COG URL Dump Skill

Purpose

Transform raw URLs into structured, insightful knowledge entries through intelligent content extraction, categorization, and integration with the user's knowledge base. Quick capture with automatic insight generation.

When to Invoke

• User shares a URL they want to save
• User says "save this link", "bookmark this", "url dump", or "save for later"
• User pastes a URL and wants to capture it
• User wants to organize web resources into their knowledge base

Agent Mode Awareness

Check agent_mode in 00-inbox/MY-PROFILE.md frontmatter:

• If agent_mode: team — delegate content extraction, analysis, and categorization to a sub-agent while handling user interaction directly. The sub-agent fetches URL content, generates insights, and returns structured results for filing.
• If agent_mode: solo (default) — handle everything directly in the conversation. No delegation.

Pre-Flight Check

Before executing, check for user profile:

1. Look for 00-inbox/MY-PROFILE.md in the vault
2. If NOT found:

   Welcome to COG! It looks like this is your first time.
   
   Before we start, let's quickly set up your profile (takes 2 minutes).
   
   Would you like to run onboarding first, or should I proceed with default settings?
   

3. If found:
   • Read the profile to get user's interests and projects
   • Use interests to help with auto-categorization
   • Check for existing booklet categories in 05-knowledge/booklets/

Process Flow

1. User Interaction & Input Collection

• Accept URL(s) from the user (single URL or batch)
• Optionally accept user's quick note about why they're saving this
• Accept any format: bare URL, markdown link, or with notes

Prompt:

What URL(s) would you like to save?
(You can paste one or more URLs, optionally with a note about why you're saving it)

2. URL Validation & Fetch

• Validate URL format
• Check if URL is accessible
• Detect duplicate URLs in existing knowledge base
• Fetch the web page content

Content Extraction

Extract from the page:

• Page Title: [extracted-title]
• Meta Description: [if available]
• Author: [if detected]
• Published Date: [if detected]
• Word Count: [estimated]
• Read Time: [X minutes]
• Main Content: [extracted body text]
• Key Headings: [list of H1/H2s]

3. Category Selection

Default Categories:

• Articles & Blogs: Long-form content, tutorials, opinion pieces
• Tools & Resources: Software, utilities, services, APIs
• Reference: Documentation, specs, standards
• Research: Papers, studies, academic content
• Inspiration: Design, ideas, creative references
• Videos & Media: YouTube, podcasts, multimedia
• News & Updates: Industry news, announcements
• Project-Specific: Related to a specific project (offer project list from MY-PROFILE.md)
• To Review: Unsure, save for later categorization

Custom Categories:

• Check 05-knowledge/booklets/ for existing custom categories
• Offer to create new category if needed

Auto-suggestion: Based on content analysis, suggest the most likely category but let user confirm or change.

4. Content Analysis and Processing

Phase 1: Content Classification

Determine:

• Content Category: [article|tool|reference|research|video|news|etc]
• Primary Topics: [topic1, topic2, topic3]
• Tone: [informative|opinion|tutorial|news|etc]
• Quality Assessment: [high|medium|low]
• Credibility Indicators: [author credentials, citations, etc]

Phase 2: Insight Extraction

Generate:

• Executive Summary: [2-3 sentences]
• Key Insights:
  1. [Insight 1 with context]
  2. [Insight 2 with context]
  3. [Insight 3 with context]
• Notable Quotes: [if any stand out]
• Action Items: [practical takeaways]

Phase 3: Relevance Assessment

Analyze:

• User Interest Match: [high|medium|low] - [which interests from profile]
• Project Relevance: [project-name] - [why relevant]
• Knowledge Gap: [yes|no] - [what gap it fills]
• Timeliness: [evergreen|current|dated]
• Uniqueness: [novel|common|duplicate-adjacent]

Phase 4: Cross-Reference

Identify connections to:

• Related Bookmarks: [existing similar saves]
• Related Braindumps: [if content connects]
• Related Projects: [if applicable]
• Suggested Tags: [tag1, tag2, tag3]

5. Generate Structured Output

Create bookmark file with this structure:

---
type: "url-bookmark"
category: "[category-name]"
domain: "[source-domain.com]"
date_saved: "YYYY-MM-DD"
date_accessed: "YYYY-MM-DD HH:MM"
url: "[original-url]"
title: "[page-title]"
author: "[author-if-available]"
published: "[publish-date-if-available]"
tags: ["#bookmark", "#category-tag", "#topic-tags"]
relevance: "[high|medium|low]"
status: "unread"
related_projects: ["project1", "project2"]
confidence: "[high|medium|low]"
---

# [Title]

## Quick Summary
[2-3 sentence summary of the content]

## Key Insights
- **Insight 1:** [description with context]
- **Insight 2:** [description with context]
- **Insight 3:** [description with context]

## Why This Matters
[Connection to user's interests/projects. What makes this worth saving?]

## User Note
[Original user note if provided, otherwise omit section]

## Content Highlights
[Key excerpts or quotes from the content - 200-400 words max]

## Practical Takeaways
- [ ] [Action item 1 if applicable] 📅 [YYYY-MM-DD = date +1 week from today]
- [ ] [Action item 2 if applicable] 📅 [YYYY-MM-DD = date +1 week from today]

## Related Knowledge
- **Similar Bookmarks:** [[bookmark1]], [[bookmark2]]
- **Connected Projects:** [[project1]]
- **Related Notes:** [[note1]], [[note2]]

## Source Details
| Field | Value |
|-------|-------|
| Domain | [domain] |
| Author | [author or "Unknown"] |
| Published | [date or "Unknown"] |
| Word Count | [~X words] |
| Read Time | [~X minutes] |

## Processing Notes
- **Extracted:** [timestamp]
- **Category Confidence:** [percentage]
- **Review Needed:** [yes|no] - [reason if yes]

---

*Processed by COG URL Curator*

Save to appropriate location:

• Standard: 05-knowledge/booklets/[category-slug]/[title-slug]-YYYY-MM-DD.md
• Project-specific: 04-projects/[project-slug]/resources/[title-slug]-YYYY-MM-DD.md
• Mixed/Unclear: 00-inbox/url-[title-slug]-YYYY-MM-DD.md

6. Tool/Resource Special Handling

For tools and software, use enhanced template:

---
type: "url-tool"
category: "tools"
domain: "[domain]"
url: "[url]"
title: "[tool-name]"
date_saved: "YYYY-MM-DD"
pricing: "[free|freemium|paid|enterprise]"
tags: ["#tool", "#category-tags"]
status: "to-evaluate"
---

# [Tool Name]

## What It Does
[1-2 sentence description]

## Key Features
- Feature 1
- Feature 2
- Feature 3

## Use Cases
- Use case 1
- Use case 2

## Pricing
[Pricing details if available]

## Why It's Relevant
[Connection to user's work/interests]

## Evaluation Status
- [ ] Sign up / try demo 📅 [YYYY-MM-DD = date +3 days from today]
- [ ] Test key features 📅 [YYYY-MM-DD = date +1 week from today]
- [ ] Compare with alternatives 📅 [YYYY-MM-DD = date +1 week from today]
- [ ] Decision: [use|pass|revisit] 📅 [YYYY-MM-DD = date +2 weeks from today]

## Notes
[Space for user's evaluation notes]

---

*Processed by COG URL Curator*

7. Batch Processing

For multiple URLs:

Processing [X] URLs...

1. [URL 1] → [category] → Saved to [path]
2. [URL 2] → [category] → Saved to [path]
3. [URL 3] → [category] → Saved to [path]

Summary:
- Articles: 2 saved
- Tools: 1 saved
- Total: 3 URLs processed

8. Confirm Completion

• Confirm file(s) created
• Show user: "URL saved to [file path]"
• Show quick summary: title, category, key insight preview
• Ask if they want to:
  • Add another URL
  • Deep-dive into the content
  • Connect to specific project or braindump

Booklet Structure

URLs are organized into "booklets" (category folders):

05-knowledge/
└── booklets/
    ├── articles/
    │   ├── _index.md (category overview - auto-created)
    │   └── [article-entries].md
    ├── tools/
    │   ├── _index.md
    │   └── [tool-entries].md
    ├── reference/
    │   ├── _index.md
    │   └── [reference-entries].md
    ├── research/
    │   ├── _index.md
    │   └── [research-entries].md
    ├── inspiration/
    │   ├── _index.md
    │   └── [inspiration-entries].md
    ├── videos/
    │   ├── _index.md
    │   └── [video-entries].md
    └── [custom-category]/
        ├── _index.md
        └── [entries].md

Category Index Template

When creating a new category, also create an index file:

---
type: "booklet-index"
category: "[category-name]"
created: "YYYY-MM-DD"
last_updated: "YYYY-MM-DD"
entry_count: 0
---

# [Category Name] Booklet

## Description
[What this category contains]

## Recent Additions
[Auto-updated list - most recent 10 entries]

## Top Entries
[Manually curated or most-accessed entries]

## Tags in This Category
[List of common tags used]

## Related Categories
- [[other-category-1]]
- [[other-category-2]]

YAML Formatting Requirements

CRITICAL: All YAML frontmatter must use proper Obsidian-compatible formatting:

• All string values MUST be quoted with double quotes
• Arrays MUST use quoted strings: ["item1", "item2", "item3"]
• URLs MUST be quoted to handle special characters
• Boolean values should NOT be quoted: true or false
• Ensure proper YAML syntax to prevent parsing errors in Obsidian

Examples:

# CORRECT
type: "url-bookmark"
url: "https://example.com/path?query=value"
tags: ["#bookmark", "#article", "#ai"]
relevance: "high"
reviewed: false

# INCORRECT
type: url-bookmark
url: https://example.com/path?query=value
tags: [#bookmark, #article, #ai]
relevance: high
reviewed: "false"

Verification Protocols

Content Accuracy

• Title Verification: Ensure extracted title matches page
• Author Attribution: Verify author if stated
• Date Accuracy: Confirm publication date if shown
• Summary Fidelity: Ensure summary accurately represents content

Categorization Verification

• Category Fit: Confirm content matches selected category
• Tag Relevance: Verify tags accurately describe content
• Interest Alignment: Confirm relevance assessment is accurate
• Project Connection: Verify project relevance if claimed

Quality Checks

• Completeness: All required fields populated
• Formatting: Proper markdown and YAML syntax
• Links: All internal links valid
• Metadata: Frontmatter properly formatted

Uncertainty Handling

When Content is Unclear

• Paywalled Content: Note limitation, extract available preview
• Dynamic Content: Note if content may change
• Complex Content: Flag for manual review if needed
• Non-English: Note language, provide translation if possible

Confidence Indicators

• High Confidence (90%+): Clear content with obvious categorization
• Medium Confidence (70-89%): Generally clear with some ambiguity
• Low Confidence (50-69%): Significant ambiguity requiring user input
• Very Low Confidence (<50%): Major uncertainty, save to inbox

Always explicitly state confidence levels and reasoning in processing notes.

Loop Engineering

URL capture is a fetch-retry loop with a quality gate, not a single fetch-and-file. See .claude/skills/loop-engineering/SKILL.md for the shared vocabulary.

The loop (per URL): fetch → if the fetch fails or returns an empty/blocked body, retry a different way (https vs http, reader mode, an archive snapshot) → once content is present, run the quality gate → file it, or escalate to the user / save to inbox with a Review Needed flag.

The verifier (deterministic):

• Fetch returned a non-empty body (not a paywall stub or error page).
• Required fields are populated: title, at least one key insight, a category.
• YAML frontmatter is valid (see YAML Formatting Requirements).
• Category confidence clears the threshold. Below ~70%, the loop does not silently guess.

Termination conditions (layered):

• Goal met: content extracted and the quality gate passes → save to the category folder.
• Retry cap: stop after ~3 fetch attempts → save what was extracted with a low-confidence flag (see Uncertainty Handling), do not invent missing fields.
• Hard stop on paywall / login wall: note the limitation, capture the available preview, do not loop forever.
• Human escalation: confidence below threshold → present the best guess and ask the user to confirm category, rather than filing it wrong.

Patterns: reflect-retry (each failed fetch picks a different method) + evaluator (the quality gate) + human-in-the-loop (low-confidence escalation).

In-loop context: once insights and metadata are extracted, drop the raw page body. For batch input, process each URL as its own independent loop so one bad URL never stalls the rest.

Integration with Other Skills

Immediate Follow-up

After URL capture, suggest:

• /braindump - Capture thoughts about the URL
• /knowledge-consolidation - Integrate into knowledge frameworks
• Daily brief will surface relevant saved URLs

Cross-Referencing

Automatically check for connections to:

• Active projects (from MY-PROFILE.md)
• Recent braindumps
• Competitive watchlist companies (if exists)
• User interests

Success Metrics

• Speed of capture (< 30 seconds for single URL)
• Accurate categorization with user confirmation
• Useful insight extraction
• Proper integration with existing knowledge
• Easy retrieval and discovery later
• High confidence in extractions

Learning and Adaptation

Pattern Learning

• Track which bookmarks get revisited
• Learn user's categorization preferences
• Improve relevance scoring based on engagement
• Refine insight extraction based on what user finds useful

Continuous Improvement

• Monitor categorization accuracy over time
• Adapt to user's preferred tag taxonomy
• Learn domain-specific terminology
• Improve cross-referencing accuracy