name: domain-researcher
description: Comprehensive methodology for researching any domain to build expert AI agents. Use when designing agents and need to understand best practices, novel approaches, tool ecosystems, and ethical guidelines for a specialty. Systematic research workflow with validation criteria.

Domain Researcher

Systematic methodology for researching any domain to build AI agents with genuine expertise.

Purpose

Research any specialty deeply enough to design agents that demonstrate true domain expertise, not surface-level knowledge. Combines current best practices (2024-2025) with novel but stable techniques.

When to Use

Use this skill when:
- Designing an agent for an unfamiliar domain
- Need to validate best practices for a specialty
- Researching tool ecosystems and methodologies
- Identifying domain-specific ethical requirements
- Building knowledge base for agent specification

Pairs with: agent-builder skill (use this to research, agent-builder to design)

Research Workflow

Step 1: Domain Mapping

Extract Keywords from Requirements

Identify core concepts, tools, methodologies, and terminology:
- Main domain (e.g., "legal contract analysis")
- Sub-domains (e.g., "risk assessment", "clause extraction")
- Standard tools (e.g., "PDF parsing", "NLP libraries")
- Industry terminology (e.g., "force majeure", "indemnification")

Create Research Query Matrix

Build search queries across multiple dimensions:

[Domain] + best practices 2025
[Domain] + AI agent implementation
[Domain] + workflow automation
[Domain] + tool ecosystem
[Domain] + ethical guidelines
[Sub-domain] + specialized techniques
[Tool/Framework] + integration patterns

Step 2: Best Practices Research

Search Current Methodologies

Use WebSearch with date-specific queries:
- Priority: 2024-2025 sources (most current)
- Acceptable: 2022-2023 (stable practices)
- Caution: Pre-2022 (may be outdated)

Look For:
- Industry standards and certifications
- Professional organization guidelines
- Academic research papers
- Technical documentation from authoritative sources
- Case studies from successful implementations

Validate Through Multiple Sources

A practice is "best practice" when:
- ✅ Mentioned in 3+ independent authoritative sources
- ✅ Recommended by industry leaders or standards bodies
- ✅ Demonstrated in real-world implementations
- ✅ Updated within last 2-3 years

Example: Financial Data Analysis Agent

Search queries:

"financial data analysis best practices 2025"
"AI agents for financial analysis implementation"
"financial data processing workflows automation"
"financial analysis Python libraries ecosystem"

Findings:
- Best practices: SOX compliance, audit trails, source citation
- Tools: pandas, numpy, statsmodels, plotly
- Standards: GAAP, IFRS awareness

Step 3: Novel Approaches Research

Identify Innovative Techniques

Search for cutting-edge but production-ready approaches:
- "latest advances in [domain] AI"
- "[domain] innovative techniques 2024"
- "emerging tools for [domain] automation"

Stability Criteria

A novel approach is "stable" when:
- ✅ Has production deployments (not just research papers)
- ✅ Supported by maintained open-source libraries
- ✅ Documented performance benchmarks
- ✅ Clear migration path from traditional methods
- ✅ Community adoption (GitHub stars, Stack Overflow discussions)

Balance Innovation with Reliability

✅ GOOD: "Use transformer-based NER models for entity extraction"
   (Novel: transformers, Stable: widely adopted, libraries mature)

⚠️ CAUTION: "Use experimental quantum NLP algorithm"
   (Novel: yes, Stable: no - research-only, no libraries)

✅ GOOD: "Implement RAG with vector databases for proprietary knowledge"
   (Novel: RAG pattern, Stable: proven in production, tools available)

Step 4: Tool Ecosystem Analysis

Identify Standard Tools

Research the dominant tools, libraries, and frameworks for the domain:
- Programming languages (Python, JavaScript, etc.)
- Core libraries (data processing, ML, visualization)
- APIs and services (external integrations)
- File format handlers (PDF, Excel, JSON, etc.)
- Data sources (databases, APIs, web scraping)

Evaluation Criteria

Tools should be:
- Actively maintained - Recent commits, responsive maintainers
- Well-documented - Clear API docs, examples, tutorials
- Widely adopted - Large user base, community support
- Compatible - Works with modern Python/Node versions
- Licensed appropriately - Open source or commercially viable

Tool Integration Patterns

Document how tools combine:

Example: PDF Contract Analysis Agent

Data Flow:
1. PDF Ingestion → PyPDF2 or pdfplumber
2. Text Extraction → pdfminer.six
3. Entity Recognition → spaCy or transformers
4. Clause Detection → Custom regex + NLP models
5. Risk Analysis → Domain-specific rules + LLM
6. Report Generation → jinja2 templates + markdown

Step 5: Ethical Guidelines Research

Domain-Specific Ethics

Every domain has ethical considerations. Research:
- Professional codes of ethics (medical, legal, financial)
- Regulatory requirements (GDPR, HIPAA, SOX)
- Industry standards for fairness and bias
- Privacy expectations and data handling

Search Queries:

"[domain] regulatory requirements"
"[domain] compliance standards"
"[domain] fairness guidelines"
"[domain] data privacy requirements"

Document Ethical Constraints

Create a list of:
- Required compliance: Mandatory regulatory requirements
- Industry standards: Professional association guidelines
- Bias risks: Known problematic patterns to avoid
- Privacy boundaries: Data handling restrictions

Example: Healthcare Diagnostic Agent

Ethics research findings:

Required Compliance:
- HIPAA for patient data
- FDA guidelines for clinical decision support

Industry Standards:
- AMA ethics guidelines
- Clear disclaimer: "Not medical advice, consult physician"
- Explain diagnostic reasoning (transparency)
- Cite medical literature sources

Bias Risks:
- Training data may underrepresent certain demographics
- Historical medical bias in treatment recommendations
- Language/cultural barriers in symptom description

Privacy:
- No storage of identifiable patient information
- Encrypted data transmission
- Audit logs for all data access

Step 6: Knowledge Compilation

Synthesize Research into Structured Knowledge

Organize findings:

1. Core Concepts
- Domain fundamentals the agent must understand
- Industry terminology and definitions
- Key processes and workflows

2. Methodologies
- Standard approaches for common tasks
- Best practices from research
- Novel techniques to incorporate

3. Tool Specifications
- Required libraries and versions
- API endpoints and authentication
- Data format schemas

4. Ethical Requirements
- Regulatory compliance (on top of Coalition base ethics)
- Domain-specific standards
- Privacy requirements

5. Edge Cases
- Known challenges in the domain
- Error handling strategies
- Escalation criteria

Research Output Template

Use this structure to document findings:

# [Domain] Agent - Research Summary

## Domain Overview
[Brief description of the domain and agent purpose]

## Best Practices Identified
1. [Practice 1] - Source: [Citation]
2. [Practice 2] - Source: [Citation]
...

## Novel Approaches (Stable)
1. [Approach 1] - Adoption: [Evidence] - Tools: [Libraries]
2. [Approach 2] - Adoption: [Evidence] - Tools: [Libraries]
...

## Tool Ecosystem
### Core Libraries
- [Library 1] ([version]) - Purpose: [What it does]
- [Library 2] ([version]) - Purpose: [What it does]

### External Services/APIs
- [Service 1] - Use case: [When to use]
- [Service 2] - Use case: [When to use]

### Data Sources
- [Source 1] - Format: [Type] - Access: [How to access]

## Domain-Specific Compliance
### Regulatory Requirements
- [Regulation 1] - Implication: [What it means for agent]
- [Regulation 2] - Implication: [What it means for agent]

### Industry Standards
- [Standard 1] - How to implement: [Approach]

### Privacy Requirements
- [Requirement 1] - Implementation: [How to meet it]

## Edge Cases & Limitations
- [Edge case 1] - Handling: [Strategy]
- [Limitation 1] - Mitigation: [Approach]

## Implementation Recommendations
[Synthesized recommendations based on all research]

Quality Checklist

Before proceeding to agent design, verify:

• [ ] Researched from 3+ authoritative sources
• [ ] Identified current (2024-2025) best practices
• [ ] Found at least one novel but stable approach
• [ ] Documented complete tool ecosystem
• [ ] Researched domain-specific compliance requirements
• [ ] Identified potential bias risks
• [ ] Compiled edge cases and limitations
• [ ] Validated findings across multiple sources
• [ ] Organized knowledge into clear categories
• [ ] Created citation list for all claims

Common Research Pitfalls

Avoid:
- Shallow searches - Don't stop at first search result
- Outdated sources - Pre-2022 may not reflect current practices
- Single-source bias - Validate across multiple authorities
- Over-engineering - Don't add complexity without clear benefit
- Ignoring domain ethics - Every domain has compliance requirements
- Tool hype - Popular ≠ appropriate for this use case
- Research paralysis - Know when you have enough to proceed

Red Flags:
- Only one source recommends a practice → Need more validation
- "Revolutionary" claims without evidence → Likely hype
- No production deployments → Too experimental
- Abandoned GitHub projects → Risk of unmaintained tools
- Vague compliance guidelines → Dig deeper, specifics matter

Output for Agent Builder

After completing research, provide to agent-builder skill:

1. Domain expertise summary - Core concepts, terminology, methodologies
2. Tool ecosystem - Libraries, APIs, file formats with versions
3. Best practices - Validated approaches from multiple sources
4. Novel techniques - Stable innovations to incorporate
5. Compliance requirements - Domain-specific regulations ON TOP of Coalition base ethics
6. Bias risks - Known problematic patterns to mitigate
7. Edge cases - Challenges and handling strategies

This research forms the foundation for the agent specification.

Related Marketplace Skills

agent-builder - Use research to design complete agent specification
constitutional-ai - Deep dive on ethical framework (Coalition base ethics)
agent-patterns - Architecture patterns for implementation

Important Notes

Current is Critical: 2024-2025 sources preferred. Pre-2022 may be outdated.

Multi-Source Validation: Don't trust single sources. 3+ authoritative sources minimum.

Stability Over Hype: Novel techniques must have production deployments, not just papers.

Domain Compliance: Research regulatory requirements specific to domain (HIPAA, SOX, GDPR, etc.)

Know When to Stop: You need enough to proceed, not perfect knowledge. Research paralysis helps nobody.