Discover

Research methodology for AI agents conducting online discovery. Emphasizes external exploration over local reflection.

Purpose

Enable systematic discovery of techniques, libraries, tools, and approaches within a technical domain. Transform open-ended exploration into structured, actionable knowledge.

When to Use

• •Surveying a technical landscape (e.g., "What spec-to-SQL tools exist?")
• •Discovering libraries, frameworks, or approaches for a problem domain
• •Learning about unfamiliar technology areas
• •Gathering options before making architectural decisions
• •Building knowledge foundations for implementation planning

Research Methodology

Phase 1: Frame the Domain

Before searching, decompose the inquiry into searchable facets:

Domain: {primary subject}
├── Core terms: {canonical vocabulary}
├── Adjacent terms: {synonyms, related concepts}
├── Directions: {input → output transformations}
└── Ecosystem: {languages, platforms, communities}

Example: "spec to SQL"

Domain: Specification-driven database schema generation
├── Core: schema generation, DDL, migration, code generation
├── Adjacent: ORM, type-safe SQL, declarative schemas
├── Directions: spec→SQL, SQL→types, bidirectional
└── Ecosystem: TypeScript, Go, Python, Rust, PostgreSQL, MySQL

Phase 2: Search Strategy

Execute searches in waves, building on discoveries:

Wave 1 - Canonical Tools Search for established, well-documented solutions:

• •"{domain} tools libraries {year}"
• •"{input format} to {output format} generation"
• •"best {domain} tools"

Wave 2 - Format-Specific Search by specification format:

• •OpenAPI/Swagger → SQL
• •JSON Schema → DDL
• •GraphQL → database
• •Protobuf → SQL
• •TypeScript types → SQL

Wave 3 - Ecosystem-Specific Search by language/platform:

• •"{language} {domain} library"
• •"{framework} schema migration"

Wave 4 - Reverse Direction Often reveals bidirectional tools:

• •SQL → TypeScript types
• •Database → OpenAPI
• •Schema → documentation

Phase 3: Categorize Discoveries

Organize findings into a structured taxonomy:

Category: {transformation type}
├── Tool: {name}
│   ├── Input: {specification format}
│   ├── Output: {generated artifact}
│   ├── Languages: {supported}
│   └── URL: {reference}

Standard Categories:

1. •Declarative Schema DSLs - Purpose-built schema languages
2. •ORM Migration Systems - Code-first schema generation
3. •Specification Parsers - OpenAPI, JSON Schema, GraphQL converters
4. •Type-to-Schema - Language type systems to DDL
5. •AI-Assisted - Natural language to schema
6. •Bidirectional - Schema introspection and generation

Phase 4: Synthesize Findings

Structure the synthesis for actionability:

## {Domain} Landscape

### Approaches
| Approach | When to Use | Trade-offs |
|----------|-------------|------------|
| {name}   | {use case}  | {pros/cons}|

### Tool Matrix
| Tool | Input | Output | Maturity | Ecosystem |
|------|-------|--------|----------|-----------|

### Recommendations
- For {scenario}: Consider {tool} because {reason}

Research Best Practices

Query Construction

Effective patterns:

• •Include year for currency: "topic 2026"
• •Use quotes for exact phrases: "JSON Schema to PostgreSQL"
• •Combine terms: "{input} to {output} {language}"
• •Search alternatives: "{topic} alternatives competitors"

Avoid:

• •Single-word queries
• •Overly broad searches
• •Queries without context anchors

Progressive Refinement

1. •Start broad, identify vocabulary
2. •Use discovered terms in subsequent searches
3. •Follow references and "see also" patterns
4. •Cross-reference multiple sources

Source Evaluation

High signal sources:

• •Official documentation
• •GitHub repositories with activity
• •Technical blog posts with code examples
• •Conference talks and papers
• •Stack Overflow accepted answers

Low signal sources:

• •AI-generated listicles without depth
• •Outdated documentation (check dates)
• •Marketing pages without technical content

Citation Discipline

Always include sources with findings:

**{Tool Name}** - {brief description}
- Source: [{Title}]({URL})
- Last updated: {date if available}

Anti-Patterns

Avoid these research failures:

1. •Premature Closure - Stopping at first result
2. •Echo Chamber - Only searching one ecosystem
3. •Recency Bias - Ignoring mature, stable tools
4. •Novelty Bias - Chasing latest over proven
5. •Local Reflection - Using training data instead of searching
6. •Shallow Synthesis - Listing without categorizing

Example Application

Query: "What spec-to-SQL techniques and libraries exist?"

Execution:

1. •Frame: Specification → SQL DDL generation landscape
2. •Wave 1: "spec to SQL schema generation tools 2026"
3. •Wave 2: "OpenAPI to SQL", "JSON Schema to PostgreSQL", "GraphQL to database"
4. •Wave 3: "Prisma schema migration", "Drizzle ORM SQL", "TypeScript to SQL"
5. •Wave 4: "SQL to TypeScript types", "database to OpenAPI"
6. •Categorize by approach type
7. •Synthesize with tool matrix and recommendations

Additional Resources

Reference Files

• •references/spec-to-sql-landscape.md - Comprehensive catalog of spec-to-SQL tools
• •references/research-patterns.md - Extended methodology and examples

Example Output

• •examples/discovery-report.md - Sample discovery report structure