ReadItLater Processor — Clawdbot Skill

Automatically process, summarize, and organize your Obsidian ReadItLater inbox. Turns a pile of saved links into actionable knowledge.

What It Does

When triggered (via heartbeat or manual), this skill:

•Scans your Obsidian ReadItLater inbox folder for unprocessed items
•Reads each item and classifies it (tweet, article, GitHub repo, video, note, other)
•Summarizes the content with key takeaways
•Tags items with relevant Obsidian tags (#tool, #ai, #nixos, #dev, etc.)
•Rates relevance (🔥 must-read, ⭐ interesting, 📝 reference, ⏭️ skip)
•Marks processed with #agent-work-done tag
•Generates a digest — a single markdown file with all processed items summarized

Configuration

Set these in your HEARTBEAT.md or pass as context:

yaml

readitlater:
  inbox_path: "/path/to/vault/ReadItLater Inbox/"
  output_path: "/path/to/vault/ReadItLater Digests/"
  batch_size: 20                    # items per run
  skip_patterns:                    # skip these by filename
    - "Youtube -"                   # YouTube saves (can't extract content)
    - "Article 20"                  # Instagram/dead links
  tag_processed: "#agent-work-done" # tag added to processed items
  relevance_filter: "dev"           # focus: dev, ai, general, all

Usage

In HEARTBEAT.md

markdown

## Task: Process ReadItLater
Process unprocessed items from Obsidian ReadItLater inbox.
- Path: /home/user/vault/ReadItLater Inbox/
- Batch: 20 items per cycle
- Focus on: dev tools, AI, NixOS content
- Skip: YouTube saves, Instagram links
- Tag processed items with #agent-work-done
- Write digest to memory/YYYY-MM-DD.md

Manual Trigger

Just tell your Clawdbot: "Process my ReadItLater inbox" and it will follow this skill.

Processing Logic

Item Classification

Type	Detection	Processing
Tweet	`[[Tweet]]` tag or twitter.com URL	Extract author, quote, key claim
GitHub Repo	`github.com` URL, `[[Article]]`	Stars, language, what it does, install method
Article	`[[Article]]` tag	Title, source, TL;DR (3 sentences max)
Note/Voice	`[[Textsnippet]]` tag	Transcription context, action items
Video	YouTube URL	Note as unwatchable from CLI, skip or extract title
Dead Link	Empty content or error page	Mark as dead, skip

Relevance Scoring

•🔥 Must-read: Directly useful (tools to install, techniques to apply, money opportunities)
•⭐ Interesting: Worth knowing about (trends, new projects, ecosystem news)
•📝 Reference: Save for later lookup (documentation, tutorials, configs)
•⏭️ Skip: Not relevant (entertainment, broken links, off-topic)

Output: Digest Format

markdown

# ReadItLater Digest — YYYY-MM-DD

## 🔥 Must-Read
### [Tool Name](url)
**Type**: GitHub Repo | **Stars**: 5.2k | **Language**: Rust
One-line summary of what it does and why you care.
**Action**: `cargo install tool-name`

## ⭐ Interesting
### [Article Title](url)
**Source**: blog.example.com
TL;DR in 2-3 sentences.

## 📝 Reference
- [Config Guide](url) — How to set up X with Y
- [API Docs](url) — Reference for Z integration

## ⏭️ Skipped (12 items)
- 8 YouTube videos (can't process from CLI)
- 3 Instagram reels (no content extractable)
- 1 dead link

Integration with Other Skills

Night Shift

Add as a rotation task in the night-shift cycle. Processes 20 items per heartbeat, tracks progress.

Morning Brief

Feed the digest into a morning summary: "While you slept, I processed 20 saved items. Here are the 3 you should look at..."

QMD Search

After processing, items are indexed by qmd for fast semantic search across your entire vault.

Tips

•Batch size 20 is a sweet spot — enough to make progress, not so much it burns context
•Skip YouTube unless you have transcript extraction set up (Readeck can do this)
•Tag immediately after reading, even if summary is brief — prevents re-processing
•Voice notes (Textsnippet) often contain Lucas's TODO items — flag these as action items
•GitHub repos: check stars count and last commit date to gauge if worth installing

Requirements

•Obsidian vault with ReadItLater plugin (or any folder of markdown files)
•Clawdbot with file read/write access
•Optional: web_fetch for enriching summaries of linked content
•Optional: qmd for post-processing search indexing