doc-daily-digest

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Daily Digest

每日摘要

Process an Obsidian daily note by classifying raw items, fetching/researching content, creating structured notes, and replacing raw items with wikilinks.

通过分类原始条目、抓取/调研内容、创建结构化笔记、用wikilinks替换原始条目，处理Obsidian每日笔记。

Prerequisites

前置条件

Skill	Required	Purpose
doc-obsidian	Yes	Vault CRUD via notesmd-cli + search via qmd
res-x	For X/Twitter URLs	Fetch tweet content via xAI
res-deep	For loose ideas	Multi-round research

Optional dependency:

scrapling

-- headless browser fallback for sites that block WebFetch (403, empty content, captcha pages). Install with:

uv tool install 'scrapling[all]'

技能	是否必填	用途
doc-obsidian	是	通过notesmd-cli实现知识库CRUD，通过qmd实现搜索
res-x	处理X/Twitter URL时需要	通过xAI抓取推文内容
res-deep	处理零散想法时需要	多轮调研

可选依赖：

scrapling

-- 针对拦截WebFetch的站点（403、内容为空、验证码页面）的无头浏览器降级方案。安装命令：

uv tool install 'scrapling[all]'

Workflow

工作流

Step 0: Setup → Step 1: Scan & Classify → Step 2: Process → Step 3: Create Notes → Step 4: Update Daily → Step 5: Re-index & Report

Step 0: Setup → Step 1: Scan & Classify → Step 2: Process → Step 3: Create Notes → Step 4: Update Daily → Step 5: Re-index & Report

Step 0: Setup

步骤0：初始化设置

Run all three checks:

bash

undefined

运行以下三项检查：

bash

undefined

1. Vault path

1. 知识库路径

VAULT=$(notesmd-cli print-default --path-only)

2. Read daily note (today or user-specified date)

2. 读取每日笔记（今日或用户指定日期）

DATE=$(date '+%Y-%m-%d') notesmd-cli print "$DATE"

3. xAI key (needed for res-x and res-deep full mode)

3. xAI密钥（res-x和res-deep完整模式需要）

security find-generic-password -s "xai-api" -w ~/Library/Keychains/claude-keys.keychain-db 2>/dev/null && echo "XAI_AVAILABLE=true" || echo "XAI_AVAILABLE=false"


If user specifies a date, use that instead of today.

security find-generic-password -s "xai-api" -w ~/Library/Keychains/claude-keys.keychain-db 2>/dev/null && echo "XAI_AVAILABLE=true" || echo "XAI_AVAILABLE=false"


如果用户指定了日期，则使用指定日期代替今日。

Step 1: Scan & Classify

步骤1：扫描与分类

Parse the daily note and classify every item. Items live in these sections:

## Notes

## Log

## Links

解析每日笔记并对每个条目分类。条目位于以下区块：

## Notes

、

## Log

、

## Links

。

Classification Rules

分类规则

Type	Pattern	Action
Skip	`[[wikilink]]` anywhere in line	Already processed — skip
Skip	Section headers ( `##` ), frontmatter, empty lines, task checkboxes	Structural — skip
X tweet	URL matching `https://(x\.com\|twitter\.com)/\w+/status/\d+`	Fetch via res-x
X article	URL matching `https://(x\.com\|twitter\.com)/i/article/[\w-]+`	Fetch via res-x
GitHub repo	URL matching `https://github\.com/[\w-]+/[\w-]+`	WebFetch repo page
Web URL	Any other `https://...` URL	WebFetch page
Loose idea	Non-empty text that is not a URL, not a wikilink, not structural	Deep research via res-deep

类型	匹配规则	处理动作
跳过	行内任意位置包含 `[[wikilink]]`	已处理，直接跳过
跳过	区块标题（ `##` 开头）、前置元数据、空行、任务复选框	结构元素，直接跳过
X推文	URL匹配 `https://(x\.com\|twitter\.com)/\w+/status/\d+`	通过res-x抓取
X文章	URL匹配 `https://(x\.com\|twitter\.com)/i/article/[\w-]+`	通过res-x抓取
GitHub仓库	URL匹配 `https://github\.com/[\w-]+/[\w-]+`	WebFetch抓取仓库页面
网页URL	其他所有 `https://...` 格式的URL	WebFetch抓取页面
零散想法	非空文本，且不是URL、不是wikilink、不是结构元素	通过res-deep开展深度调研

Present Classification

分类结果展示

Before processing, show the user a classification table:

undefined

处理前，向用户展示分类表格：

undefined

Daily Digest: {DATE}

#	Section	Type	Item (truncated)	Action
1	Links	X tweet	https://x.com/user/status/123...	res-x fetch
2	Notes	Loose idea	Train a model to click on...	res-deep
3	Links	GitHub	https://github.com/org/repo	WebFetch
4	Log	Skip	[[already-processed]] — ...	skip


Ask user to confirm or exclude items before proceeding. User may:
- Approve all
- Exclude specific items by number
- Change action for an item (e.g., skip an idea, or upgrade a URL to res-deep)

#	区块	类型	条目（截断）	处理动作
1	Links	X tweet	https://x.com/user/status/123...	res-x fetch
2	Notes	Loose idea	Train a model to click on...	res-deep
3	Links	GitHub	https://github.com/org/repo	WebFetch
4	Log	Skip	[[already-processed]] — ...	skip


继续操作前请用户确认或排除条目。用户可执行以下操作：
- 全部批准
- 按编号排除特定条目
- 修改某个条目的处理动作（例如：跳过某个想法，或将某个URL升级为res-deep处理）

Step 2: Process Items

步骤2：处理条目

Process approved items. Run independent fetches in parallel where possible.

处理已批准的条目。可并行执行的独立抓取任务尽量并行运行。

X/Twitter URLs

X/Twitter URL

Requires xAI key (XAI_AVAILABLE=true).

bash

uv run ~/.claude/skills/res-x/scripts/x_fetch.py fetch "URL1" "URL2" "URL3"

The script batches 3 URLs per API call. Extract from results:

Author handle and display name
Full tweet text
Engagement metrics (likes, reposts, replies, views)
Thread context and quoted tweets if present

If XAI_AVAILABLE=false, report that X URLs require xAI key and skip them.

需要xAI密钥（XAI_AVAILABLE=true）。

bash

uv run ~/.claude/skills/res-x/scripts/x_fetch.py fetch "URL1" "URL2" "URL3"

脚本每次API调用批量处理3个URL。从结果中提取：

作者账号和显示名称
完整推文文本
互动指标（点赞、转发、回复、浏览量）
线程上下文和引用的推文（如果有）

如果XAI_AVAILABLE=false，提示用户X URL需要xAI密钥并跳过这些条目。

GitHub URLs

GitHub URL

WebFetch: https://github.com/{owner}/{repo}
Prompt: "Extract: repo name, description, star count, language, license, last update date, and a 2-3 sentence summary of what this project does based on the README."

Scrapling fallback: If WebFetch returns 403, empty content, a captcha page, or a blocked response, retry using the auto-escalation protocol from cli-web-scrape:

scrapling extract get "URL" /tmp/scrapling-fallback.md

→ Read → validate content

If content is thin (JS-only shell, no data) →

scrapling extract fetch "URL" /tmp/scrapling-fallback.md --network-idle --disable-resources

→ Read → validate

If still blocked →

scrapling extract stealthy-fetch "URL" /tmp/scrapling-fallback.md --solve-cloudflare

All tiers fail → note failure and move on

WebFetch: https://github.com/{owner}/{repo}
Prompt: "Extract: repo name, description, star count, language, license, last update date, and a 2-3 sentence summary of what this project does based on the README."

Scrapling降级方案： 如果WebFetch返回403、空内容、验证码页面或被拦截的响应，使用cli-web-scrape的自动升级协议重试：

scrapling extract get "URL" /tmp/scrapling-fallback.md

→ 读取 → 验证内容

如果内容过短（仅JS外壳，无数据）→

scrapling extract fetch "URL" /tmp/scrapling-fallback.md --network-idle --disable-resources

→ 读取 → 验证

如果仍然被拦截 →

scrapling extract stealthy-fetch "URL" /tmp/scrapling-fallback.md --solve-cloudflare

所有层级都失败 → 记录失败并继续处理下一个条目

Web URLs

网页URL

WebFetch: {URL}
Prompt: "Extract: page title, author if available, publication date if available, and a 3-5 sentence summary of the key content."

Scrapling fallback: If WebFetch returns 403, empty content, a captcha page, or a blocked response, retry using the auto-escalation protocol from cli-web-scrape:

scrapling extract get "URL" /tmp/scrapling-fallback.md

→ Read → validate content

If content is thin (JS-only shell, no data) →

scrapling extract fetch "URL" /tmp/scrapling-fallback.md --network-idle --disable-resources

→ Read → validate

If still blocked →

scrapling extract stealthy-fetch "URL" /tmp/scrapling-fallback.md --solve-cloudflare

All tiers fail → note failure and move on

WebFetch: {URL}
Prompt: "Extract: page title, author if available, publication date if available, and a 3-5 sentence summary of the key content."

Scrapling降级方案： 如果WebFetch返回403、空内容、验证码页面或被拦截的响应，使用cli-web-scrape的自动升级协议重试：

scrapling extract get "URL" /tmp/scrapling-fallback.md

→ 读取 → 验证内容

如果内容过短（仅JS外壳，无数据）→

scrapling extract fetch "URL" /tmp/scrapling-fallback.md --network-idle --disable-resources

→ 读取 → 验证

如果仍然被拦截 →

scrapling extract stealthy-fetch "URL" /tmp/scrapling-fallback.md --solve-cloudflare

所有层级都失败 → 记录失败并继续处理下一个条目

Loose Ideas

零散想法

Invoke res-deep skill with the idea text as the query. Use

quick

depth (1 round, 10-15 sources) unless user requests deeper research.

For ideas, the res-deep output becomes the note body directly.

以想法文本为查询词调用res-deep技能。默认使用

quick

深度（1轮调研，10-15个信息源），除非用户要求更深层级的调研。

对于想法类条目，res-deep的输出直接作为笔记正文。

Step 3: Create Notes

步骤3：创建笔记

For each processed item, create an Obsidian note.

为每个处理完成的条目创建一条Obsidian笔记。

Note Naming

笔记命名规则

Type	Naming Pattern	Example
X tweet	`{topic}-{descriptor}` from content	`scrapling-undetectable-web-scraping`
X article	`{author}-x-article-{date}`	`irabukht-x-article-2026-02-23`
GitHub repo	`{repo-name}`	`scrapling` or `huggingface-skills-agent-plugins`
Web page	`{topic}-{descriptor}` from title	`kubernetes-practical-learning-path`
Loose idea	`{concept}-{descriptor}`	`agent-sort-through-the-slop`
Deep research	`{topic}-deep-research`	`scrapling-deep-research`

All names: kebab-case, lowercase, no special characters.

Check for existing notes with same name before creating. If exists, append

-2

or ask user.

类型	命名模式	示例
X推文	从内容提取 `{主题}-{描述符}`	`scrapling-undetectable-web-scraping`
X文章	`{作者}-x-article-{日期}`	`irabukht-x-article-2026-02-23`
GitHub仓库	`{仓库名称}`	`scrapling` 或 `huggingface-skills-agent-plugins`
网页	从标题提取 `{主题}-{描述符}`	`kubernetes-practical-learning-path`
零散想法	`{概念}-{描述符}`	`agent-sort-through-the-slop`
深度调研	`{主题}-deep-research`	`scrapling-deep-research`

所有名称使用小写kebab-case格式，不含特殊字符。

创建前检查是否已有同名笔记，如果存在，追加

-2

后缀或询问用户处理方式。

Note Structure

笔记结构

For X tweets / web pages / GitHub repos (quick captures):

bash

notesmd-cli create "NOTE_NAME" --content "---
tags: [TYPE_TAG]
source: SOURCE_URL
author: AUTHOR
date: DATE
---

X推文/网页/GitHub仓库（快速捕获类）：

bash

notesmd-cli create "NOTE_NAME" --content "---
tags: [TYPE_TAG]
source: SOURCE_URL
author: AUTHOR
date: DATE
---

TITLE

Key Points

核心要点

Point 1
Point 2
Point 3

要点1
要点2
要点3

Summary

摘要

Brief paragraph summarizing the content.

简要概括内容的段落。

Source

来源

Original"


Type tags: `tweet` for X, `github` for GitHub, `web` for web pages, `idea` for ideas.

**For deep research (ideas):**

The res-deep skill produces its own structured output. Create the note with that output as body, adding frontmatter:

```bash
notesmd-cli create "NOTE_NAME" --content "---
tags: [idea, research]
date: DATE
---

{res-deep output here}"

Original"


类型标签：X类内容用`tweet`，GitHub类用`github`，网页用`web`，想法用`idea`。

**深度调研（想法类）：**

res-deep技能会生成自包含的结构化输出。创建笔记时使用该输出作为正文，添加前置元数据：

```bash
notesmd-cli create "NOTE_NAME" --content "---
tags: [idea, research]
date: DATE
---

{res-deep output here}"

Step 4: Update Daily Note

步骤4：更新每日笔记

For each processed item, replace the raw text in the daily note with a wikilink.

将每个已处理条目的原始文本替换为wikilink。

Wikilink Format by Section

不同区块的wikilink格式

## Links section (URLs from bookmarks/saves):

- [[note-name]] — @author: summary with key metrics (stars, likes, etc.)

## Notes section (ideas and thoughts):

- [[note-name]] — Brief: what the idea/research covers

## Log section (activity entries):

- [[note-name]] — Summary of what was captured

## Links区块（书签/收藏的URL）：

- [[note-name]] — @author: 包含核心指标（星数、点赞数等）的摘要

## Notes区块（想法和思考）：

- [[note-name]] — 简介：该想法/调研覆盖的内容

## Log区块（活动记录）：

- [[note-name]] — 捕获内容的摘要

Edit Procedure

编辑流程

Read the daily note:
```
notesmd-cli print "$DATE"
```

Resolve vault path:

VAULT=$(notesmd-cli print-default --path-only)

Use the Edit tool to replace each raw item with its wikilink line
Replace one item at a time to avoid Edit conflicts
Verify the final note by reading it again

读取每日笔记：
```
notesmd-cli print "$DATE"
```

解析知识库路径：

VAULT=$(notesmd-cli print-default --path-only)

使用编辑工具将每个原始条目替换为对应的wikilink行
每次替换一个条目，避免编辑冲突
再次读取最终笔记验证修改结果

Rules

规则

Preserve existing wikilinks — never modify already-processed lines
Keep section structure intact (## headers, empty lines between items)
If an item spans multiple lines (e.g., a paragraph idea), replace all lines with one wikilink line
The wikilink summary should be concise (under 120 chars) but include key metrics when available

保留已有wikilinks — 永远不要修改已处理的行
保持区块结构完整（##标题、条目之间的空行）
如果条目跨多行（例如段落形式的想法），将所有行替换为一行wikilink
wikilink摘要应简洁（不超过120字符），如有核心指标需包含在内

Step 5: Re-index & Report

步骤5：重新索引与报告

Re-index Vault

重新索引知识库

bash

qmd update && qmd embed

bash

qmd update && qmd embed

Summary Report

摘要报告

Present a summary table:

undefined

展示摘要表格：

undefined

Digest Complete: {DATE}

摘要处理完成: {DATE}

#	Type	Note Created	Status
1	X tweet	[[note-name]]	Created
2	Loose idea	[[note-name]]	Created (res-deep quick)
3	GitHub	[[note-name]]	Created
4	Web URL	—	Failed (403)

Notes created: 3 Items skipped: 2 (already processed) Items failed: 1 Vault re-indexed: Yes

undefined

#	类型	已创建笔记	状态
1	X tweet	[[note-name]]	已创建
2	零散想法	[[note-name]]	已创建（res-deep快速模式）
3	GitHub	[[note-name]]	已创建
4	网页URL	—	处理失败（403）

已创建笔记：3条跳过条目：2条（已处理）处理失败：1条知识库已重新索引：是

undefined

Modes

运行模式

Full (default)

全量模式（默认）

Process all unprocessed items in the daily note.

"Process my daily note" / "Daily digest"

处理每日笔记中所有未处理的条目。

"处理我的每日笔记" / "每日摘要"

Selective

选择模式

Process only specific items or sections.

"Process only the links in today's daily note" "Digest just the X URLs"

仅处理特定条目或区块。

"仅处理今日每日笔记中的链接" "只消化X URL"

Date Override

指定日期模式

Process a specific date's daily note.

"Process yesterday's daily note" "Digest 2026-02-20"

处理指定日期的每日笔记。

"处理昨天的每日笔记" "消化2026-02-20的笔记"

Dry Run

试运行模式

Classify and show the table (Step 1) without processing.

"What's unprocessed in my daily note?" "Show me what needs digesting"

仅执行分类并展示表格（步骤1），不实际处理。

"我的每日笔记里还有什么未处理的？" "告诉我需要消化的内容"

Constraints

约束

DO:

Always run Step 0 (vault path + daily note + xAI check) first
Present classification table and wait for user approval before processing
Process items in parallel where independent (multiple WebFetch calls, multiple X URLs in one batch)
Check for existing notes before creating to avoid duplicates
Read the daily note before editing — never guess content
Resolve vault path dynamically via
```
notesmd-cli print-default --path-only
```

DON'T:

Process items the user excluded from the classification table
Modify already-processed wikilink lines
Hardcode vault paths
Skip the classification approval step
Run res-deep at default/deep depth unless user explicitly requests it — use quick for daily digest
Create notes without frontmatter

必须做：

永远先执行步骤0（知识库路径 + 每日笔记 + xAI检查）
处理前展示分类表格，等待用户批准
独立任务尽量并行处理（多个WebFetch调用、单批次处理多个X URL）
创建笔记前检查是否存在同名文件，避免重复
编辑前读取每日笔记 — 永远不要猜测内容
通过
```
notesmd-cli print-default --path-only
```
动态解析知识库路径

禁止做：

处理用户在分类表格中排除的条目
修改已处理的wikilink行
硬编码知识库路径
跳过分类确认步骤
除非用户明确要求，否则不要使用默认/深度层级的res-deep调研 — 每日摘要默认使用快速模式
创建没有前置元数据的笔记