Create Pattern

{"url": "USER_PROVIDED_URL", "return_format": "markdown"}

PDF File

Use Read tool with the PDF path.

YouTube Video

Use mcp__web_reader__webReader with the YouTube URL (transcript extraction).

Codebase/Repository

Use Glob to find key files, then Read to analyze:

• **/*.md - Documentation
• **/*.py - Python implementation
• **/*.ts - TypeScript implementation
• README files

Pasted Text

Analyze directly (no tool needed).

Phase 2: Analyze & Extract

From the source content, extract:

1. Pattern Title - What is this pattern called?
2. Problem Statement - What challenge does it solve?
3. Solution - What's the core approach/mechanism?
4. Category - Which of 8 categories fits best?
5. Source URL - The original source
6. Author/Originator - Who created/originated this?
7. Tags - 3+ relevant keywords
8. Key Insights - Any novel techniques, constraints, trade-offs

Generate candidate metadata:

• Title: descriptive, concise
• Category: select from allowed categories (see reference below)
• Tags: generate 5-10 relevant tags
• Status: default to "emerging" unless evidence suggests otherwise

Phase 3: Match Against Existing Patterns

Search for similar patterns in patterns/ directory:

Step 3a: Read All Pattern Frontmatter

Use Glob to find all .md files in patterns/, then Grep with output_mode content to extract frontmatter from each:

grep -A 10 "^title:" patterns/*.md

Or read multiple pattern files in parallel to compare:

• Problem statements
• Solution approaches
• Categories
• Tags
• Sources

Step 3b: Semantic Comparison

For each existing pattern, assess similarity using these signals:

Primary Signals (weight: 30% each)

1. Problem Statement - Same fundamental challenge?
2. Solution Mechanism - Same core approach/technique?
3. Category - Same category?

Secondary Signals (weight: 10% each) 4. Tag Overlap - Shared keywords? 5. Source/Author - Same contributor's follow-up work? 6. Title - Similar naming/description?

Step 3c: Calculate Confidence Score

Score = Σ(matching_signals × weights)

• All 3 primary match → 90%+ confidence
• 2 primary + 1+ secondary → 70-90% confidence
• 1 primary + 2+ secondary → 50-70% confidence
• 0-1 primary only → <50% confidence

Step 3d: Identify Top Matches

List top 3 matching patterns with:

• Pattern title and file path
• Confidence score
• Matching rationale (which signals matched)

Phase 4: Decision

Based on confidence score:

High Confidence (>80%) → Update Existing

Update the top-matching existing pattern with new source.

Low Confidence (<50%) → Create New

Create a brand new pattern file.

Medium Confidence (50-80%) → Ask User

Present the top match and ask:

This source seems related to:
- [Match Pattern] (X% confidence)
  - Similar problem: [problem summary]
  - Similar solution: [solution summary]

Should I:
1. Create a new pattern
2. Update the existing pattern with this new source

Phase 5a: Create New Pattern

Step 5a-1: Generate Metadata

---