权限确认是 JSON‑RPC `permission.request`，通过 `SessionConfig.on_permission_request` 回调处理（缺失或异常会默认拒绝）。  \n`session.on()` 只能用于审计/关联工具事件（如 `tool.user_requested`、`tool.execution_start`），不能替代确认逻辑。", "model": "gpt-5.2-codex"}
+{"index": 44, "question": "Provide best practices and common pitfalls for Confirmation Events: programmatic handling of 'confirmation_required'.", "answer": "SDK 源码未定义 `confirmation_required` 事件，确认流程必须用 `SessionConfig.on_permission_request` 处理，缺失或异常会默认拒绝（来源：本地 SDK 源码）。  \n最佳实践/坑：  \n- ✅ 对 `kind` 建立白名单策略（如仅允许 `read/mcp`），记录决策并用 `session.on` 仅做审计关联 `tool.user_requested`  \n- ✅ `resume_session` 时重新注册 `on_permission_request`/hooks  \n- ❌ 误以为 `session.on()` 可完成确认、返回非规范 dict 或抛异常导致全部拒绝", "model": "gpt-5.2-codex"}
+{"index": 45, "question": "Show a minimal example snippet for Confirmation Events: programmatic handling of 'confirmation_required'.", "answer": "SDK 里没有 `confirmation_required` 事件，需用 `SessionConfig.on_permission_request` 处理权限确认。  \n最小示例：  \n```python\nfrom copilot import CopilotClient\n\nclient = CopilotClient()\nawait client.start()\n\nsession = await client.create_session({\n    \"model\": \"gpt-5\",\n    \"on_permission_request\": lambda req, _: {\"kind\": \"approved\"},\n})\n```\n建议/落地：在 pipe 中统一注册 on_permission_request，并在 resume_session 时重复注册。", "model": "gpt-5.2-codex"}
+{"index": 46, "question": "List recommended configuration defaults for Confirmation Events: programmatic handling of 'confirmation_required'.", "answer": "推荐默认值（SDK 无 confirmation_required 事件，改用 on_permission_request，来源：本地 SDK 源码）：- 对 kind 建白名单（常见只放行 read/mcp）- 未命中默认返回 denied-by-rules 并给 rules - resume_session 时重注册 on_permission_request/hooks。  \n建议/落地：pipe 启动即注入 on_permission_request 并记录审批日志。", "model": "gpt-5.2-codex"}
+{"index": 47, "question": "How does Confirmation Events: programmatic handling of 'confirmation_required' relate to building a custom Agent?", "answer": "SDK 没有 `confirmation_required` 事件，权限确认由 `create_session`/`resume_session` 的 `on_permission_request` 处理，且对同一 session 的所有 custom agent 生效（来源：本地 SDK 源码）。  \n建议/落地：在 pipe 启动时注入 `on_permission_request` 白名单策略并在 resume 时重注册，以约束高权限工具。", "model": "gpt-5.2-codex"}
+{"index": 48, "question": "Give a concise overview of Skills: Conflict resolution and precedence defaults.", "answer": "在本地 SDK 源码中，Skills 只通过 SessionConfig 的 `skill_directories` 与 `disabled_skills` 暴露，未定义冲突解析/优先级规则，因此行为取决于 CLI 运行时实现（来源：本地 SDK 源码 copilot/types.py）。  \n为可控性，建议确保技能名唯一、仅配置单一目录或由你自行按顺序拼接目录，并用 `disabled_skills` 明确屏蔽同名/不应启用的技能。  \n建议/落地：在无状态后端每次 `create_session`/`resume_session` 时显式传入同一份 `skill_directories` 与 `disabled_skills`，避免因默认加载顺序差异导致行为漂移。", "model": "gpt-5.2-codex"}
+{"index": 49, "question": "Provide best practices and common pitfalls for Skills: Conflict resolution and precedence defaults.", "answer": "SDK 源码仅暴露 SessionConfig/ResumeSessionConfig 的 `skill_directories` 和 `disabled_skills`，未定义冲突解析或优先级，行为由 CLI 侧决定（来源：本地 SDK 源码 copilot/types.py）。  \n最佳实践：技能名全局唯一、只配置单一目录或显式排序，并用 `disabled_skills` 禁用冲突项；常见坑是依赖默认加载顺序或 resume 未重传这两个字段导致行为漂移。  \n建议/落地：在 pipe 启动与 resume 时统一注入同一份 `skill_directories`/`disabled_skills` 并记录变更。", "model": "gpt-5.2-codex"}
+{"index": 50, "question": "Show a minimal example snippet for Skills: Conflict resolution and precedence defaults.", "answer": "SDK 里技能只通过 `skill_directories` 与 `disabled_skills` 配置，冲突/优先级未在源码定义，需自行避免同名。  \n最小示例：  \n```python\nsession = await client.create_session({\n    \"model\": \"gpt-5\",\n    \"skill_directories\": [\"/opt/skills\"],\n    \"disabled_skills\": [\"legacy_skill\"],\n})\n```  \n建议：技能名唯一并在 `resume_session` 时重传这两项。", "model": "gpt-5.2-codex"}
+{"index": 51, "question": "List recommended configuration defaults for Skills: Conflict resolution and precedence defaults.", "answer": "推荐默认值（基于 SDK 仅暴露 `skill_directories`/`disabled_skills`）：  \n- `skill_directories`: 明确设置为单一目录（或你自己控制顺序的列表），不要依赖 CLI 默认搜索顺序  \n- `disabled_skills`: 默认空列表；如存在同名/旧技能则显式禁用  \n- Resume/Replay：始终复用同一份配置以避免“隐式优先级”漂移", "model": "gpt-5.2-codex"}
+{"index": 52, "question": "How does Skills: Conflict resolution and precedence defaults relate to building a custom Agent?", "answer": "Skills 是会话级配置（`skill_directories`/`disabled_skills`），会影响所有自定义 Agent 的可用能力与提示行为；SDK 不定义冲突优先级，因此需要你在构建自定义 Agent 时确保技能命名唯一并显式禁用冲突项（来源：本地 SDK 源码）。  \n建议/落地：为 custom agent 固定同一份 skills 配置，并在 resume 时重复传入，避免行为漂移。", "model": "gpt-5.2-codex"}
+{"index": 53, "question": "Give a concise overview of Debugging: Tracing JSON-RPC traffic in the SDK.", "answer": "本地 SDK 源码里 JsonRpcClient 没有内置 JSON‑RPC trace，只能通过 `CopilotClient({\"log_level\":\"debug\"})` 打开 CLI 侧日志（非原始包）（来源：本地 SDK 源码）。  \n要抓实际 JSON‑RPC 流量，可 monkey‑patch `JsonRpcClient._send_message`/`_read_message` 或包装 stdio 进行 tee，最小示例如下。  \n建议/落地：在 pipe 中开启 patch 并对日志做脱敏与采样。  \n```python\nfrom copilot.jsonrpc import JsonRpcClient\n\norig_send = JsonRpcClient._send_message\nasync def _send(self, message):\n    print(\">>\", message)\n    return await orig_send(self, message)\nJsonRpcClient._send_message = _send\n\norig_read = JsonRpcClient._read_message\ndef _read(self):\n    msg = orig_read(self)\n    if msg:\n        print(\"<<\", msg)\n    return msg\nJsonRpcClient._read_message = _read\n```", "model": "gpt-5.2-codex"}
+{"index": 54, "question": "Provide best practices and common pitfalls for Debugging: Tracing JSON-RPC traffic in the SDK.", "answer": "源码里 JsonRpcClient 没有内置 trace，只能用 `CopilotClient({\"log_level\":\"debug\"})` 查看 CLI 侧日志或 monkey‑patch `JsonRpcClient._send_message/_read_message` 做 tee（来源：本地 SDK 源码）。  \n最佳实践：仅在调试启用、对 github_token/API key/内容脱敏、采样/限流、只读不改消息并注意线程安全；常见坑：记录完整 prompt/工具参数导致泄露，或同步打印阻塞 IO 造成超时/断连。", "model": "gpt-5.2-codex"}
+{"index": 55, "question": "Show a minimal example snippet for Debugging: Tracing JSON-RPC traffic in the SDK.", "answer": "最小示例（monkey‑patch `JsonRpcClient` 打印收发）：  \n```python\nfrom copilot.jsonrpc import JsonRpcClient\n\norig_send = JsonRpcClient._send_message\nasync def _send(self, msg):\n    print(\">>\", msg)\n    return await orig_send(self, msg)\nJsonRpcClient._send_message = _send\n\norig_read = JsonRpcClient._read_message\ndef _read(self):\n    msg = orig_read(self)\n    if msg:\n        print(\"<<\", msg)\n    return msg\nJsonRpcClient._read_message = _read\n```\n建议：仅在调试启用并脱敏日志。", "model": "gpt-5.2-codex"}
+{"index": 56, "question": "List recommended configuration defaults for Debugging: Tracing JSON-RPC traffic in the SDK.", "answer": "推荐默认值（基于 SDK 源码）：- 生产环境关闭 trace；- 调试时 `CopilotClient({\"log_level\":\"debug\"})`；- 如需 JSON‑RPC 报文，采用临时 monkey‑patch 并启用脱敏/采样；- 不修改消息内容，避免线程阻塞。  \n建议/落地：pipe 中通过环境变量开关调试与采样率。", "model": "gpt-5.2-codex"}
+{"index": 57, "question": "How does Debugging: Tracing JSON-RPC traffic in the SDK relate to building a custom Agent?", "answer": "JSON‑RPC trace 能看到 custom_agents 在 `session.create/session.resume` 的实际 payload，以及后续 `tool.call`/`permission.request`/`hooks.invoke` 是否按预期触发，从而验证自定义 Agent 的能力与权限边界（来源：本地 SDK 源码）。  \n建议/落地：仅在开发环境开启并脱敏日志；用它核对 customAgents/tools/权限决策与 MCP 配置。", "model": "gpt-5.2-codex"}
+{"index": 58, "question": "Give a concise overview of Billing & Policies: How seat management affects SDK features.", "answer": "- 访问与政策：SDK 依赖 Copilot CLI，需有效 Copilot 订阅且组织/企业未禁用 CLI，否则无法使用（源：https://raw.githubusercontent.com/github/copilot-sdk/main/README.md，https://raw.githubusercontent.com/github/copilot-cli/main/README.md）  \n- 计费与认证：每次 prompt 计入 Copilot premium requests 配额，PAT 需 Copilot Requests 权限（源：https://raw.githubusercontent.com/github/copilot-sdk/main/README.md，https://raw.githubusercontent.com/github/copilot-cli/main/README.md）\n\n建议/落地：启动先 `get_auth_status()`，失败时提示分配席位/启用 CLI 或更新 token", "model": "gpt-5.2-codex"}
+{"index": 59, "question": "Provide best practices and common pitfalls for Billing & Policies: How seat management affects SDK features.", "answer": "最佳实践：确认用户/组织席位允许 Copilot CLI/SDK 使用，启动前用 `get_auth_status()` 预检，使用 PAT 时确保具备 Copilot Requests 权限（源：https://raw.githubusercontent.com/github/copilot-sdk/main/README.md，https://raw.githubusercontent.com/github/copilot-cli/main/README.md）。  \n常见坑：未分配席位或组织/企业策略禁用 CLI 导致 SDK 失效、PAT 权限不足或 premium requests 用尽引发认证/配额错误（源：https://raw.githubusercontent.com/github/copilot-sdk/main/README.md，https://raw.githubusercontent.com/github/copilot-cli/main/README.md）。  \n建议/落地：在 pipe 启动时做席位与鉴权检测并给出明确提示/自助链接，记录每次请求计费与配额告警阈值以便限流。", "model": "gpt-5.2-codex"}
+{"index": 60, "question": "Show a minimal example snippet for Billing & Policies: How seat management affects SDK features.", "answer": "基于 SDK README，最小示例是在启动时用 `get_auth_status()` 检查是否已分配席位/可用认证（源：https://raw.githubusercontent.com/github/copilot-sdk/main/README.md）。  \n```python\nimport asyncio\nfrom copilot import CopilotClient\n\nasync def main():\n    client = CopilotClient()\n    await client.start()\n\n    auth = await client.get_auth_status()\n    if not auth.isAuthenticated:\n        raise RuntimeError(\"Copilot 未认证/未分配席位，请先登录或分配权限\")\n\n    # ...正常创建 session...\n    await client.stop()\n\nasyncio.run(main())\n```\n建议/落地：pipe 启动先做该检查，失败时提示分配席位或更新 PAT 权限。", "model": "gpt-5.2-codex"}
+{"index": 61, "question": "List recommended configuration defaults for Billing & Policies: How seat management affects SDK features.", "answer": "推荐默认值：CopilotClient 用 `use_logged_in_user=True`、`auto_start=True`、`log_level=\"info\"`；如用 PAT 则用 `github_token` 注入并确保具备 Copilot Requests 权限，且启动前 `get_auth_status()` 失败即按席位/策略限制处理（来源：https://raw.githubusercontent.com/github/copilot-sdk/main/README.md ，https://raw.githubusercontent.com/github/copilot-cli/main/README.md）。  \n建议/落地：pipe 启动先做 auth 预检并给出续订/分配席位提示，避免会话中途才触发权限或配额错误。", "model": "gpt-5.2-codex"}
+{"index": 62, "question": "How does Billing & Policies: How seat management affects SDK features relate to building a custom Agent?", "answer": "席位与组织策略决定 Copilot CLI/SDK 是否可用，因此自定义 Agent 只是同一会话上的配置，前提仍是有有效订阅与允许的策略，否则 Agent 无法运行（源：https://raw.githubusercontent.com/github/copilot-sdk/main/README.md ，https://raw.githubusercontent.com/github/copilot-cli/main/README.md）。  \n建议/落地：在创建自定义 Agent 前先 `get_auth_status()` 预检并对配额/政策限制做降级提示。", "model": "gpt-5.2-codex"}
+{"index": 63, "question": "How to configure MCP servers in ~/.copilot/config.json for SDK usage?", "answer": "我在 CLI 文档里未看到 `~/.copilot/config.json` 的 MCP 配置结构；CLI 仅提供 `/mcp add|edit|delete|show` 命令管理配置（来源：Copilot CLI 文档）。  \n因此建议用 CLI 写入配置，并在 SDK 中用 `config_dir` 指向该配置目录；或直接在 SDK 的 `create_session(mcp_servers=...)` 传入。  \n如果你确认存在 `~/.copilot/config.json` 的具体 schema，请提供示例或路径，我可按你的文件格式给出精确配置。", "model": "gpt-5.2-codex"}
+{"index": 64, "question": "What CLI environment variables force 'Agent' mode vs 'Generic' mode?", "answer": "CLI 文档未提到任何“通过环境变量强制 Agent 模式/Generic 模式”的开关（源：Copilot CLI 文档）。  \n已知与模式相关的是命令/参数，例如 `--agent` 选择自定义 agent，或交互内 `/agent` 切换；但不是环境变量。  \n如果你指的是内部实验开关，请提供具体变量名或版本线索，我可继续追索。", "model": "gpt-5.2-codex"}
+{"index": 65, "question": "Explain the 'confirmation' flow in CLI and how it maps to SDK events.", "answer": "- CLI：工具权限弹窗确认，可用 `/allow-all` 或 `/yolo` 自动批准。（源: https://raw.githubusercontent.com/github/copilot-cli/main/README.md）  \n- SDK：对应 JSON‑RPC `permission.request`→`on_permission_request` 返回 `approved/denied-*`，`session.on()` 只用于审计 tool.execution_*，无独立 confirmation 事件。（源: 本地 SDK 源码 copilot/client.py、copilot/session.py、copilot/types.py）\n\n示例：\n```python\nfrom copilot import CopilotClient\nfrom copilot.generated.session_events import SessionEventType\n\nclient = CopilotClient()\nawait client.start()\nsession = await client.create_session({\n    \"model\": \"gpt-5\",\n    \"on_permission_request\": lambda req, _: {\"kind\": \"approved\"},\n})\nsession.on(lambda ev: ev.type == SessionEventType.TOOL_EXECUTION_START and print(ev.data.tool_name))\n```\n\n建议/落地：非交互模式实现 `on_permission_request` 并记录 `tool.execution_*` 事件。", "model": "gpt-5.2-codex"}
+{"index": 66, "question": "Does the CLI support 'dry-run' permission checks for tools?", "answer": "CLI 文档未提到“dry-run”权限检查，只有正常确认与 `/allow-all`、`/yolo` 自动批准（源：https://raw.githubusercontent.com/github/copilot-cli/main/README.md）。SDK 侧仅有 `permission.request`→`on_permission_request` 与 `hooks.on_pre_tool_use` 的拦截/拒绝（本地 SDK 源码），不提供不执行的试运行；建议/落地：pipe 中用 `on_permission_request` 记录/拒绝高风险工具并在 hooks 做审计。", "model": "gpt-5.2-codex"}
+{"index": 67, "question": "What are the undocumented requirements for 'workspace' context updates?", "answer": "CLI 文档与本地 SDK 源码都未提到“workspace context updates”的未公开要求；SDK 只暴露 `working_directory` 与 `session.workspace_path`，以及 `attachments` 真实路径（来源：本地 SDK 源码）。  \n如果你指的是 CLI 的“工作区/会话文件”更新规则，请提供具体页面或日志片段，我可以进一步定位。", "model": "gpt-5.2-codex"}
+{"index": 68, "question": "How does the CLI handle 'device code' re-authentication automatically?", "answer": "CLI 文档未说明“自动 device code 重新认证”的具体机制；只提到未登录时使用 `/login`，PAT 可通过 `GH_TOKEN/GITHUB_TOKEN` 提供（源：Copilot CLI 文档）。因此无法确认是否有自动 re-auth 或缓存刷新策略。  \n若需无交互场景，请使用 SDK 的 `github_token` 注入并自行实现 device flow 轮询与持久化。", "model": "gpt-5.2-codex"}
--- a/plugins/debug/copilot_sdk_research/outputs/copilot_sdk_guide_20260130_114642.md
+++ b/plugins/debug/copilot_sdk_research/outputs/copilot_sdk_guide_20260130_114642.md
--- a/plugins/debug/copilot_sdk_research/test_multi_rule_permission.py
+++ b/plugins/debug/copilot_sdk_research/test_multi_rule_permission.py
@@ -0,0 +1,186 @@
+"""
+Test multi-rule permission control pattern (VSCode-style)
+Tests ordered rule matching like VSCode's chat.tools.terminal.autoApprove
+SAFE VERSION: Uses harmless commands (echo/ls) only. No rm, no git.
+"""
+
+import argparse
+import asyncio
+import json
+import logging
+import re
+from typing import Any, Dict, List, Tuple
+
+from copilot import CopilotClient
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s - %(levelname)s - %(message)s",
+)
+logger = logging.getLogger(__name__)
+
+
+def build_multi_rule_handler(rules_json: str):
+    """
+    Build permission handler with ordered rules (VSCode-style)
+    """
+    try:
+        rules = json.loads(rules_json) if rules_json else {}
+    except json.JSONDecodeError as e:
+        logger.error("Invalid rules JSON: %s", e)
+        rules = {}
+
+    async def on_permission_request(request: Dict[str, Any], context: Dict[str, str]):
+        kind = request.get("kind")
+        command = request.get("fullCommandText", "") or request.get("command", "")
+
+        # Always approve read and url
+        if kind in ("read", "url"):
+            return {"kind": "approved"}
+
+        # For shell commands, apply ordered rules
+        if kind == "shell" and command:
+            for pattern, approved in rules.items():
+                try:
+                    if re.match(pattern, command):
+                        if approved:
+                            logger.info(
+                                "✅ Approved (rule match): pattern=%r command=%r",
+                                pattern,
+                                command,
+                            )
+                            return {"kind": "approved"}
+                        else:
+                            logger.warning(
+                                "❌ Denied (rule match): pattern=%r command=%r",
+                                pattern,
+                                command,
+                            )
+                            return {
+                                "kind": "denied-by-rules",
+                                "rules": [
+                                    {"kind": "multi-rule-deny", "pattern": pattern}
+                                ],
+                            }
+                except re.error as exc:
+                    logger.error("Invalid pattern %r: %s", pattern, exc)
+                    continue
+
+        # Default deny for shell without matching rule
+        logger.warning("❌ Denied (no matching rule): command=%r", command)
+        return {"kind": "denied-by-rules", "rules": [{"kind": "no-rule-match"}]}
+
+    return on_permission_request
+
+
+async def run_test(model: str, rules_json: str, prompt: str) -> Tuple[bool, str]:
+    """Run a single test and return (approved, response)"""
+    try:
+        client = CopilotClient()
+        await client.start()
+
+        session = await client.create_session(
+            {
+                "model": model,
+                "on_permission_request": build_multi_rule_handler(rules_json),
+            }
+        )
+
+        # Set a short timeout
+        try:
+            response = await asyncio.wait_for(
+                session.send_and_wait({"prompt": prompt}), timeout=15.0
+            )
+        except asyncio.TimeoutError:
+            logger.error("Test Timed Out")
+            return (False, "Timeout")
+        finally:
+            await client.stop()
+
+        content = response.data.content
+        # Heuristics to detect denial in response
+        denied_keywords = [
+            "不允许",
+            "无法",
+            "对不起",
+            "Sorry",
+            "can't",
+            "cannot",
+            "not have permission",
+            "denied",
+        ]
+        is_denied = any(kw in content for kw in denied_keywords)
+
+        return (not is_denied, content)
+    except Exception as e:
+        logger.error("Test failed: %s", e)
+        return (False, str(e))
+
+
+async def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", default="gpt-4.1", help="Model ID")
+    args = parser.parse_args()
+
+    # LOGIC TEST RULES
+    # 1. Deny "echo secret" explicitly (Specific Deny)
+    # 2. Allow "echo" anything else (General Allow)
+    # 3. Allow "ls" (General Allow)
+    # 4. Deny everything else (Default Deny)
+    logic_test_rules = {
+        "^echo\\s+secret": False,  # Higher priority: Deny specific echo
+        "^echo": True,  # Lower priority: Allow general echo
+        "^ls": True,  # Allow ls
+        ".*": False,  # Deny everything else (e.g. whoami)
+    }
+
+    rules_json = json.dumps(logic_test_rules)
+
+    test_cases = [
+        # 1. Matches Rule 2 (^echo) -> Should be Approved
+        ("Allowed: Normal Echo", "请执行: echo 'hello world'", True),
+        # 2. Matches Rule 3 (^ls) -> Should be Approved
+        ("Allowed: LS", "请执行: ls -la", True),
+        # 3. Matches Rule 1 (^echo\s+secret) -> Should be DENIED
+        # This proves the ORDER matters. If it matched Rule 2 first, it would be allowed.
+        ("Denied: Restricted Echo", "请执行: echo secret data", False),
+        # 4. Matches Rule 4 (.*) -> Should be DENIED
+        ("Denied: Unknown Command", "请执行: whoami", False),
+    ]
+
+    logger.info("=" * 80)
+    logger.info("Safe Multi-Rule Logic Test (Proving Precedence)")
+    logger.info("Rules: %s", json.dumps(logic_test_rules, indent=2))
+    logger.info("=" * 80)
+
+    results = []
+    for i, (name, prompt, expected) in enumerate(test_cases, 1):
+        logger.info("\n[Test %d/%d] %s", i, len(test_cases), name)
+        logger.info("  Prompt: %s", prompt)
+
+        approved, response = await run_test(args.model, rules_json, prompt)
+        passed = approved == expected
+
+        status = "✅ PASS" if passed else "❌ FAIL"
+        results.append((name, passed))
+
+        logger.info(
+            "  Expected: %s, Got: %s - %s",
+            "Approved" if expected else "Denied",
+            "Approved" if approved else "Denied",
+            status,
+        )
+        logger.info("  Response: %s", response[:100].replace("\n", " "))
+
+    # Summary
+    logger.info("\n" + "=" * 80)
+    logger.info("Test Summary")
+    logger.info("=" * 80)
+    passed_count = sum(1 for _, passed in results if passed)
+    for name, passed in results:
+        logger.info("%s %s", "✅" if passed else "❌", name)
+    logger.info("Total: %d/%d tests passed", passed_count, len(results))
+
+
+if __name__ == "__main__":
+    asyncio.run(main())
--- a/plugins/debug/copilot_sdk_research/test_permission_comprehensive.py
+++ b/plugins/debug/copilot_sdk_research/test_permission_comprehensive.py
@@ -0,0 +1,202 @@
+"""
+Comprehensive Permission Control Test Suite
+Tests all permission control scenarios for GitHub Copilot SDK
+"""
+
+import argparse
+import asyncio
+import logging
+import re
+from typing import Any, Dict, List, Tuple
+
+from copilot import CopilotClient
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s - %(levelname)s - %(message)s",
+)
+logger = logging.getLogger(__name__)
+
+
+def build_permission_handler(allow_all: bool, allow_shell: bool, pattern: str):
+    async def on_permission_request(request: Dict[str, Any], context: Dict[str, str]):
+        kind = request.get("kind")
+        # Shell requests use 'fullCommandText' not 'command'
+        command = request.get("fullCommandText", "") or request.get("command", "")
+
+        if allow_all:
+            logger.info("✅ Approved (allow-all): kind=%s command=%r", kind, command)
+            return {"kind": "approved"}
+
+        if kind in ("read", "url"):
+            logger.info("✅ Approved (safe): kind=%s", kind)
+            return {"kind": "approved"}
+
+        if kind == "shell":
+            if allow_shell:
+                logger.info("✅ Approved (allow-shell): command=%r", command)
+                return {"kind": "approved"}
+
+            if pattern and command:
+                try:
+                    if re.match(pattern, command):
+                        logger.info(
+                            "✅ Approved (regex match): pattern=%r command=%r",
+                            pattern,
+                            command,
+                        )
+                        return {"kind": "approved"}
+                except re.error as exc:
+                    logger.error("Invalid regex pattern: %s (%s)", pattern, exc)
+
+        logger.warning("❌ Denied: kind=%s command=%r", kind, command)
+        return {"kind": "denied-by-rules", "rules": [{"kind": "test-suite"}]}
+
+    return on_permission_request
+
+
+async def run_test(
+    model: str, allow_all: bool, allow_shell: bool, pattern: str, prompt: str
+) -> Tuple[bool, str]:
+    """Run a single test and return (success, response)"""
+    try:
+        client = CopilotClient()
+        await client.start()
+
+        session = await client.create_session(
+            {
+                "model": model,
+                "on_permission_request": build_permission_handler(
+                    allow_all=allow_all,
+                    allow_shell=allow_shell,
+                    pattern=pattern,
+                ),
+            }
+        )
+
+        response = await session.send_and_wait({"prompt": prompt})
+        await client.stop()
+
+        content = response.data.content
+        # Check if response indicates success or denial
+        denied_keywords = [
+            "不允许",
+            "无法",
+            "对不起",
+            "Sorry",
+            "can't",
+            "cannot",
+            "not have permission",
+        ]
+        is_denied = any(kw in content for kw in denied_keywords)
+
+        return (not is_denied, content)
+    except Exception as e:
+        logger.error("Test failed with exception: %s", e)
+        return (False, str(e))
+
+
+async def main():
+    parser = argparse.ArgumentParser(
+        description="Comprehensive permission control test suite."
+    )
+    parser.add_argument("--model", default="gpt-4.1", help="Model ID for testing.")
+
+    args = parser.parse_args()
+
+    # Test cases: (name, allow_all, allow_shell, pattern, prompt, expected_approved)
+    test_cases = [
+        ("Default Deny Shell", False, False, "", "请执行: ls -la", False),
+        ("Allow All", True, False, "", "请执行: ls -la", True),
+        ("Allow Shell", False, True, "", "请执行: pwd", True),
+        ("Regex Match: ^ls", False, False, "^ls", "请执行: ls -la", True),
+        ("Regex No Match: ^ls vs pwd", False, False, "^ls", "请执行: pwd", False),
+        (
+            "Regex Complex: ^(ls|pwd|echo)",
+            False,
+            False,
+            "^(ls|pwd|echo)",
+            "请执行: pwd",
+            True,
+        ),
+        (
+            "Regex Complex No Match: git",
+            False,
+            False,
+            "^(ls|pwd|echo)",
+            "请执行: git status",
+            False,
+        ),
+        (
+            "Read Permission (Always Allow)",
+            False,
+            False,
+            "",
+            "Read the file: README.md",
+            True,
+        ),
+    ]
+
+    results = []
+    logger.info("=" * 80)
+    logger.info("Starting Comprehensive Permission Control Test Suite")
+    logger.info("Model: %s", args.model)
+    logger.info("=" * 80)
+
+    for i, (name, allow_all, allow_shell, pattern, prompt, expected) in enumerate(
+        test_cases, 1
+    ):
+        logger.info("\n[Test %d/%d] %s", i, len(test_cases), name)
+        logger.info(
+            "  Config: allow_all=%s, allow_shell=%s, pattern=%r",
+            allow_all,
+            allow_shell,
+            pattern,
+        )
+        logger.info("  Prompt: %s", prompt)
+
+        approved, response = await run_test(
+            args.model, allow_all, allow_shell, pattern, prompt
+        )
+        passed = approved == expected
+
+        status = "✅ PASS" if passed else "❌ FAIL"
+        results.append((name, passed))
+
+        logger.info(
+            "  Expected: %s, Got: %s - %s",
+            "Approved" if expected else "Denied",
+            "Approved" if approved else "Denied",
+            status,
+        )
+        logger.info(
+            "  Response: %s",
+            response[:100] + "..." if len(response) > 100 else response,
+        )
+
+    # Summary
+    logger.info("\n" + "=" * 80)
+    logger.info("Test Summary")
+    logger.info("=" * 80)
+    passed_count = sum(1 for _, passed in results if passed)
+    total_count = len(results)
+
+    for name, passed in results:
+        logger.info("%s %s", "✅" if passed else "❌", name)
+
+    logger.info("-" * 80)
+    logger.info(
+        "Total: %d/%d tests passed (%.1f%%)",
+        passed_count,
+        total_count,
+        100 * passed_count / total_count,
+    )
+
+    if passed_count == total_count:
+        logger.info("🎉 All tests passed!")
+    else:
+        logger.warning("⚠️  Some tests failed. Please review the logs.")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())
--- a/plugins/debug/copilot_sdk_research/test_shell_permission_pattern.py
+++ b/plugins/debug/copilot_sdk_research/test_shell_permission_pattern.py
@@ -0,0 +1,90 @@
+import argparse
+import asyncio
+import logging
+import re
+from typing import Any, Dict
+
+from copilot import CopilotClient
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s - %(levelname)s - %(message)s",
+)
+logger = logging.getLogger(__name__)
+
+
+def build_permission_handler(allow_all: bool, allow_shell: bool, pattern: str):
+    async def on_permission_request(request: Dict[str, Any], context: Dict[str, str]):
+        kind = request.get("kind")
+        # Shell requests use 'fullCommandText' not 'command'
+        command = request.get("fullCommandText", "") or request.get("command", "")
+        logger.info("permission.request FULL: %s", request)
+        logger.info("permission.request kind=%s command=%r", kind, command)
+
+        if allow_all:
+            return {"kind": "approved"}
+
+        if kind in ("read", "url"):
+            return {"kind": "approved"}
+
+        if kind == "shell":
+            if allow_shell:
+                return {"kind": "approved"}
+
+            if pattern and command:
+                try:
+                    if re.match(pattern, command):
+                        return {"kind": "approved"}
+                except re.error as exc:
+                    logger.error("Invalid regex pattern: %s (%s)", pattern, exc)
+
+        return {"kind": "denied-by-rules", "rules": [{"kind": "debug-shell-pattern"}]}
+
+    return on_permission_request
+
+
+async def main():
+    parser = argparse.ArgumentParser(
+        description="Test shell permission regex with GitHub Copilot SDK."
+    )
+    parser.add_argument(
+        "--pattern", default="", help="Regex pattern for auto-approving shell commands."
+    )
+    parser.add_argument(
+        "--allow-shell", action="store_true", help="Auto-approve all shell commands."
+    )
+    parser.add_argument(
+        "--allow-all", action="store_true", help="Auto-approve all permission requests."
+    )
+    parser.add_argument(
+        "--prompt",
+        default="请执行: ls -la",
+        help="Prompt to trigger a shell tool request.",
+    )
+    parser.add_argument("--model", default="gpt-5-mini", help="Model ID for testing.")
+
+    args = parser.parse_args()
+
+    client = CopilotClient()
+    await client.start()
+
+    session = await client.create_session(
+        {
+            "model": args.model,
+            "on_permission_request": build_permission_handler(
+                allow_all=args.allow_all,
+                allow_shell=args.allow_shell,
+                pattern=args.pattern,
+            ),
+        }
+    )
+
+    logger.info("Sending prompt: %s", args.prompt)
+    response = await session.send_and_wait({"prompt": args.prompt})
+    logger.info("Response: %s", response.data.content)
+
+    await client.stop()
+
+
+if __name__ == "__main__":
+    asyncio.run(main())
--- a/plugins/debug/github-copilot-sdk/guides/COPILOT_TOOLS_QUICKSTART.md
+++ b/plugins/debug/github-copilot-sdk/guides/COPILOT_TOOLS_QUICKSTART.md
--- a/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_GUIDE.md
+++ b/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_GUIDE.md
--- a/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_GUIDE_CN.md
+++ b/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_GUIDE_CN.md
--- a/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_USAGE.md
+++ b/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_USAGE.md
--- a/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_USAGE_CN.md
+++ b/plugins/debug/github-copilot-sdk/guides/NATIVE_TOOL_DISPLAY_USAGE_CN.md
--- a/plugins/debug/github-copilot-sdk/guides/OPENWEBUI_FUNCTION_INTEGRATION.md
+++ b/plugins/debug/github-copilot-sdk/guides/OPENWEBUI_FUNCTION_INTEGRATION.md
--- a/plugins/debug/github-copilot-sdk/guides/SESSIONCONFIG_INTEGRATION_GUIDE.md
+++ b/plugins/debug/github-copilot-sdk/guides/SESSIONCONFIG_INTEGRATION_GUIDE.md
--- a/plugins/debug/github-copilot-sdk/guides/TOOLS_USAGE.md
+++ b/plugins/debug/github-copilot-sdk/guides/TOOLS_USAGE.md
--- a/plugins/debug/github-copilot-sdk/guides/TOOL_IMPLEMENTATION_GUIDE.md
+++ b/plugins/debug/github-copilot-sdk/guides/TOOL_IMPLEMENTATION_GUIDE.md
--- a/plugins/debug/github-copilot-sdk/guides/WORKFLOW.md
+++ b/plugins/debug/github-copilot-sdk/guides/WORKFLOW.md
--- a/plugins/debug/github-copilot-sdk/guides/WORKFLOW_CN.md
+++ b/plugins/debug/github-copilot-sdk/guides/WORKFLOW_CN.md
--- a/plugins/debug/github-copilot-sdk/test_capabilities.py
+++ b/plugins/debug/github-copilot-sdk/test_capabilities.py
--- a/plugins/debug/github-copilot-sdk/test_injection.py
+++ b/plugins/debug/github-copilot-sdk/test_injection.py
--- a/plugins/debug/mcp_test.py
+++ b/plugins/debug/mcp_test.py
@@ -0,0 +1,47 @@
+import asyncio
+import os
+from copilot import CopilotClient
+
+
+async def main():
+    token = os.getenv("GH_TOKEN") or os.getenv("GITHUB_TOKEN")
+    if not token:
+        print(
+            "Error: GH_TOKEN (or GITHUB_TOKEN) environment variable not set. Please export GH_TOKEN=... before running."
+        )
+        return
+
+    client = CopilotClient()
+    await client.start()
+
+    async def on_permission_request(request, _ctx):
+        if request.get("kind") == "mcp":
+            return {"kind": "approved"}
+        return {"kind": "approved"}
+
+    session = await client.create_session(
+        {
+            "model": "gpt-5-mini",
+            "mcp_servers": {
+                "github": {
+                    "type": "http",
+                    "url": "https://api.githubcopilot.com/mcp/",
+                    "headers": {"Authorization": f"Bearer {token}"},
+                    "tools": ["*"],
+                }
+            }
+        }
+    )
+
+    result = await session.send_and_wait(
+        {
+            "prompt": "Use GitHub MCP tools to find the owner of the 'awesome-openwebui' repository.",
+        },timeout=1000
+    )
+    print(result.data.content)
+
+    await client.stop()
+
+
+if __name__ == "__main__":
+    asyncio.run(main())
--- a/plugins/filters/github_copilot_sdk_files_filter/README.md
+++ b/plugins/filters/github_copilot_sdk_files_filter/README.md
@@ -0,0 +1,53 @@
+# GitHub Copilot SDK Files Filter
+
+**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.1.2 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **License:** MIT
+
+This is a dedicated **companion filter plugin** designed specifically for the [GitHub Copilot SDK Pipe](https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4).
+
+Its core mission is to **protect user-uploaded files from being "pre-processed" by the OpenWebUI core system, ensuring that the Copilot Agent receives the raw files for autonomous analysis.**
+
+## 🎯 Why is this needed?
+
+In OpenWebUI's default workflow, when you upload a file (e.g., PDF, Excel, Python script), OpenWebUI automatically initiates a **RAG (Retrieval-Augmented Generation)** process: parsing the file, vectorizing it, extracting text, and injecting it into the prompt.
+
+While useful for standard models, this is often disruptive for a **Copilot SDK Agent**:
+
+1. **Agent Needs Raw Files**: The Agent may need to run Python code to read an Excel file or analyze a full directory structure, not chopped-up text fragments.
+2. **Context Pollution**: Large amounts of text injected by RAG consume tokens and can confuse the Agent about "where the file is."
+3. **Control & Performance**: Bypassing the extraction step speeds up the response and gives the Agent full autonomy over how to handle the data.
+
+**This plugin acts as a "bodyguard" to solve these issues.**
+
+## 🚀 How it Works
+
+When you select a Copilot model (name containing `copilot_sdk`) in OpenWebUI and send a file:
+
+1. **Intercept**: This plugin runs with high priority (Priority 0), before RAG and other filters.
+2. **Relocate**: Detecting a Copilot model, it moves the `files` list from the request to a secure custom field `copilot_files`.
+3. **Hide**: It clears the original `files` field.
+4. **Pass**: The OpenWebUI core sees an empty `files` list and **does not trigger RAG**.
+5. **Deliver**: The subsequent Copilot SDK Pipe plugin checks `copilot_files`, retrieves file information, and automatically copies them into the Agent's isolated workspace.
+
+## 📦 Installation & Configuration
+
+### 1. Installation
+
+Import this plugin on the OpenWebUI **Functions** page.
+
+### 2. Enable
+
+Ensure this Filter is enabled globally or in chat settings.
+
+### 3. Configuration (Valves)
+
+Default settings work for most users:
+
+| Parameter | Description | Default |
+| :--- | :--- | :--- |
+| **priority** | Execution priority. **Must be lower than OpenWebUI RAG priority**. | `0` |
+| **target_model_keyword** | Keyword to identify Copilot models for interception. | `copilot_sdk` |
+
+## ⚠️ Important Notes
+
+* **Must be used with Copilot SDK Pipe**: If you install this plugin without the main Pipe plugin, uploaded files will simply "disappear" (as no subsequent plugin will look for them).
+* **Gemini Filter Compatibility**: Fully compatible with the Gemini Multimodal Filter. Just ensure priorities don't conflict.
--- a/plugins/filters/github_copilot_sdk_files_filter/README_CN.md
+++ b/plugins/filters/github_copilot_sdk_files_filter/README_CN.md
@@ -0,0 +1,51 @@
+# GitHub Copilot SDK 文件过滤器
+
+**作者:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **版本:** 0.1.2 | **项目:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **许可证:** MIT
+
+这是一个专门为 [GitHub Copilot SDK Pipe](https://openwebui.com/posts/github_copilot_official_sdk_pipe_ce96f7b4) 设计的**伴侣过滤器插件**。
+
+它的核心使命是：**保护用户上传的文件不被 OpenWebUI 核心系统“抢先处理”，确保 Copilot Agent 能够接收到原始文件并进行自主分析。**
+
+## 🎯 为什么需要它？
+
+在 OpenWebUI 的默认流程中，当你上传一个文件（如 PDF、Excel、Python 脚本）时，OpenWebUI 会自动启动 **RAG（检索增强生成）** 流程：解析文件、向量化、提取文本并注入到提示词中。
+
+虽然这对普通模型很有用，但对于 **Copilot SDK Agent** 来说，这往往是干扰：
+
+1. **Agent 需要原始文件**：Agent 可能需要运行 Python 代码读取 Excel，或者分析完整的代码结构，而不是被切碎的文本片段。
+2. **上下文污染**：RAG 注入的大量文本会消耗 Token，且容易让 Agent 混淆“文件在哪里”。
+3. **控制权与性能**：绕过提取步骤可以加快响应速度，并赋予 Agent 处理数据的完全自主权。
+
+**本插件就是为了解决这个问题而生的“保镖”。**
+
+## 🚀 功能原理
+
+当你在 OpenWebUI 中选择了一个 Copilot 模型（名称包含 `copilot_sdk`）并发送文件时：
+
+1. **拦截 (Intercept)**：本插件会以极高的优先级（Priority 5）运行，先于 RAG 和其他过滤器。
+2. **搬运 (Relocate)**：它检测到模型是 Copilot，便将请求中的 `files`（文件列表）移动到一个安全的自定义字段 `copilot_files` 中。
+3. **隐身 (Hide)**：它清空原始的 `files` 字段。
+4. **放行 (Pass)**：OpenWebUI 核心看到 `files` 为空，便**不会触发 RAG**。
+5. **交付 (Deliver)**：后续的 Copilot SDK Pipe 插件会检查 `copilot_files`，从中获取文件信息，并自动将其复制到 Agent 的独立工作区中。
+
+## 📦 安装与配置
+
+### 1. 安装
+
+在 OpenWebUI 的 **Functions** 页面导入此插件。
+
+### 2. 启用
+
+确保在全局或对话设置中启用了此 Filter。
+
+### 3. 配置 (Valves)
+
+| 参数 | 说明 | 默认值 |
+| :--- | :--- | :--- |
+| **priority** | 过滤器的执行优先级。**必须小于 OpenWebUI RAG 的优先级**。 | `5` |
+| **target_model_keyword** | 用户识别 Copilot 模型的关键词。只有包含此关键词的模型才会触发。 | `copilot_sdk` |
+
+## ⚠️ 注意事项
+
+* **必须配合 Copilot SDK Pipe 使用**：如果你没有安装主 Pipe 插件，本插件将导致上传的文件“凭空消失”。
+* **Gemini Filter 兼容性**：已完美兼容 Gemini 多模态过滤器。只要优先级设置正确，它们可以共存互不干扰。
--- a/plugins/filters/github_copilot_sdk_files_filter/github_copilot_sdk_files_filter.py
+++ b/plugins/filters/github_copilot_sdk_files_filter/github_copilot_sdk_files_filter.py
@@ -0,0 +1,67 @@
+"""
+title: GitHub Copilot SDK Files Filter
+id: github_copilot_sdk_files_filter
+author: Fu-Jie
+author_url: https://github.com/Fu-Jie/awesome-openwebui
+funding_url: https://github.com/open-webui
+version: 0.1.2
+openwebui_id: 403a62ee-a596-45e7-be65-fab9cc249dd6
+description: A specialized filter to bypass OpenWebUI's default RAG for GitHub Copilot SDK models. It moves uploaded files to a safe location ('copilot_files') so the Copilot Pipe can process them natively without interference.
+"""
+
+from pydantic import BaseModel, Field
+from typing import Optional, Callable, Awaitable
+
+
+class Filter:
+    class Valves(BaseModel):
+        priority: int = Field(
+            default=0,
+            description="Priority level. Must be lower than RAG processors to intercept files effectively.",
+        )
+        target_model_keyword: str = Field(
+            default="copilot_sdk",
+            description="Keyword to identify Copilot models (e.g., 'copilot_sdk').",
+        )
+
+    def __init__(self):
+        self.valves = self.Valves()
+
+    async def inlet(
+        self,
+        body: dict,
+        __user__: Optional[dict] = None,
+        __model__: Optional[dict] = None,
+        __metadata__: Optional[dict] = None,
+        __event_emitter__: Callable[[dict], Awaitable[None]] = None,
+    ) -> dict:
+        # Determine the actual model ID
+        base_model_id = None
+        if __model__:
+            if "openai" in __model__:
+                base_model_id = __model__["openai"].get("id")
+            else:
+                base_model_id = __model__.get("info", {}).get("base_model_id")
+
+        current_model = base_model_id if base_model_id else body.get("model", "")
+
+        # Check if it's a Copilot model
+        if self.valves.target_model_keyword.lower() in current_model.lower():
+            # If files exist, move them to 'copilot_files' and clear 'files'
+            # This prevents OpenWebUI from triggering RAG on these files
+            if "files" in body and body["files"]:
+                file_count = len(body["files"])
+                if __event_emitter__:
+                    await __event_emitter__(
+                        {
+                            "type": "status",
+                            "data": {
+                                "description": f"Managed {file_count} files for Copilot (RAG Bypassed)",
+                                "done": True,
+                            },
+                        }
+                    )
+                body["copilot_files"] = body["files"]
+                body["files"] = []
+
+        return body
--- a/plugins/filters/github_copilot_sdk_files_filter/github_copilot_sdk_files_filter_cn.py
+++ b/plugins/filters/github_copilot_sdk_files_filter/github_copilot_sdk_files_filter_cn.py
@@ -0,0 +1,66 @@
+"""
+title: GitHub Copilot SDK 文件过滤器(GitHub Copilot SDK Files Filter)
+id: github_copilot_sdk_files_filter
+author: Fu-Jie
+author_url: https://github.com/Fu-Jie/awesome-openwebui
+funding_url: https://github.com/open-webui
+version: 0.1.2
+description: 一个专门的过滤器，用于绕过 OpenWebUI 默认的 RAG 机制，针对 GitHub Copilot SDK 模型。它将上传的文件移动到安全位置 ('copilot_files')，以便 Copilot Pipe 可以原生处理它们而不受干扰。
+"""
+
+from pydantic import BaseModel, Field
+from typing import Optional, Callable, Awaitable
+
+
+class Filter:
+    class Valves(BaseModel):
+        priority: int = Field(
+            default=0,
+            description="优先级。必须低于 RAG 处理器的优先级，以便有效拦截文件。",
+        )
+        target_model_keyword: str = Field(
+            default="copilot_sdk",
+            description="用于识别 Copilot 模型的关键词（例如 'copilot_sdk'）。",
+        )
+
+    def __init__(self):
+        self.valves = self.Valves()
+
+    async def inlet(
+        self,
+        body: dict,
+        __user__: Optional[dict] = None,
+        __model__: Optional[dict] = None,
+        __metadata__: Optional[dict] = None,
+        __event_emitter__: Callable[[dict], Awaitable[None]] = None,
+    ) -> dict:
+        # Determine the actual model ID
+        base_model_id = None
+        if __model__:
+            if "openai" in __model__:
+                base_model_id = __model__["openai"].get("id")
+            else:
+                base_model_id = __model__.get("info", {}).get("base_model_id")
+
+        current_model = base_model_id if base_model_id else body.get("model", "")
+
+        # Check if it's a Copilot model
+        if self.valves.target_model_keyword.lower() in current_model.lower():
+            # If files exist, move them to 'copilot_files' and clear 'files'
+            # This prevents OpenWebUI from triggering RAG on these files
+            if "files" in body and body["files"]:
+                file_count = len(body["files"])
+                if __event_emitter__:
+                    await __event_emitter__(
+                        {
+                            "type": "status",
+                            "data": {
+                                "description": f"已为 Copilot 管理 {file_count} 个文件 (已绕过 RAG)",
+                                "done": True,
+                            },
+                        }
+                    )
+                body["copilot_files"] = body["files"]
+                body["files"] = []
+
+        return body
--- a/plugins/filters/web_gemini_multimodel_filter/web_gemini_multimodel.py
+++ b/plugins/filters/web_gemini_multimodel_filter/web_gemini_multimodel.py
@@ -3,6 +3,7 @@ title: Gemini 多模态过滤器（含字幕增强）
 author: Gemini Adapter
 author_url: https://github.com/Fu-Jie
 funding_url: https://github.com/Fu-Jie/awesome-openwebui
+
 version: 0.3.2
 description: >
    一个强大的过滤器，为 OpenWebUI 中的任何模型提供多模态能力：PDF、Office、图片、音频、视频等。
@@ -492,6 +493,13 @@ class Filter:
        print(
            f"🤖 Checking model: {current_model} (Target: {self.valves.target_model_keyword})"
        )
+        # 0. SKIP COPILOT MODELS
+        # If the model is a Copilot Pipe model, we must NOT process files here.
+        # Copilot Pipe has its own file handling logic (copy to workspace).
+        if "copilot_sdk" in current_model.lower():
+            print(f"⏩ Skipping Gemini Filter for Copilot model: {current_model}")
+            return body
+

        # Check if model matches target keyword
        is_target_model = (
--- a/plugins/pipes/github-copilot-sdk/INTERNAL_KNOWLEDGE.md
+++ b/plugins/pipes/github-copilot-sdk/INTERNAL_KNOWLEDGE.md
@@ -0,0 +1,73 @@
+# GitHub Copilot SDK Pipe - Internal Architecture & Knowledge
+
+This document details the core runtime mechanisms, file layout, and deep integration logic of the GitHub Copilot SDK plugin within the OpenWebUI environment.
+
+## 1. Core Environmental Context
+
+### 1.1 Filesystem Layout
+
+| Path | Description | Permissions |
+| :--- | :--- | :--- |
+| `/app/backend` | OpenWebUI backend Python source code | Read-Only |
+| `/app/build` | OpenWebUI frontend assets (Artifacts renderer location) | Read-Only |
+| `/root/.copilot/` | SDK core configuration and state storage | **Full Control** |
+| `/app/backend/data/copilot_workspace/` | Plugin-specific persistent workspace | **Read/Write** |
+
+### 1.2 Identity Mapping Mechanism
+
+* **Session ID Binding**: The plugin strictly maps OpenWebUI's `Chat ID` to the Copilot SDK's `Session ID`.
+* **Outcome**: Every chat session has its own isolated physical directory: `/root/.copilot/session-state/{chat_id}/`.
+
+## 2. TODO List Persistence (TODO Intelligence)
+
+### 2.1 Data Source
+
+TODO data is **not** stored in a standalone database table for basic operations but is captured from the session event stream via the `update_todo` tool:
+
+* **Storage File**: `/root/.copilot/session-state/{chat_id}/events.jsonl`
+* **Detection**: Scans for the latest `tool.execution_complete` event where `toolName` is `update_todo`.
+
+### 2.2 Data Format (NDJSON)
+
+```json
+{
+  "type": "tool.execution_complete",
+  "data": {
+    "toolName": "update_todo",
+    "result": { 
+      "detailedContent": "TODO List content...", 
+      "toolTelemetry": { "metrics": { "total_items": 3 } } 
+    }
+  }
+}
+```
+
+## 3. Toolchain Integration
+
+The plugin harmonizes three distinct tool systems:
+
+1. **Copilot Native**: Built-in capabilities like `bash`, `edit`, and `task`.
+2. **OpenWebUI Ecosystem**: Dynamically mounts local Python tools and built-in `Web Search` via `get_tools`.
+3. **MCP (Model Context Protocol)**: External capability extensions via servers like `GDMap` or `Pandoc`.
+
+## 4. Security & Permissions
+
+### 4.1 Admin Mode (God Mode)
+
+When `__user__['role'] == 'admin'`:
+
+* `ADMIN_EXTENSIONS` are enabled.
+* Access to `DATABASE_URL` via environment variables is permitted.
+* `bash` can be used to diagnose internal state within `/root/.copilot/`.
+
+### 4.2 Regular User Mode
+
+* `USER_RESTRICTIONS` are strictly enforced.
+* Probing environment variables or database credentials is prohibited.
+* `bash` activity is strictly confined to the isolated workspace.
+
+## 5. Common Maintenance
+
+* **Reset Session**: Delete the `/root/.copilot/session-state/{chat_id}` directory.
+* **Clear Cache**: Disable `ENABLE_TOOL_CACHE` in Valves.
+* **View Logs**: Check latest logs under `/root/.copilot/logs/`.
--- a/plugins/pipes/github-copilot-sdk/INTERNAL_KNOWLEDGE_CN.md
+++ b/plugins/pipes/github-copilot-sdk/INTERNAL_KNOWLEDGE_CN.md
@@ -0,0 +1,70 @@
+# GitHub Copilot SDK Pipe - 内部架构与通用知识 (Internal Knowledge)
+
+本文档记录了 GitHub Copilot SDK 插件在 OpenWebUI 环境下的核心运行机制、文件布局及深度集成逻辑。
+
+## 1. 核心环境上下文 (Environment)
+
+### 1.1 文件系统布局
+
+| 路径 | 说明 | 权限 |
+| :--- | :--- | :--- |
+| `/app/backend` | OpenWebUI 后端 Python 源码 | 只读 |
+| `/app/build` | OpenWebUI 前端静态资源 (Artifacts 渲染器位置) | 只读 |
+| `/root/.copilot/` | SDK 核心配置与状态存储 | **完全控制** |
+| `/app/backend/data/copilot_workspace/` | 插件指定的持久化工作区 | **读写自如** |
+
+### 1.2 身份映射机制
+
+* **Session ID 绑定**: 插件强制将 OpenWebUI 的 `Chat ID` 映射为 Copilot SDK 的 `Session ID`。
+* **结果**: 每个对话窗口都有独立的物理存储目录：`/root/.copilot/session-state/{chat_id}/`。
+
+## 2. TODO List 存储机制 (TODO Intelligence)
+
+### 2.1 数据源
+
+TODO 数据**并不**仅存储在独立数据库，而是通过 `update_todo` 工具写入会话事件流：
+
+* **文件**: `/root/.copilot/session-state/{chat_id}/events.jsonl`
+* **识别**: 查找类型为 `tool.execution_complete` 且 `toolName` 为 `update_todo` 的最新 JSON 行。
+
+### 2.2 数据格式 (NDJSON)
+
+```json
+{
+  "type": "tool.execution_complete",
+  "data": {
+    "toolName": "update_todo",
+    "result": { "detailedContent": "TODO List内容...", "toolTelemetry": { "metrics": { "total_items": 39 } } }
+  }
+}
+```
+
+## 3. 工具体系 (Toolchain)
+
+插件整合了三套工具系统：
+
+1. **Copilot Native**: SDK 内置的 `bash`, `edit`, `task` 等。
+2. **OpenWebUI Ecosystem**: 通过 `get_tools` 挂载的本地 Python 脚本和内置 `Web Search`。
+3. **MCP (Model Context Protocol)**: 通过 `GDMap`, `Pandoc` 等服务器扩展的外部能力。
+
+## 4. 安全与权限 (Security)
+
+### 4.1 管理员模式 (God Mode)
+
+当 `__user__['role'] == 'admin'` 时：
+
+* 启用 `ADMIN_EXTENSIONS`。
+* 允许通过环境变量获取 `DATABASE_URL`。
+* 允许使用 `bash` 诊断 `/root/.copilot/` 内部状态。
+
+### 4.2 普通用户模式
+
+* 启用 `USER_RESTRICTIONS`。
+* 严禁探测环境变量和数据库。
+* 限制 `bash` 只能在工作区内活动。
+
+## 5. 常见维护操作
+
+* **重置会话**: 删除 `/root/.copilot/session-state/{chat_id}` 目录。
+* **清理缓存**: 在 Valves 中关闭 `ENABLE_TOOL_CACHE`。
+* **查看日志**: 检查 `/root/.copilot/logs/` 下的最新日志。
--- a/plugins/pipes/github-copilot-sdk/README.md
+++ b/plugins/pipes/github-copilot-sdk/README.md
@@ -1,103 +1,118 @@
 # GitHub Copilot SDK Pipe for OpenWebUI

-**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.3.0 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **License:** MIT
+**Author:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **Version:** 0.6.0 | **Project:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **License:** MIT

-This is an advanced Pipe function for [OpenWebUI](https://github.com/open-webui/open-webui) that allows you to use GitHub Copilot models (such as `gpt-5`, `gpt-5-mini`, `claude-sonnet-4.5`) directly within OpenWebUI. It is built upon the official [GitHub Copilot SDK for Python](https://github.com/github/copilot-sdk), providing a native integration experience.
+This is an advanced Pipe function for [OpenWebUI](https://github.com/open-webui/open-webui) that integrates the official [GitHub Copilot SDK](https://github.com/github/copilot-sdk). It enables you to use **GitHub Copilot models** (e.g., `gpt-5.2-codex`, `claude-sonnet-4.5`,`gemini-3-pro`, `gpt-5-mini`) **AND** your own models via **BYOK** (OpenAI, Anthropic) directly within OpenWebUI, providing a unified agentic experience with **strict User & Chat-level Workspace Isolation**.

-## 🚀 What's New (v0.3.0) - The Power of "Unified Ecosystem"
+> [!IMPORTANT]
+> **Essential Companion**
+> To unlock file handling and data analysis capabilities, you must install the [GitHub Copilot SDK Files Filter](https://openwebui.com/posts/403a62ee-a596-45e7-be65-fab9cc249dd6).

-* **🔌 Zero-Config Tool Bridge**: Automatically transforms your existing OpenWebUI Functions (Tools) into Copilot-compatible tools. **Copilot now has total access to your entire WebUI toolset!**
-* **🔗 Dynamic MCP Discovery**: Seamlessly connects to MCP servers defined in **Admin Settings -> Connections**. No configuration files required—it just works.
-* **⚡ High-Performance Async Engine**: Background CLI updates and optimized event-driven streaming ensure lightning-fast responses without UI lag.
-* **🛡️ Robust Interoperability**: Advanced sanitization and dynamic Pydantic model generation ensure smooth integration even with complex third-party tools.
+> [!TIP]
+> **No Subscription Required for BYOK**
+> If you are using your own API keys (BYOK mode with OpenAI/Anthropic), **you do NOT need a GitHub Copilot subscription.** A subscription is only required to access GitHub's official models.
+
+---
+
+## ✨ v0.6.0 Updates (What's New)
+
+- **👥 User & Chat Management**: Physical management architecture (`user_id/chat_id`) for absolute resource independence.
+- **🤖 Empowering Agent Autonomy**: Automatic synchronization of raw files to the workspace, enabling direct Python-based analysis of Excel/CSV.
+- **🔧 OpenAPI & External Tool Fixes**: Full support for tools mounted via OpenAPI servers.
+- **📊 Cost Control**: Enhanced **Billing Multiplier Limits** (`MAX_MULTIPLIER`, e.g., set to 0 for free models only) and **Model Keyword Filtering** (`EXCLUDE_KEYWORDS`) for precise cost management.
+- **🧠 Persistent TODO Lists**: Database-backed task tracking that persists across sessions.
+
+---

 ## ✨ Key Capabilities

-* **🌉 The Ultimate Bridge**: The first and only plugin that creates a seamless bridge between **OpenWebUI Tools** and **GitHub Copilot SDK**.
-* **🚀 Official & Native**: Built directly on the official Python SDK, providing the most stable and authentic Copilot experience.
-* **🌊 Advanced Streaming (Thought Process)**: Supports full model reasoning/thinking display with typewriter effects.
-* **🖼️ Intelligent Multimodal**: Full support for images and attachments, enabling Copilot to "see" your workspace.
-* **🛠️ Effortless Setup**: Automatic CLI detection, version enforcement, and dependency management.
-* **🔑 Dual-Layer Security**: Supports secure OAuth flow for Chat and standard PAT for extended MCP capabilities.
+- **🔑 Flexible Auth & BYOK**: Official Copilot subscriptions (PAT) or Bring Your Own Key (OpenAI/Anthropic).
+- **🔌 Universal Tool Protocol**: Native support for **MCP (Model Context Protocol)**, OpenAPI, and OpenWebUI built-in tools.
+- **🛡️ Sandbox Workspace Isolation**: Strict per-session sandboxing for data privacy and security.
+- **♾️ Infinite Session Management**: Smart context window management with automatic compaction for indefinite conversation capability.
+- **🧠 Deep Database Integration**: Real-time persistence of TOD·O lists for long-running workflows.
+- **🌊 Advanced Streaming**: Full support for thinking process/Chain of Thought visualization.
+- **🖼️ Intelligent Multimodal**: Vision capabilities and raw file analysis support.
+- **⚡ Interactive Artifacts**: Automatically renders HTML/JS apps generated by the agent directly in the chat interface.

-## Installation & Configuration
+---

-### 1) Import Function
+## ⚙️ Core Configuration (Valves)

-1. Open OpenWebUI.
-2. Go to **Workspace** -> **Functions**.
-3. Click **+** (Create Function).
-4. Paste the content of `github_copilot_sdk.py` (or `github_copilot_sdk_cn.py` for Chinese) completely.
-5. Save.
+### 1. Administrator Settings (Base)

-### 2) Configure Valves (Settings)
+Administrators define the default behavior for all users in the function settings.

-Find "GitHub Copilot" in the function list and click the **⚙️ (Valves)** icon to configure:
-
-| Parameter | Description | Default |
+| Valve | Default | Description |
 | :--- | :--- | :--- |
-| **GH_TOKEN** | **(Required)** GitHub Access Token (PAT or OAuth Token). Access to Chat. | - |
-| **DEBUG** | Whether to enable debug logs (output to browser console). | `False` |
-| **LOG_LEVEL** | Copilot CLI log level: none, error, warning, info, debug, all. | `error` |
-| **SHOW_THINKING** | Show model reasoning/thinking process (requires streaming + model support). | `True` |
-| **COPILOT_CLI_VERSION** | Specific Copilot CLI version to install/enforce. | `0.0.405` |
-| **EXCLUDE_KEYWORDS** | Exclude models containing these keywords (comma separated). | - |
-| **WORKSPACE_DIR** | Restricted workspace directory for file operations. | - |
-| **INFINITE_SESSION** | Enable Infinite Sessions (automatic context compaction). | `True` |
-| **COMPACTION_THRESHOLD** | Background compaction threshold (0.0-1.0). | `0.8` |
-| **BUFFER_THRESHOLD** | Buffer exhaustion threshold (0.0-1.0). | `0.95` |
-| **TIMEOUT** | Timeout for each stream chunk (seconds). | `300` |
-| **CUSTOM_ENV_VARS** | Custom environment variables (JSON format). | - |
-| **REASONING_EFFORT** | Reasoning effort level: low, medium, high. `xhigh` is supported for some models. | `medium` |
-| **ENFORCE_FORMATTING** | Add formatting instructions to system prompt for better readability. | `True` |
-| **ENABLE_MCP_SERVER** | Enable Direct MCP Client connection (Recommended). | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | Enable OpenWebUI Tools (includes defined and server tools). | `True` |
+| `GH_TOKEN` | `""` | Global GitHub Token (Requires 'Copilot Requests' permission). |
+| `ENABLE_OPENWEBUI_TOOLS` | `True` | Enable OpenWebUI Tools (includes defined Tools and Built-in Tools). |
+| `ENABLE_OPENAPI_SERVER` | `True` | Enable OpenAPI Tool Server connection. |
+| `ENABLE_MCP_SERVER` | `True` | Enable Direct MCP Client connection (Recommended). |
+| `REASONING_EFFORT` | `medium` | Reasoning effort level: low, medium, high. |
+| `SHOW_THINKING` | `True` | Show model reasoning/thinking process. |
+| `INFINITE_SESSION` | `True` | Enable Infinite Sessions (automatic context compaction). |
+| `MAX_MULTIPLIER` | `1.0` | Max allowed billing multiplier (0x for free models only). |
+| `EXCLUDE_KEYWORDS` | `""` | Exclude models containing these keywords (comma separated). |
+| `TIMEOUT` | `300` | Timeout for each stream chunk (seconds). |
+| `BYOK_TYPE` | `openai` | BYOK Provider Type: `openai`, `anthropic`. |
+| `BYOK_BASE_URL` | `""` | BYOK Base URL (e.g., <https://api.openai.com/v1>). |
+| `BYOK_MODELS` | `""` | BYOK Model List (comma separated). Leave empty to fetch from API. |
+| `CUSTOM_ENV_VARS` | `""` | Custom environment variables (JSON format). |
+| `DEBUG` | `False` | Enable this to see detailed logs in your browser console. |

-#### User Valves (per-user overrides)
+### 2. User Settings (Individual Overrides)

-These optional settings can be set per user (overrides global Valves):
+Standard users can override these settings in their individual Profile/Function settings.

-| Parameter | Description | Default |
-| :--- | :--- | :--- |
-| **GH_TOKEN** | Personal GitHub Token (overrides global setting). | - |
-| **REASONING_EFFORT** | Reasoning effort level (low/medium/high/xhigh). | - |
-| **DEBUG** | Enable technical debug logs. | `False` |
-| **SHOW_THINKING** | Show model reasoning/thinking process. | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | Enable OpenWebUI Tools (overrides global). | `True` |
-| **ENABLE_MCP_SERVER** | Enable MCP server loading (overrides global). | `True` |
-| **ENFORCE_FORMATTING** | Enforce formatting guidelines (overrides global). | `True` |
+| Valve | Description |
+| :--- | :--- |
+| `GH_TOKEN` | Use your personal GitHub Token. |
+| `REASONING_EFFORT` | Individual reasoning effort preference. |
+| `SHOW_THINKING` | Show model reasoning/thinking process. |
+| `MAX_MULTIPLIER` | Maximum allowed billing multiplier override. |
+| `EXCLUDE_KEYWORDS` | Exclude models containing these keywords. |
+| `BYOK_API_KEY` | Use your personal OpenAI/Anthropic API Key. |
+
+---
+
+## 🎯 Use Cases (What can you do?)
+
+- **📁 Fully Autonomous DevOps**: Agent analyzes code, runs tests, and applies patches within its isolated sandbox.
+- **📊 Deep Data Auditing**: Directly process raw Excel/CSV data via Python (bypassing RAG) and generate visual reports.
+- **📝 Long-Task Management**: Automatically decomposes complex requests and persists TOD·O progress across sessions.
+
+---

 ## ⭐ Support

-If this plugin has been useful, a star on [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) is a big motivation for me. Thank you for the support.
+If this plugin has been useful, a **Star** on [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) would be a great motivation for me. Thank you!

-### Get Token
+---

-To use GitHub Copilot, you need a GitHub Personal Access Token (PAT) with appropriate permissions.
+## 🚀 Installation & Configuration

-**Steps to generate your token:**
+### 1) Import Function
+
+1. Open OpenWebUI, go to **Workspace** -> **Functions**.
+2. Click **+** (Create Function), paste the content of `github_copilot_sdk.py`.
+3. Save and ensure it is enabled.
+
+### 2) Get Token

 1. Visit [GitHub Token Settings](https://github.com/settings/tokens?type=beta).
-2. Click **Generate new token (fine-grained)**.
-3. **Repository access**: Select **Public Repositories** (simplest) or **All repositories**.
-4. **Permissions**:
-    * If you chose **All repositories**, you must click **Account permissions**.
-    * Find **Copilot Requests**, and select **Access**.
-5. Generate and copy the Token.
+2. Create **Fine-grained token**, granting **Account permissions** -> **Copilot Requests** access.
+3. Paste the generated Token into the `GH_TOKEN` field in Valves.

-## 📋 Dependencies
+---

-This Pipe will automatically attempt to install the following dependencies:
+## 📋 Troubleshooting & Dependencies

-* `github-copilot-sdk` (Python package)
-* `github-copilot-cli` (Binary file, installed via official script)
+- **Agent ignores files?**: Ensure the Files Filter is enabled, otherwise RAG will interfere with raw binaries.
+- **No progress bar?**: The bar only appears when the Agent uses the `update_todo` tool.
+- **Dependencies**: This Pipe automatically installs `github-copilot-sdk` (Python) and `github-copilot-cli` (Binary).

-## Troubleshooting ❓
-
-* **Images not recognized**:
-  * Ensure `MODEL_ID` is a model that supports multimodal input.
-* **Thinking not shown**:
-  * Ensure **streaming is enabled** and the selected model supports reasoning output.
+---

 ## Changelog

--- a/plugins/pipes/github-copilot-sdk/README_CN.md
+++ b/plugins/pipes/github-copilot-sdk/README_CN.md
@@ -1,104 +1,123 @@
 # GitHub Copilot SDK 官方管道

-**作者:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **版本:** 0.3.0 | **项目:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **许可证:** MIT
+**作者:** [Fu-Jie](https://github.com/Fu-Jie/awesome-openwebui) | **版本:** 0.6.0 | **项目:** [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) | **许可证:** MIT

-这是一个用于 [OpenWebUI](https://github.com/open-webui/open-webui) 的高级 Pipe 函数，允许你直接在 OpenWebUI 中使用 GitHub Copilot 模型（如 `gpt-5`, `gpt-5-mini`, `claude-sonnet-4.5`）。它基于官方 [GitHub Copilot SDK for Python](https://github.com/github/copilot-sdk) 构建，提供了原生级的集成体验。
+这是一个用于 [OpenWebUI](https://github.com/open-webui/open-webui) 的高级 Pipe 函数，深度集成了 **GitHub Copilot SDK**。它不仅支持 **GitHub Copilot 官方模型**（如 `gpt-5.2-codex`, `claude-sonnet-4.5`, `gemini-3-pro`, `gpt-5-mini`），还支持 **BYOK (自带 Key)** 模式对接自定义服务商（OpenAI, Anthropic），并具备**严格的用户与会话级工作区隔离**能力，提供统一且安全的 Agent 交互体验。

-## 🚀 最新特性 (v0.3.0) - “统一生态”的力量
+> [!IMPORTANT]
+> **核心伴侣组件**
+> 如需启用文件处理与数据分析能力，请务必安装 [GitHub Copilot SDK Files Filter](https://openwebui.com/posts/403a62ee-a596-45e7-be65-fab9cc249dd6)。

-* **🔌 零配置工具桥接 (Unified Tool Bridge)**: 自动将您现有的 OpenWebUI Functions (工具) 转换为 Copilot 兼容工具。**Copilot 现在可以无缝调用您手头所有的 WebUI 工具！**
-* **🔗 动态 MCP 自动发现**: 直接联动 OpenWebUI **管理面板 -> 连接**。无需编写任何配置文件，即插即用，瞬间扩展 Copilot 能力边界。
-* **⚡ 高性能异步引擎**: 异步 CLI 更新检查与高度优化的事件驱动流式处理，确保对话毫秒级响应。
-* **🛡️ 卓越的兼容性**: 独创的动态 Pydantic 模型生成技术，确保复杂工具参数在 Copilot 端也能得到精准验证。
+> [!TIP]
+> **BYOK 模式无需订阅**
+> 如果您使用自带的 API Key (BYOK 模式对接 OpenAI/Anthropic)，**您不需要 GitHub Copilot 官方订阅**。只有在访问 GitHub 官方模型时才需要订阅。

-## ✨ 核心能力
+---

-* **🌉 强大的生态桥接**: 首个且唯一完美打通 **OpenWebUI Tools** 与 **GitHub Copilot SDK** 的插件。
-* **🚀 官方原生产体验**: 基于官方 Python SDK 构建，提供最稳定、最纯正的 Copilot 交互体验。
-* **🌊 深度推理展示**: 完整支持模型思考过程 (Thinking Process) 的流式渲染。
-* **🖼️ 智能多模态**: 支持图像识别与附件上传，让 Copilot 拥有视觉能力。
-* **🛠️ 极简部署流程**: 自动检测环境、自动下载 CLI、自动管理依赖，全自动化开箱即用。
-* **🔑 安全认证体系**: 完美支持 OAuth 授权与 PAT 模式，兼顾便捷与安全性。
+## ✨ 0.6.0 更新内容 (What's New)

-## 安装与配置
+- **👥 多用户与会话管理**: 采用 `user_id/chat_id` 的物理隔离架构，确保资源独立与稳健管理。
+- **🤖 赋予 Agent 文件自主权**: 自动将上传的文件同步至物理工作区，支持 Python 直接分析 Excel/CSV。
+- **🔧 OpenAPI & 外部工具修复**: 完美支持通过 OpenAPI 服务器挂载的工具调用。
+- **📊 计费与成本控制**: 增强的**计费倍率限制** (`MAX_MULTIPLIER`，例如设为 0 即仅限免费模型) 和**模型关键词过滤** (`EXCLUDE_KEYWORDS`)，实现更精准的成本管控。
+- **🧠 数据库持久化 TODO**: 任务进度跨会话保存，Agent 拥有更持久的任务记忆。

-### 1. 导入函数
+---

-1. 打开 OpenWebUI。
-2. 进入 **Workspace** -> **Functions**。
-3. 点击 **+** (创建函数)。
-4. 将 `github_copilot_sdk_cn.py` 的内容完整粘贴进去。
-5. 保存。
+## ✨ 核心能力 (Key Capabilities)

-### 2. 配置 Valves (设置)
+- **🔑 灵活鉴权与 BYOK**: 支持 GitHub Copilot 官方订阅 (PAT) 或自带 Key (OpenAI/Anthropic)。
+- **🔌 通用工具协议**: 原生支持 **MCP (Model Context Protocol)**、OpenAPI 以及 OpenWebUI 内置工具。
+- **🛡️ 物理级工作区隔离**: 强制执行严格的用户特定沙箱，确保数据隐私与文件安全。
+- **♾️ 无限会话管理**: 智能上下文窗口管理与自动压缩算法，支持无限时长的对话交互。
+- **🧠 深度数据库集成**: 实时持久化 TOD·O 列表到 UI 进度条。
+- **🌊 深度推理展示**: 完整支持模型思考过程 (Thinking Process) 的流式渲染。
+- **🖼️ 智能多模态**: 完整支持图像识别与附件上传分析。
+- **⚡ 交互式伪影 (Artifacts)**: 自动渲染 Agent 生成的 HTML/JS 应用程序，直接在聊天界面交互。

-在函数列表中找到 "GitHub Copilot"，点击 **⚙️ (Valves)** 图标进行配置：
+---

-| 参数 | 说明 | 默认值 |
+## ⚙️ 核心配置参数 (Valves)
+
+### 1. 管理员配置 (基础设置)
+
+管理员可在函数设置中定义全局默认行为。
+
+| 参数 | 默认值 | 说明 |
 | :--- | :--- | :--- |
-| **GH_TOKEN** | **(必填)** GitHub 访问令牌 (PAT 或 OAuth Token)。用于聊天。 | - |
-| **DEBUG** | 是否开启调试日志（输出到浏览器控制台）。 | `False` |
-| **LOG_LEVEL** | Copilot CLI 日志级别: none, error, warning, info, debug, all。 | `error` |
-| **SHOW_THINKING** | 是否显示模型推理/思考过程（需开启流式 + 模型支持）。 | `True` |
-| **COPILOT_CLI_VERSION** | 指定安装/强制使用的 Copilot CLI 版本。 | `0.0.405` |
-| **EXCLUDE_KEYWORDS** | 排除包含这些关键词的模型（逗号分隔）。 | - |
-| **WORKSPACE_DIR** | 文件操作的受限工作区目录。 | - |
-| **INFINITE_SESSION** | 启用无限会话（自动上下文压缩）。 | `True` |
-| **COMPACTION_THRESHOLD** | 后台压缩阈值 (0.0-1.0)。 | `0.8` |
-| **BUFFER_THRESHOLD** | 缓冲区耗尽阈值 (0.0-1.0)。 | `0.95` |
-| **TIMEOUT** | 每个流式分块超时（秒）。 | `300` |
-| **CUSTOM_ENV_VARS** | 自定义环境变量 (JSON 格式)。 | - |
-| **REASONING_EFFORT** | 推理强度级别: low, medium, high. `xhigh` 仅部分模型支持。 | `medium` |
-| **ENFORCE_FORMATTING** | 在系统提示词中添加格式化指导。 | `True` |
-| **ENABLE_MCP_SERVER** | 启用直接 MCP 客户端连接 (建议)。 | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | 启用 OpenWebUI 工具 (包括自定义和服务器工具)。 | `True` |
+| `GH_TOKEN` | `""` | 全局 GitHub Token (需具备 'Copilot Requests' 权限)。 |
+| `ENABLE_OPENWEBUI_TOOLS` | `True` | 启用 OpenWebUI 工具 (包括定义工具和内置工具)。 |
+| `ENABLE_OPENAPI_SERVER` | `True` | 启用 OpenAPI 工具服务器连接。 |
+| `ENABLE_MCP_SERVER` | `True` | 启用直接 MCP 客户端连接 (推荐)。 |
+| `REASONING_EFFORT` | `medium` | 推理强度：low, medium, high。 |
+| `SHOW_THINKING` | `True` | 显示模型推理/思考过程。 |
+| `INFINITE_SESSION` | `True` | 启用无限会话 (自动上下文压缩)。 |
+| `MAX_MULTIPLIER` | `1.0` | 最大允许的模型计费倍率 (0x 为仅限免费模型)。 |
+| `EXCLUDE_KEYWORDS` | `""` | 排除包含这些关键字的模型 (逗号分隔)。 |
+| `TIMEOUT` | `300` | 每个流数据块的超时时间 (秒)。 |
+| `BYOK_TYPE` | `openai` | BYOK 服务商类型：`openai`, `anthropic`。 |
+| `BYOK_BASE_URL` | `""` | BYOK 基础 URL (例如: <https://api.openai.com/v1)。> |
+| `BYOK_MODELS` | `""` | BYOK 模型列表 (逗号分隔)。留空则从 API 获取。 |
+| `CUSTOM_ENV_VARS` | `""` | 自定义环境变量 (JSON 格式)。 |
+| `DEBUG` | `False` | 开启此项以在前端控制台输出详细调试日志。 |

-#### 用户 Valves（按用户覆盖）
+### 2. 用户配置 (个人覆盖)

-以下设置可按用户单独配置（覆盖全局 Valves）：
+普通用户可在各自的个人设置中根据需要覆盖以下参数。

-| 参数 | 说明 | 默认值 |
-| :--- | :--- | :--- |
-| **GH_TOKEN** | 个人 GitHub Token（覆盖全局设置）。 | - |
-| **REASONING_EFFORT** | 推理强度级别（low/medium/high/xhigh）。 | - |
-| **DEBUG** | 是否启用技术调试日志。 | `False` |
-| **SHOW_THINKING** | 是否显示思考过程。 | `True` |
-| **ENABLE_OPENWEBUI_TOOLS** | 启用 OpenWebUI 工具（覆盖全局设置）。 | `True` |
-| **ENABLE_MCP_SERVER** | 启用动态 MCP 服务器加载（覆盖全局设置）。 | `True` |
-| **ENFORCE_FORMATTING** | 强制启用格式化指导（覆盖全局设置）。 | `True` |
+| 参数 | 说明 |
+| :--- | :--- |
+| `GH_TOKEN` | 使用个人的 GitHub Token。 |
+| `REASONING_EFFORT`| 个人偏好的推理强度。 |
+| `SHOW_THINKING` | 显示模型推理/思考过程。 |
+| `MAX_MULTIPLIER` | 最大允许的模型计费倍率覆盖。 |
+| `EXCLUDE_KEYWORDS` | 排除包含这些关键字的模型。 |
+| `BYOK_API_KEY` | 使用个人的 OpenAI/Anthropic API Key。 |

-## ⭐ 支持
+---

-如果这个插件对你有帮助，欢迎到 [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) 点个 Star，这将是我持续改进的动力，感谢支持。
+## 🎯 典型应用场景 (Use Cases)

-### 获取 Token
+- **📁 全自主仓库维护**: Agent 在隔离工作区内自动分析代码、运行测试并应用补丁。
+- **📊 深度财务数据审计**: 直接通过 Python 加载 Excel/CSV 原始数据（绕过 RAG），生成图表并实时预览。
+- **📝 长任务项目管理**: 自动拆解复杂任务并持久化 TOD·O 进度，跨会话跟踪执行状态。

-要使用 GitHub Copilot，您需要一个具有适当权限的 GitHub 个人访问令牌 (PAT)。
+---

-**获取步骤：**
+## ⭐ 支持与交流 (Support)

-1. 访问 [GitHub 令牌设置](https://github.com/settings/tokens?type=beta)。
-2. 点击 **Generate new token (fine-grained)**。
-3. **Repository access**: 选择 **Public Repositories** (最简单) 或 **All repositories**。
-4. **Permissions**:
-    * 如果您选择了 **All repositories**，则必须点击 **Account permissions**。
-    * 找到 **Copilot Requests**，选择 **Access**。
-5. 生成并复制令牌。
+如果这个插件对您有所帮助，请在 [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui) 项目上点个 **Star** 💫，这是对我最大的鼓励。

-## 📋 依赖说明
+---

-该 Pipe 会自动尝试安装以下依赖（如果环境中缺失）：
+## 🚀 安装与配置 (Installation)

-* `github-copilot-sdk` (Python 包)
-* `github-copilot-cli` (二进制文件，通过官方脚本安装)
+### 1) 导入函数

-## 故障排除 (Troubleshooting) ❓
+1. 打开 OpenWebUI，前往 **工作区** -> **函数**。
+2. 点击 **+** (创建函数)，完整粘贴 `github_copilot_sdk_cn.py` 的内容。
+3. 点击保存并确保已启用。

-* **图片及多模态使用说明**：
-  * 确保 `MODEL_ID` 是支持多模态的模型。
-* **看不到思考过程**：
-  * 确认已开启**流式输出**，且所选模型支持推理输出。
+### 2) 获取 Token (Get Token)

-## 更新日志
+1. 访问 [GitHub Token 设置](https://github.com/settings/tokens?type=beta)。
+2. 创建 **Fine-grained token**，授予 **Account permissions** -> **Copilot Requests** 访问权限。
+3. 将生成的 Token 填入插件的 `GH_TOKEN` 配置项中。

-完整历史请查看 GitHub 项目： [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)
+### 3) 配套插件 (强烈推荐)
+
+为了获得最佳的文件处理体验，请安装 [GitHub Copilot SDK Files Filter](https://openwebui.com/posts/403a62ee-a596-45e7-be65-fab9cc249dd6)。
+
+---
+
+## 📋 常见问题与依赖 (Troubleshooting)
+
+- **Agent 无法识别文件？**: 请确保已安装并启用了 Files Filter 插件，否则原始文件会被 RAG 干扰。
+- **看不到 TODO 进度条？**: 进度条仅在 Agent 使用 `update_todo` 工具（通常是处理复杂任务）时出现。
+- **依赖安装**: 本管道会自动尝试安装 `github-copilot-sdk` (Python 包) 和 `github-copilot-cli` (官方二进制)。
+
+---
+
+## 更新日志 (Changelog)
+
+完整历史记录请见 GitHub: [Awesome OpenWebUI](https://github.com/Fu-Jie/awesome-openwebui)
--- a/plugins/pipes/github-copilot-sdk/debug/test_system_message_resume.py
+++ b/plugins/pipes/github-copilot-sdk/debug/test_system_message_resume.py
@@ -0,0 +1,424 @@
+#!/usr/bin/env python3
+"""
+Copilot SDK System Message Test Script
+Tests whether system_message is properly applied during session.resume
+
+This script verifies the bug hypothesis:
+- session.resume with system_message config may not reliably update the system prompt
+
+Test scenarios:
+1. Create a new session with a custom system message
+2. Resume the same session with a DIFFERENT system message
+3. Ask the model to describe its current system instructions
+
+Requirements:
+- github-copilot-sdk>=0.1.23
+"""
+
+import asyncio
+import os
+import sys
+import time
+
+from copilot import CopilotClient
+from copilot.types import SessionConfig
+from copilot.generated.session_events import SessionEventType
+
+
+# Test system messages
+SYSTEM_MSG_A = """You are a helpful assistant named "ALPHA". 
+When asked about your name or identity, you MUST respond: "I am ALPHA, the first assistant."
+Always start your responses with "[ALPHA]:" prefix.
+"""
+
+SYSTEM_MSG_B = """You are a helpful assistant named "BETA". 
+When asked about your name or identity, you MUST respond: "I am BETA, the second assistant."
+Always start your responses with "[BETA]:" prefix.
+"""
+
+
+async def send_and_get_response(session, prompt: str) -> str:
+    """Send a message and collect the full response using event subscription."""
+    full_response = ""
+    response_complete = asyncio.Event()
+
+    def event_handler(event):
+        nonlocal full_response
+        if event.type == SessionEventType.ASSISTANT_MESSAGE_DELTA:
+            delta = getattr(event.data, "content", "") or ""
+            print(delta, end="", flush=True)
+            full_response += delta
+        elif event.type == SessionEventType.ASSISTANT_MESSAGE:
+            # Final complete message
+            content = getattr(event.data, "content", "") or ""
+            if content and not full_response:
+                full_response = content
+                print(content, end="", flush=True)
+        elif event.type == SessionEventType.SESSION_IDLE:
+            response_complete.set()
+        elif event.type == SessionEventType.ASSISTANT_TURN_END:
+            response_complete.set()
+
+    # Subscribe to events
+    unsubscribe = session.on(event_handler)
+
+    try:
+        # Send the message
+        await session.send({"prompt": prompt, "mode": "immediate"})
+        # Wait for completion (with timeout)
+        await asyncio.wait_for(response_complete.wait(), timeout=120)
+        print()  # newline after completion
+    finally:
+        unsubscribe()
+
+    return full_response
+
+
+async def test_new_session_system_message(client: CopilotClient):
+    """Test 1: New session with system message A"""
+    print("\n" + "=" * 60)
+    print("TEST 1: New Session with System Message A (ALPHA)")
+    print("=" * 60)
+
+    session_config = SessionConfig(
+        session_id="test-session-001",
+        model="gpt-5-mini",
+        streaming=True,
+        system_message={
+            "mode": "replace",
+            "content": SYSTEM_MSG_A,
+        },
+    )
+
+    session = await client.create_session(config=session_config)
+    print(f"✅ Created new session: {session.session_id}")
+
+    print("\n📤 Asking: 'What is your name?'")
+    print("📥 Response: ", end="")
+    response = await send_and_get_response(session, "What is your name?")
+
+    if "ALPHA" in response:
+        print("✅ SUCCESS: Model correctly identified as ALPHA")
+    else:
+        print("⚠️ WARNING: Model did NOT identify as ALPHA")
+
+    return session
+
+
+async def test_resume_session_with_new_system_message(
+    client: CopilotClient, session_id: str
+):
+    """Test 2: Resume session with DIFFERENT system message B"""
+    print("\n" + "=" * 60)
+    print("TEST 2: Resume Session with System Message B (BETA)")
+    print("=" * 60)
+
+    resume_config = {
+        "model": "gpt-5-mini",
+        "streaming": True,
+        "system_message": {
+            "mode": "replace",
+            "content": SYSTEM_MSG_B,
+        },
+    }
+
+    print(f"📋 Resume config includes system_message with mode='replace'")
+    print(f"📋 New system_message content: BETA identity")
+
+    session = await client.resume_session(session_id, resume_config)
+    print(f"✅ Resumed session: {session.session_id}")
+
+    print("\n📤 Asking: 'What is your name now? Did your identity change?'")
+    print("📥 Response: ", end="")
+    response = await send_and_get_response(
+        session, "What is your name now? Did your identity change?"
+    )
+
+    if "BETA" in response:
+        print("✅ SUCCESS: System message was updated to BETA")
+        return True
+    elif "ALPHA" in response:
+        print("❌ BUG CONFIRMED: System message was NOT updated (still ALPHA)")
+        return False
+    else:
+        print("⚠️ INCONCLUSIVE: Model response doesn't clearly indicate identity")
+        return None
+
+
+async def test_resume_without_system_message(client: CopilotClient, session_id: str):
+    """Test 3: Resume session without specifying system_message"""
+    print("\n" + "=" * 60)
+    print("TEST 3: Resume Session WITHOUT System Message")
+    print("=" * 60)
+
+    resume_config = {
+        "model": "gpt-4o",
+        "streaming": True,
+        # No system_message specified
+    }
+
+    session = await client.resume_session(session_id, resume_config)
+    print(f"✅ Resumed session: {session.session_id}")
+
+    print("\n📤 Asking: 'What is your name? Tell me your current identity.'")
+    print("📥 Response: ", end="")
+    response = await send_and_get_response(
+        session, "What is your name? Tell me your current identity."
+    )
+
+    if "ALPHA" in response:
+        print(
+            "ℹ️ Without system_message: Model still remembers ALPHA from original session"
+        )
+    elif "BETA" in response:
+        print("ℹ️ Without system_message: Model remembers BETA from Test 2")
+    else:
+        print("ℹ️ Model identity unclear")
+
+
+async def main():
+    print("=" * 60)
+    print("🧪 Copilot SDK System Message Resume Test")
+    print("=" * 60)
+    print(f"Time: {time.strftime('%Y-%m-%d %H:%M:%S')}")
+    print(f"Testing with SDK from: {CopilotClient.__module__}")
+
+    # Create client with explicit CLI path if provided
+    cli_path = os.environ.get("COPILOT_CLI_PATH")
+    client_config = {"log_level": "info"}
+    if cli_path:
+        client_config["cli_path"] = cli_path
+
+    client = CopilotClient(client_config)
+
+    try:
+        await client.start()
+        print("✅ Client started successfully")
+
+        # Test 1: Create new session with system message A
+        session = await test_new_session_system_message(client)
+        session_id = session.session_id
+
+        # Wait a bit before resuming
+        print("\n⏳ Waiting 2 seconds before resume test...")
+        await asyncio.sleep(2)
+
+        # Test 2: Resume with different system message B
+        bug_confirmed = await test_resume_session_with_new_system_message(
+            client, session_id
+        )
+
+        # Test 3: Resume without system message
+        await test_resume_without_system_message(client, session_id)
+
+        # Summary
+        print("\n" + "=" * 60)
+        print("📊 TEST SUMMARY (Native Copilot)")
+        print("=" * 60)
+        if bug_confirmed is False:
+            print(
+                "❌ BUG CONFIRMED: session.resume does NOT apply system_message updates"
+            )
+            print("   The system message from create_session persists even when")
+            print("   resume_session specifies a different system_message.")
+            print("\n   WORKAROUND: Inject system context into user prompt instead.")
+        elif bug_confirmed is True:
+            print("✅ NO BUG: session.resume correctly updates system_message")
+        else:
+            print("⚠️ INCONCLUSIVE: Could not determine if bug exists")
+
+    except Exception as e:
+        print(f"❌ Error: {e}")
+        import traceback
+
+        traceback.print_exc()
+    finally:
+        await client.stop()
+        print("\n✅ Client stopped")
+
+
+# =============================================================================
+# BYOK OpenAI Test
+# =============================================================================
+
+
+async def test_byok_new_session(client: CopilotClient, provider_config: dict):
+    """BYOK Test 1: New session with BYOK provider and system message A"""
+    print("\n" + "=" * 60)
+    print("BYOK TEST 1: New Session with BYOK Provider + System Message A (ALPHA)")
+    print("=" * 60)
+    print(
+        f"📋 Provider: {provider_config.get('type')} @ {provider_config.get('base_url')}"
+    )
+
+    session_config = SessionConfig(
+        session_id="byok-test-session-001",
+        model="gpt-4o",  # or your model name
+        streaming=True,
+        provider=provider_config,
+        system_message={
+            "mode": "replace",
+            "content": SYSTEM_MSG_A,
+        },
+    )
+
+    session = await client.create_session(config=session_config)
+    print(f"✅ Created BYOK session: {session.session_id}")
+
+    print("\n📤 Asking: 'What is your name?'")
+    print("📥 Response: ", end="")
+    response = await send_and_get_response(session, "What is your name?")
+
+    if "ALPHA" in response:
+        print("✅ SUCCESS: Model correctly identified as ALPHA")
+    else:
+        print("⚠️ WARNING: Model did NOT identify as ALPHA")
+
+    return session
+
+
+async def test_byok_resume_with_new_system_message(
+    client: CopilotClient, session_id: str, provider_config: dict
+):
+    """BYOK Test 2: Resume BYOK session with DIFFERENT system message B"""
+    print("\n" + "=" * 60)
+    print("BYOK TEST 2: Resume BYOK Session with System Message B (BETA)")
+    print("=" * 60)
+
+    resume_config = {
+        "model": "gpt-4o",
+        "streaming": True,
+        "provider": provider_config,
+        "system_message": {
+            "mode": "replace",
+            "content": SYSTEM_MSG_B,
+        },
+    }
+
+    print(f"📋 Resume config includes system_message with mode='replace'")
+    print(f"📋 New system_message content: BETA identity")
+    print(
+        f"📋 Provider: {provider_config.get('type')} @ {provider_config.get('base_url')}"
+    )
+
+    session = await client.resume_session(session_id, resume_config)
+    print(f"✅ Resumed BYOK session: {session.session_id}")
+
+    print("\n📤 Asking: 'What is your name now? Did your identity change?'")
+    print("📥 Response: ", end="")
+    response = await send_and_get_response(
+        session, "What is your name now? Did your identity change?"
+    )
+
+    if "BETA" in response:
+        print("✅ SUCCESS: System message was updated to BETA")
+        return True
+    elif "ALPHA" in response:
+        print("❌ BUG CONFIRMED: System message was NOT updated (still ALPHA)")
+        return False
+    else:
+        print("⚠️ INCONCLUSIVE: Model response doesn't clearly indicate identity")
+        return None
+
+
+async def main_byok():
+    """Run BYOK-specific tests"""
+    print("=" * 60)
+    print("🧪 Copilot SDK BYOK System Message Resume Test")
+    print("=" * 60)
+    print(f"Time: {time.strftime('%Y-%m-%d %H:%M:%S')}")
+
+    # Get BYOK configuration from environment
+    byok_api_key = os.environ.get("BYOK_API_KEY") or os.environ.get("OPENAI_API_KEY")
+    byok_base_url = os.environ.get("BYOK_BASE_URL", "https://api.openai.com/v1")
+    byok_model = os.environ.get("BYOK_MODEL", "gpt-4o")
+
+    if not byok_api_key:
+        print(
+            "❌ Error: Please set BYOK_API_KEY or OPENAI_API_KEY environment variable"
+        )
+        print("   export BYOK_API_KEY='your_api_key'")
+        print("   export BYOK_BASE_URL='https://api.openai.com/v1'  # optional")
+        print("   export BYOK_MODEL='gpt-4o'  # optional")
+        return
+
+    provider_config = {
+        "type": "openai",
+        "base_url": byok_base_url,
+        "api_key": byok_api_key,
+    }
+
+    print(f"📋 BYOK Provider: openai @ {byok_base_url}")
+    print(f"📋 BYOK Model: {byok_model}")
+
+    # Create client
+    cli_path = os.environ.get("COPILOT_CLI_PATH")
+    client_config = {"log_level": "info"}
+    if cli_path:
+        client_config["cli_path"] = cli_path
+
+    client = CopilotClient(client_config)
+
+    try:
+        await client.start()
+        print("✅ Client started successfully")
+
+        # BYOK Test 1: Create new session with BYOK provider
+        session = await test_byok_new_session(client, provider_config)
+        session_id = session.session_id
+
+        # Wait a bit before resuming
+        print("\n⏳ Waiting 2 seconds before resume test...")
+        await asyncio.sleep(2)
+
+        # BYOK Test 2: Resume with different system message B
+        bug_confirmed = await test_byok_resume_with_new_system_message(
+            client, session_id, provider_config
+        )
+
+        # Summary
+        print("\n" + "=" * 60)
+        print("📊 BYOK TEST SUMMARY")
+        print("=" * 60)
+        if bug_confirmed is False:
+            print(
+                "❌ BYOK BUG CONFIRMED: session.resume does NOT apply system_message updates"
+            )
+            print("   In BYOK mode, the system message from create_session persists")
+            print("   even when resume_session specifies a different system_message.")
+            print("\n   WORKAROUND: Inject system context into user prompt instead.")
+        elif bug_confirmed is True:
+            print("✅ BYOK NO BUG: session.resume correctly updates system_message")
+        else:
+            print("⚠️ BYOK INCONCLUSIVE: Could not determine if bug exists")
+
+    except Exception as e:
+        print(f"❌ Error: {e}")
+        import traceback
+
+        traceback.print_exc()
+    finally:
+        await client.stop()
+        print("\n✅ Client stopped")
+
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser(
+        description="Copilot SDK System Message Resume Test"
+    )
+    parser.add_argument(
+        "--byok",
+        action="store_true",
+        help="Run BYOK (Bring Your Own Key) test instead of native Copilot test",
+    )
+    args = parser.parse_args()
+
+    if args.byok:
+        print("Running BYOK test mode...")
+        asyncio.run(main_byok())
+    else:
+        print("Running native Copilot test mode...")
+        print("(Use --byok flag for BYOK provider test)")
+        asyncio.run(main())
--- a/plugins/pipes/github-copilot-sdk/github_copilot_sdk.py
+++ b/plugins/pipes/github-copilot-sdk/github_copilot_sdk.py
--- a/plugins/pipes/github-copilot-sdk/github_copilot_sdk_cn.py
+++ b/plugins/pipes/github-copilot-sdk/github_copilot_sdk_cn.py
--- a/plugins/pipes/github-copilot-sdk/scripts/sync_to_workspace.py
+++ b/plugins/pipes/github-copilot-sdk/scripts/sync_to_workspace.py
@@ -0,0 +1,103 @@
+import os
+import sys
+import json
+import asyncio
+from typing import List, Dict
+
+# 1. 尝试导入 OpenWebUI 环境
+try:
+    from open_webui.models.models import Models, ModelForm, ModelMeta, ModelParams
+    from open_webui.internal.db import get_session
+except ImportError:
+    print("❌ 错误: 无法导入 OpenWebUI 模块。请确保在 OpenWebUI 环境（如 conda ai）中运行此脚本。")
+    sys.exit(1)
+
+# 2. 导入 Copilot SDK
+try:
+    from copilot import CopilotClient
+except ImportError:
+    print("❌ 错误: 无法导入 copilot SDK。请运行: pip install github-copilot-sdk==0.1.23")
+    sys.exit(1)
+
+async def fetch_real_models() -> List[Dict]:
+    gh_token = os.environ.get("GH_TOKEN")
+    if not gh_token:
+        print("❌ 错误: 未设置 GH_TOKEN 环境变量。")
+        return []
+
+    client = CopilotClient()
+    try:
+        await client.start()
+        raw_models = await client.list_models()
+        processed = []
+        for m in raw_models:
+            m_id = getattr(m, "id", str(m))
+            # 提取倍率
+            billing = getattr(m, "billing", {})
+            if not isinstance(billing, dict): billing = vars(billing)
+            multiplier = billing.get("multiplier", 1)
+            
+            # 提取能力
+            cap = getattr(m, "capabilities", None)
+            vision = False
+            reasoning = False
+            if cap:
+                supports = getattr(cap, "supports", {})
+                if not isinstance(supports, dict): supports = vars(supports)
+                vision = supports.get("vision", False)
+                reasoning = supports.get("reasoning_effort", False)
+
+            processed.append({
+                "id": m_id,
+                "name": f"GitHub Copilot ({m_id})",
+                "vision": vision,
+                "reasoning": reasoning,
+                "multiplier": multiplier
+            })
+        return processed
+    except Exception as e:
+        print(f"❌ 获取模型失败: {e}")
+        return []
+    finally:
+        await client.stop()
+
+async def sync_to_db():
+    models = await fetch_real_models()
+    if not models: return
+
+    print(f"🔄 发现 {len(models)} 个 Copilot 模型，正在同步到工作区...")
+    
+    # 默认管理员 ID
+    admin_user_id = "00000000-0000-0000-0000-000000000000" 
+
+    for m in models:
+        # 对应插件中的 ID 格式
+        full_id = f"copilot-{m['id']}"
+        
+        existing = Models.get_model_by_id(full_id)
+        if existing:
+            print(f"⚠️ 跳过: {full_id} (已存在)")
+            continue
+
+        form_data = ModelForm(
+            id=full_id,
+            base_model_id=None,
+            name=m['name'],
+            meta=ModelMeta(
+                description=f"GitHub Copilot 官方模型。倍率: {m['multiplier']}x。支持推理: {m['reasoning']}。",
+                capabilities={
+                    "vision": m['vision'],
+                    "reasoning": m['reasoning']
+                }
+            ),
+            params=ModelParams()
+        )
+
+        try:
+            if Models.insert_new_model(form_data, admin_user_id):
+                print(f"✅ 成功同步: {m['name']}")
+        except Exception as e:
+            print(f"❌ 同步 {m['id']} 失败: {e}")
+
+if __name__ == "__main__":
+    asyncio.run(sync_to_db())
--- a/plugins/pipes/github-copilot-sdk/scripts/test_sdk.py
+++ b/plugins/pipes/github-copilot-sdk/scripts/test_sdk.py
@@ -0,0 +1,111 @@
+import asyncio
+import os
+import json
+import time
+from typing import Any
+from datetime import datetime
+from copilot import CopilotClient
+
+# 自定义 JSON 编码器，处理 datetime 等对象
+class SDKEncoder(json.JSONEncoder):
+    def default(self, obj):
+        if isinstance(obj, datetime):
+            return obj.isoformat()
+        if hasattr(obj, "to_dict"):
+            return obj.to_dict()
+        try:
+            return super().default(obj)
+        except TypeError:
+            return str(obj)
+
+async def run_debug_test(client, model_id, effort, prompt):
+    print(f"\n" + "🔍" * 15)
+    print(f"DEBUGGING: {model_id} | {effort.upper()}")
+    print("🔍" * 15)
+    
+    session_config = {
+        "model": model_id,
+        "reasoning_effort": effort
+    }
+    
+    queue = asyncio.Queue()
+    done = asyncio.Event()
+    SENTINEL = object()
+
+    def handler(event):
+        try:
+            etype = getattr(event, "type", "unknown")
+            if hasattr(etype, "value"): etype = etype.value
+            
+            raw_data = {}
+            # 优先尝试 SDK 自带的 to_dict
+            if hasattr(event, "to_dict"):
+                raw_data = event.to_dict()
+            elif hasattr(event, "data"):
+                data_obj = event.data
+                if isinstance(data_obj, dict):
+                    raw_data = data_obj
+                elif hasattr(data_obj, "to_dict"):
+                    raw_data = data_obj.to_dict()
+                else:
+                    raw_data = {k: v for k, v in vars(data_obj).items() if not k.startswith('_')}
+            else:
+                raw_data = {k: v for k, v in vars(event).items() if not k.startswith('_')}
+            
+            queue.put_nowait((etype, raw_data))
+            
+            if etype in ["session.idle", "session.error"]:
+                done.set()
+                queue.put_nowait(SENTINEL)
+        except Exception as e:
+            print(f"Handler Error: {e}")
+
+    try:
+        session = await client.create_session(config=session_config)
+        unsubscribe = session.on(handler)
+        
+        print(f"Sending prompt...")
+        asyncio.create_task(session.send({"prompt": prompt, "mode": "immediate"}))
+        
+        event_count = 0
+        while True:
+            try:
+                item = await asyncio.wait_for(queue.get(), timeout=180)
+                if item is SENTINEL: break
+                
+                etype, data = item
+                event_count += 1
+                
+                # 打印所有事件，不进行过滤，这样我们可以看到完整的生命周期
+                print(f"\n[#{event_count} EVENT: {etype}]")
+                print(json.dumps(data, indent=2, ensure_ascii=False, cls=SDKEncoder))
+                
+            except asyncio.TimeoutError:
+                print("\n⚠️ Timeout: No events for 180s")
+                break
+
+        unsubscribe()
+        await session.destroy()
+        
+    except Exception as e:
+        print(f"❌ Session Failed: {e}")
+
+async def main():
+    gh_token = os.environ.get("GH_TOKEN")
+    if not gh_token:
+        print("❌ Error: GH_TOKEN not set")
+        return
+
+    client = CopilotClient()
+    await client.start()
+    
+    # 使用一个简单但需要思考的问题
+    prompt = "123 * 456 等于多少？请给出思考过程。"
+    target_model = "gpt-5-mini" 
+    
+    await run_debug_test(client, target_model, "high", prompt)
+    
+    await client.stop()
+
+if __name__ == "__main__":
+    asyncio.run(main())
Author	SHA1	Message	Date
fujie	e17818bf6c	chore: re-trigger release for v0.6.0	2026-02-09 19:54:49 +08:00
fujie	c5c9fd9d57	fix(ci): fix shell script error when commit messages contain backticks	2026-02-09 19:53:23 +08:00
fujie	5c9875d390	feat: add `openwebui_id` to the GitHub Copilot SDK files filter plugin metadata.	2026-02-09 19:48:41 +08:00
fujie	9185f88d40	feat: release github-copilot-sdk v0.6.0 and files-filter v0.1.2	2026-02-09 19:45:23 +08:00
github-actions[bot]	1fcad993ea	chore: update community stats - followers increased (208 -> 210)	2026-02-09 03:10:29 +00:00
github-actions[bot]	1594ea3a20	chore: update community stats - followers increased (207 -> 208)	2026-02-09 00:44:14 +00:00
github-actions[bot]	ee681ddad9	chore: update community stats - followers increased (206 -> 207)	2026-02-08 21:11:27 +00:00
github-actions[bot]	881597fd51	chore: update community stats - followers increased (205 -> 206)	2026-02-08 07:21:06 +00:00
github-actions[bot]	5772b1c65f	chore: update community stats - plugin version updated, points increased (258 -> 259)	2026-02-08 00:53:22 +00:00
fujie	3d4b4b96e8	fix(docs): Repair broken valves table in README	2026-02-08 08:33:02 +08:00
fujie	409d2f663f	feat(copilot): Release v0.5.1 - Smarter BYOK, Tool Caching & Refined Docs	2026-02-08 08:21:32 +08:00
github-actions[bot]	de9948a5b0	chore: update community stats - followers increased (204 -> 205)	2026-02-07 23:12:20 +00:00
github-actions[bot]	0a51b4cfca	chore: update community stats - points increased (257 -> 258)	2026-02-07 18:13:00 +00:00
github-actions[bot]	f8eb096300	chore: update community stats - plugin version updated	2026-02-07 11:08:46 +00:00
fujie	f6369a1591	feat: release export-to-docx v0.4.4 w/ formatting & font fixes	2026-02-07 18:14:11 +08:00
github-actions[bot]	81634f57fa	chore: update community stats - points increased (254 -> 257)	2026-02-07 09:12:46 +00:00
github-actions[bot]	70dc62aaab	chore: update community stats - points increased (246 -> 254), followers increased (203 -> 204)	2026-02-07 06:19:35 +00:00
fujie	8762c05e04	docs: add explicit Copilot subscription requirement to README	2026-02-07 13:41:23 +08:00
github-actions[bot]	db52ac1041	chore: update community stats - plugin version updated	2026-02-07 05:23:23 +00:00
fujie	795ac34cd9	chore: archive development research artifacts and debug scripts	2026-02-07 12:49:42 +08:00
fujie	89a12a4fe8	docs: enforce English-only for commit messages and release notes	2026-02-07 12:38:46 +08:00