domain-vocab

name: domain-vocab description: This skill should be used when the user asks to "learn domain concepts", "domain vocab", "learn [domain] vocabulary", "what are key concepts in [domain]", "explore domain", "divergent exploration", or wants to build conceptual vocabulary for AI prompting. Provides structured learning of domain terminology with token priming, divergent exploration, and depth progression. version: 0.3.0

Domain Vocabulary Learning

Build conceptual vocabulary for any domain to enhance AI prompting effectiveness. This skill uses token priming to activate domain-specific knowledge, divergent exploration for creative discovery, and depth progression from popular to expert-level terminology.

Core Principle: Token Priming

"Response quality improves dramatically when you inject tokens that experts actually use, rather than just adopting an expert persona."

The skill works by injecting precisely fitting tokens that prime the model's attention toward the target domain. Like a game where visiting a region unlocks its map, the right tokens unlock domain-specific knowledge.

When to Use

• Learning a new technical domain
• Preparing to work with domain experts
• Improving AI prompt quality for specific fields
• Exploring unfamiliar territories through divergent thinking
• Deepening from popular knowledge to expert-level understanding

Core Workflow (6 Phases)

Phase 0: Token Priming

Objective: Prime the context with authentic domain tokens before concept extraction.

Actions:

1. Summon key figures (3-5 people) associated with the domain
2. Extract their vocabulary - terms/concepts these figures would actually use
3. Sample recent literature - pull tokens from arXiv, papers, or authoritative sources
4. Prime the context - use these tokens to "lay the foundation" before proceeding

Example for "DevOps":

Key Figures: Gene Kim, Jez Humble, Nicole Forsgren
Their Tokens: "deployment frequency", "lead time for changes",
              "MTTR", "change failure rate", "DORA metrics"
Recent arXiv: "AIOps", "observability-driven development", "GitOps reconciliation"

Output: Context primed with authentic expert-level tokens

Phase 1: Domain Identification

Objective: Clarify the target domain, scope, and desired depth.

Actions:

1. Confirm the domain from user input
2. If domain is broad, ask about focus area
3. Determine user's current familiarity level
4. Ask about desired depth level (L1-L4)

Depth Levels:

Output: Clear domain scope, user level, and target depth

Phase 2: Core Concept Extraction

Objective: Identify 20-30 essential terms using primed tokens.

Actions:

1. Generate concepts informed by Phase 0 priming
2. Organize by difficulty: Entry (5-8), Intermediate (10-12), Advanced (5-8)
3. Include tokens that actual experts use, not just textbook definitions
4. Mark prerequisite relationships

Quality Check: Would a domain expert recognize these as "real" terms they use?

Output: Structured concept list with authentic terminology

Phase 3: Relationship Mapping

Objective: Connect concepts showing how they relate.

Relationship Types:

• prerequisite: A must be understood before B
• related: A and B frequently appear together
• hierarchy: A is a specific case of B
• alternative: A can be chosen instead of B

Output: Relationship data for each concept

Phase 4: Practical Context

Objective: Ground concepts in real-world usage.

For each concept, provide:

1. Where it appears: Situations where this term is encountered in practice
2. Problems without it: What goes wrong without this knowledge
3. Enabled capabilities: What becomes possible with this knowledge

Output: Practical context for each concept

Phase 5: Output Generation

Objective: Present information in user's preferred format.

Available Formats:

Standard Formats

• Card: Individual concept cards with full details
• Map: Mermaid diagram showing relationships
• Interactive: Progressive exploration through conversation

Exploration Formats

• Figure Summoning: A-Z figure exploration
• arXiv Traverse: Cross-paper insight discovery
• Acronym Association: Pareidolia-based divergent thinking
• Cross-Domain Collision: Collide tokens from different domains for emergent insights

See references/exploration-modes.md for detailed templates.

Exploration Modes

Mode A: Figure Summoning

Leverage the model's associative ability by summoning domain figures.

Process:

1. Think of A-Z as first letters of names in the domain
2. For each letter, summon a relevant figure
3. Extract concepts/tokens that figure would use
4. Find non-obvious connections between figures' ideas

Example:

A: Alan Kay → "messaging", "late binding", "Dynabook"
B: Barbara Liskov → "substitution principle", "CLU", "abstraction"
C: Christopher Alexander → "pattern language", "quality without a name"
...
→ Connection: Kay's "messaging" + Alexander's "patterns" → Design Patterns movement

Mode B: arXiv Traverse

Use actual paper abstracts to fill context with meaningful tokens.

Process:

1. Identify relevant arXiv category (cs.SE, cs.DC, etc.)
2. Sample 10 recent papers using ID pattern (e.g., 2601.XXXXX)
3. Read only abstracts to extract key tokens
4. Find "non-obvious insights" connecting different papers

Why it works: Fills context with tokens that real researchers actually use.

Mode C: Acronym Association

Use pareidolia (pattern recognition) for divergent exploration.

Process:

1. Generate random 4-letter combinations (e.g., MSTK, PRLD, CVBN)
2. Read each as an acronym related to the domain
3. Interpret what each "could mean" in context
4. Use interpretations to discover new perspectives

Example for DevOps:

MSTK → "Metrics-driven Service Toolkit"
PRLD → "Progressive Rollout with Latency Detection"
CVBN → "Container Validation Before Networking"

Depth Progression

Start broad (L1), narrow down based on user interest.

L1 → L2 Transition

• Replace Wikipedia-level terms with practitioner jargon
• Add tool-specific terminology
• Include "tribal knowledge" terms

L2 → L3 Transition

• Introduce paper-specific concepts
• Add measurement/metrics terminology
• Include methodology names (e.g., "Goodhart's Law" in metrics)

L3 → L4 Transition

• Summon specific researchers by name
• Use their coined terms and frameworks
• Reference specific papers or talks

Example Progression (Observability):

L1: "monitoring", "logging", "alerting"
L2: "golden signals", "SLI/SLO/SLA", "distributed tracing"
L3: "observability-driven development", "cardinality explosion", "exemplars"
L4: "Charity Majors' o11y philosophy", "Cindy Sridharan's distributed systems observability"

Interactive Commands

AI Prompting Integration

Each concept includes guidance on prompt improvement:

Before (without domain tokens):

"Set up server monitoring for me"

After (with L2 tokens):

"Set up Prometheus to collect golden signals (latency, traffic, errors, saturation) metrics and configure a Grafana dashboard"

After (with L3 tokens):

"Implement distributed tracing with OpenTelemetry and connect high-cardinality metrics to traces using exemplars"

Additional Resources

Reference Files

• references/learning-framework.md - Learning principles and methodology
• references/output-templates.md - Templates for each output format
• references/exploration-modes.md - Detailed exploration mode guides

Example Files

• examples/software-dev.md - Software development vocabulary (30 terms)

Workflow Summary

0. PRIME context with authentic expert tokens
1. IDENTIFY domain + scope + depth level
2. EXTRACT concepts using primed vocabulary
3. MAP relationships between concepts
4. CONTEXTUALIZE with practical usage
5. OUTPUT in preferred format (including exploration modes)

The goal is conceptual fluency - using domain terms accurately in AI prompts for dramatically better results. Token priming ensures responses come from the expert distribution, not the generic one.