doc-daily-digest（文档每日摘要）

Daily Digest

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

Prerequisites

Optional dependency: scrapling -- headless browser fallback for sites that block WebFetch (403, empty content, captcha pages). Install with: uv tool install 'scrapling[all]'

Workflow

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

Run all three checks:

# 1. Vault path
VAULT=$(notesmd-cli print-default --path-only)

# 2. Read daily note (today or user-specified date)
DATE=$(date '+%Y-%m-%d')
notesmd-cli print "$DATE"

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

Step 1: Scan & Classify

Parse the daily note and classify every item. Items live in these sections: ## Notes, ## Log, ## Links.

Classification Rules

Present Classification

Before processing, show the user a classification table:

## 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)

Step 2: Process Items

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

X/Twitter URLs

Requires xAI key (XAI_AVAILABLE=true).

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.

GitHub URLs

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:

1. scrapling extract get "URL" /tmp/scrapling-fallback.md → Read → validate content
2. If content is thin (JS-only shell, no data) → scrapling extract fetch "URL" /tmp/scrapling-fallback.md --network-idle --disable-resources → Read → validate
3. If still blocked → scrapling extract stealthy-fetch "URL" /tmp/scrapling-fallback.md --solve-cloudflare
4. All tiers fail → note failure and move on

Web URLs

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:

1. scrapling extract get "URL" /tmp/scrapling-fallback.md → Read → validate content
2. If content is thin (JS-only shell, no data) → scrapling extract fetch "URL" /tmp/scrapling-fallback.md --network-idle --disable-resources → Read → validate
3. If still blocked → scrapling extract stealthy-fetch "URL" /tmp/scrapling-fallback.md --solve-cloudflare
4. All tiers fail → note failure and move on

Loose Ideas

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.

Step 3: Create Notes

For each processed item, create an Obsidian note.

Note Naming

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

Check for existing notes with same name before creating. If exists, append -2 or ask user.

Note Structure

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

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

# TITLE

## Key Points

- Point 1
- Point 2
- Point 3

## Summary

Brief paragraph summarizing the content.

## Source

- [Original](SOURCE_URL)"

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:

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

{res-deep output here}"

Step 4: Update Daily Note

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

Wikilink Format by Section

## 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

Edit Procedure

1. Read the daily note: notesmd-cli print "$DATE"
2. Resolve vault path: VAULT=$(notesmd-cli print-default --path-only)
3. Use the Edit tool to replace each raw item with its wikilink line
4. Replace one item at a time to avoid Edit conflicts
5. Verify the final note by reading it again

Rules

• 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

Step 5: Re-index & Report

Re-index Vault

qmd update && qmd embed

Summary Report

Present a summary table:

## Digest Complete: {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

Modes

Full (default)

Process all unprocessed items in the daily note.

"Process my daily note" / "Daily digest"

Selective

Process only specific items or sections.

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

Date Override

Process a specific date's daily note.

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

Dry Run

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

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

Constraints

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