Improve GitHub learning cards with DeepWiki context

Author: jjjojoj

CHANGELOG.md

@@ -26,6 +26,7 @@ The format is inspired by Keep a Changelog, and version tags follow the reposito
 - Coding-plan analysis now defaults to `glm-4.7`, instead of probing `glm-5` first
 - BigModel / z.ai `chat/completions` summary requests now disable `thinking` by default so `glm-4.7` returns final answer text instead of reasoning-only payloads
 - GitHub cards now use the exact repository name as the note title, plus a DeepWiki-inspired breakdown: what problem the repo solves, how the system is layered, which paths to read first, and how to start digging into the code
+- GitHub extractor now supplements repository metadata with DeepWiki overview context when available, and final knowledge cards prefer verified repo entrypoints / paths over raw LLM path guesses
 - Knowledge-card notes no longer dump full raw content for high-confidence web / GitHub captures; folded evidence is only shown for lower-confidence fallback cases or transcript-style media
 - Obsidian knowledge indexes now insert new cards under the correct date heading instead of appending entries into the wrong day block
 - README and `.env.example` now clarify that Feishu is unsupported and that `auto` output switches to Obsidian as soon as a vault path is configured

README.md

@@ -17,7 +17,7 @@ English | [简体中文](./README.zh-CN.md)
 - `.env` files inside the skill directory are loaded automatically, so local OpenAI-compatible settings can take effect without manual export.
 - Non-OpenAI providers such as GLM / MiniMax can now use `chat/completions` style endpoints, while official OpenAI still uses `responses`.
 - If you already configured `zai` / GLM Coding Plan inside OpenClaw, this skill can now reuse that local provider config instead of maintaining a second key by hand.
-- GitHub repositories now generate repository-specific knowledge cards and are automatically linked into `MOC/GitHub` category branches such as `AI Agent`, `SaaS`, and `FastAPI`.
+- GitHub repositories now generate repository-specific knowledge cards, enrich from DeepWiki overview pages when available, and are automatically linked into `MOC/GitHub` category branches such as `AI Agent`, `SaaS`, and `FastAPI`.
 
 ## What's New In v2.4.0
 
@@ -67,7 +67,7 @@ Stable-release validation last refreshed on `2026-04-19`:
 
 | Platform | Status | Notes |
 | --- | --- | --- |
-| GitHub | Stable | Uses GitHub API + README extraction |
+| GitHub | Stable | Uses GitHub API + README, and supplements with DeepWiki overview when available |
 | Generic web pages | Stable | Main path uses `trafilatura` |
 | WeChat | Stable | Usually succeeds via `Scrapling` |
 | Zhihu / CSDN | Stable | Real links verified |
@@ -92,7 +92,7 @@ See [docs/release-validation.md](./docs/release-validation.md) for the latest re
 
 | Capability | What it does |
 | --- | --- |
-| GitHub extractor | Pulls repo metadata, topics, stars, default branch, and README |
+| GitHub extractor | Pulls repo metadata, topics, stars, default branch, README, and DeepWiki overview context when available |
 | Web extractor | Uses `trafilatura` for article-style pages |
 | Dynamic-page fallback | Uses `Scrapling` for harder pages |
 | Media pipeline | Uses `yt-dlp` subtitles first, then `ffmpeg + whisper-cli` |
@@ -272,6 +272,7 @@ The default Obsidian note set includes:
 - YAML frontmatter for Dataview / filtering / tagging
 - `_index.md` as the vault-folder root entry point
 - GitHub cards automatically link into `MOC/GitHub` and category pages such as `AI Agent`, `SaaS`, `FastAPI`, or `Automation`
+- GitHub cards prefer student-friendly project breakdowns: what problem the repo solves, how the system is layered, which entrypoints matter, and what to read first
 - high-confidence web / GitHub cards stay clean by default; folded raw evidence only appears for fallback / partial captures or transcript-style media
 - no `sources/` directory in the default knowledge-card layout
 
@@ -304,7 +305,7 @@ Typical CLI response:
 
 Different sources use different pipelines on purpose:
 
-- GitHub repositories: `GitHub API + README`
+- GitHub repositories: `GitHub API + README (+ DeepWiki overview when available)`
 - Regular web pages: `trafilatura`
 - Dynamic / anti-bot pages: `Scrapling`
 - Media links: `yt-dlp` subtitles first

README.zh-CN.md

@@ -17,7 +17,7 @@
 - skill 目录下的 `.env` 会自动加载，不需要再手工 `export` 才能让本地配置生效。
 - 对 GLM / MiniMax 这类不支持 Responses API 的服务，现在会优先兼容 `chat/completions`。
 - 如果你已经在 OpenClaw 里配置了 `zai` / GLM Coding Plan，这个 skill 现在也可以直接复用那份本地 provider 配置，不用再单独维护第二份 key。
-- GitHub 仓库现在会生成专门的仓库知识卡片，并自动挂到 Obsidian 里的 `MOC/GitHub` 分支，再细分到 `AI Agent`、`SaaS`、`FastAPI` 等分类页。
+- GitHub 仓库现在会生成专门的仓库知识卡片；在可用时还会补充 DeepWiki overview 作为结构化上下文，并自动挂到 Obsidian 里的 `MOC/GitHub` 分支，再细分到 `AI Agent`、`SaaS`、`FastAPI` 等分类页。
 
 ## 本次更新（v2.4.0）
 
@@ -67,7 +67,7 @@ https://github.com/jjjojoj/openclaw-content-processor.git
 
 | 平台 | 状态 | 说明 |
 | --- | --- | --- |
-| GitHub | 稳定 | 走 GitHub API + README 专项抽取 |
+| GitHub | 稳定 | 走 GitHub API + README 专项抽取，可用时再补充 DeepWiki overview |
 | 普通网页 | 稳定 | 主链路为 `trafilatura` |
 | 微信公众号 | 稳定 | 通常由 `Scrapling` 处理 |
 | 知乎 / CSDN | 稳定 | 已完成真实链接验证 |
@@ -92,7 +92,7 @@ https://github.com/jjjojoj/openclaw-content-processor.git
 
 | 能力 | 说明 |
 | --- | --- |
-| GitHub 抽取 | 提取仓库描述、stars、topics、默认分支、README |
+| GitHub 抽取 | 提取仓库描述、stars、topics、默认分支、README，并在可用时补充 DeepWiki overview 上下文 |
 | 网页正文抽取 | 通过 `trafilatura` 获取文章正文 |
 | 动态页 fallback | 通过 `Scrapling` 处理更难抓的页面 |
 | 媒体处理链路 | 优先 `yt-dlp` 字幕，无字幕则 `ffmpeg + whisper-cli` |
@@ -272,6 +272,7 @@ Obsidian 模式会生成：
 - 适合 Dataview / 过滤 / 标签的 YAML frontmatter
 - 位于 Vault 根目录下的 `_index.md`
 - GitHub 卡片会自动接入 `MOC/GitHub` 以及 `AI Agent`、`SaaS`、`FastAPI`、`Automation` 等分类页
+- GitHub 卡片会优先整理成适合学生学习的项目拆解结构：项目解决什么、系统怎么分层、从哪些入口读起
 - 高置信网页 / GitHub 卡片默认不再塞整段原始内容；只有 fallback / 部分成功 / 转写型媒体时才显示折叠证据
 - 默认不再生成 `sources/` 目录
 
@@ -304,7 +305,7 @@ CLI 结束时会输出一个 JSON 概览，例如：
 
 不同来源会走不同管线：
 
-- GitHub 仓库：`GitHub API + README`
+- GitHub 仓库：`GitHub API + README（可用时补充 DeepWiki overview）`
 - 普通网页：`trafilatura`
 - 动态 / 反爬页面：`Scrapling`
 - 媒体链接：优先 `yt-dlp` 字幕