Skills: Domain Modeling

Skills: Domain Modeling

The /domain-modeling skill is a meta-skill for designing new domain-specific skills. Use it when you want to track a new type of information systematically.

Overview

When you find yourself repeatedly tracking a certain kind of information, it's time to create a dedicated skill. This skill guides you through:

1. Defining the entity types for your domain
2. Designing the TypeDB schema
3. Writing the SKILL.md documentation
4. Creating transaction scripts

When to Create a New Skill

You should create a new skill when:

• You're tracking the same type of information repeatedly
• The information has consistent structure
• You want to query across instances
• You need specialized analysis

Examples:

• Grant applications you're writing
• Conference talks you're preparing
• Research projects you're managing
• Datasets you're curating

The Curation Pattern

All skills should follow the curation workflow:

FORAGING → INGESTION → SENSEMAKING → ANALYSIS → REPORTING

When designing a new skill, consider each stage:

Stage	Questions to Answer
Foraging	Where does the information come from? URLs? APIs? Manual entry?
Ingestion	What's the raw form? How do you capture it?
Sensemaking	What entities should Claude extract? What relationships?
Analysis	What questions do you want to answer across instances?
Reporting	What views or dashboards would be useful?

Example: Conference Talks Skill

Let's design a skill for managing conference talks.

Step 1: Define the Domain

You: I want to track conference talks I'm giving. For each talk I need:
     - The conference name, date, location
     - Talk title and abstract
     - Submission deadlines and status
     - Slides and materials
     - Audience feedback

Claude: Let me help you design a schema for this...

Step 2: Identify Entities

Entity	Description	Attributes
conference	A conference event	name, date, location, url
talk	A talk submission	title, abstract, status, format
deadline	Important dates	type (submit, camera-ready, etc.), date
material	Associated files	type (slides, paper, demo), url
feedback	Audience responses	score, comments, source

Step 3: Define Relationships

Relationship	Description
at-conference	Links talk to conference
has-deadline	Links conference/talk to deadline
has-material	Links talk to materials
has-feedback	Links talk to feedback

Step 4: Design Workflow

Foraging:

• Conference CFPs from email, Twitter, websites
• Add by URL or manual entry

Ingestion:

• Script fetches CFP details
• Stores raw CFP as artifact

Sensemaking:

• Claude extracts deadlines, topics, requirements
• You add your talk details

Analysis:

• What talks are due this month?
• What's my acceptance rate?
• Which topics get best feedback?

Reporting:

• Calendar view of deadlines
• Status pipeline (submitted → accepted → presented)
• Feedback trends

Skill File Structure

A complete skill has:

.claude/skills/conference-talks/
├── SKILL.md              # Instructions for Claude
├── conference_talks.py   # TypeDB transaction scripts
└── README.md            # Developer documentation

SKILL.md Template

# Conference Talks Skill

Use this skill to manage conference talks you're preparing and giving.

## Commands

### Add a Conference
You: Add NeurIPS 2025 to my conference list
Claude: [Creates conference entity, fetches deadlines if URL provided]

### Submit a Talk
You: I'm submitting "AI for Science" to NeurIPS 2025
Claude: [Creates talk, links to conference, notes deadline]

### Update Status
You: My NeurIPS talk was accepted!
Claude: [Updates status, creates preparation timeline]

## Status Values
- idea - Not yet submitted
- submitted - Awaiting decision
- accepted - Will present
- rejected - Not accepted
- presented - Completed

Best Practices

Separation of Concerns

• Scripts handle I/O: API calls, database transactions, file operations
• Claude handles thinking: Extraction, analysis, recommendations

Consistent Naming

• Entity types: domain-entity (e.g., conference-talk)
• Relations: verb-noun (e.g., at-conference)
• Attributes: noun or adjective-noun (e.g., status, submission-deadline)

Provenance

Always link back to sources:

• Artifacts store raw content
• Fragments store extracted portions
• Notes store your annotations

Progressive Enhancement

Start simple, add complexity as needed:

1. Basic entities and relationships
2. Status tracking
3. Analysis queries
4. Dashboard views

Getting Help

You: I want to create a new skill for tracking [domain]

Claude: [Uses /domain-modeling to guide you through the design process]

Claude will help you:

• Identify entities and relationships
• Design the schema
• Write SKILL.md
• Create initial scripts