Research Methodology

Research Methodology Skill

This skill provides a systematic methodology for conducting rapid, accurate documentation research to ground implementations in truth rather than potentially stale LLM memory.

When Claude Should Use This Skill

Claude will automatically invoke this skill when:

• User mentions implementing/using a specific library or API
• User asks about current documentation for a technology
• User requests verification of API signatures or methods
• Task requires external dependencies or third-party integrations
• Updating/upgrading to new versions of frameworks

Core Principles (BRAHMA Constitution)

1. Truth over speed - But achieve both through systematic approach
2. Never guess APIs - Always retrieve from authoritative sources
3. Cite everything - Include version, URL, and section references
4. Deterministic - Same research query should yield consistent results

Research Methodology Protocol

Step 1: Rapid Assessment (< 30 seconds)

Objectives:

• Identify what needs to be researched
• Detect current versions from project files
• Clarify any ambiguities

Actions:

1. Parse request - Extract library/API names mentioned

2. Version detection - Check project dependency files:

   • package.json → Node.js projects
   • requirements.txt, pyproject.toml, Pipfile → Python
   • go.mod → Go
   • Cargo.toml → Rust
   • build.gradle, pom.xml → Java
   • *.csproj → C#/.NET
   • pubspec.yaml → Dart/Flutter
   • composer.json → PHP

3. Context gathering - Note runtime, platform, existing dependencies

Output:

Target: [library-name]
Current Version: [X.Y.Z] (detected from [file])
Platform: [Node.js 20.x / Python 3.11 / etc.]

If unclear: Ask ONE specific clarifying question rather than proceed with assumptions

Step 2: Source Prioritization (< 10 seconds)

Source hierarchy (in order of preference):

1. Official documentation (PRIMARY)

   • Project's official website
   • Official API reference
   • Official getting started guide

2. Official migration/upgrade guides (if version change)

   • Breaking changes documentation
   • Migration paths
   • Deprecation notices

3. Official release notes/changelog (for version-specific info)

   • What's new in this version
   • Bug fixes relevant to use case
   • Known issues

4. Official GitHub repository (if docs sparse)

   • README.md
   • Examples directory
   • Issue tracker (for known problems)

5. Avoid (unless no alternatives):

   • Blog posts (may be outdated)
   • Stack Overflow (may be for wrong version)
   • AI-generated content (circular hallucination risk)

Step 3: Information Retrieval (< 90 seconds)

Retrieval strategy:

1. Try context7 system (if available)
   └─ Fastest, curated, version-aware docs

2. Use WebFetch on known official doc URLs
   └─ Direct fetch from source

3. If URL unknown, use WebSearch
   Query format: "[library name] [version] official documentation"
   └─ Find the official site first, then fetch

4. Extract only relevant sections
   └─ Don't download entire docs, target specific info needed

What to extract:

• API signatures - Function names, parameters, return types
• Setup/initialization - How to configure and start using
• Code examples - Minimal working examples (with URLs to source)
• Gotchas - Known issues, breaking changes, version-specific warnings
• Best practices - Recommended usage patterns from docs

Anti-stagnation:

• Set 60-second timeout per source
• If source fetch fails, report and try next source
• If all sources fail, report what was attempted and suggest manual research

Step 4: Verification & Citation (< 30 seconds)

For every piece of information extracted:

API: someFunction(param1: Type): ReturnType
Source: official-docs.com/api-reference/someFunction [version X.Y.Z]

Verification checklist:

• ✓ Version matches project dependency
• ✓ Source is official (not third-party)
• ✓ URL links to specific section (not just homepage)
• ✓ Information is current (check doc version/date if shown)
• ✓ Examples are complete and runnable

Confidence levels:

• HIGH: Official docs for exact version, multiple sources corroborate
• MEDIUM: Official docs but slight version mismatch, or single source
• LOW: Only unofficial sources, version mismatch, or deprecated docs

Report confidence: Always indicate confidence level in research output

Step 5: Structured Output (< 30 seconds)

Deliver in ResearchPack format:

# 📋 ResearchPack: [Library Name]

## Quick Reference
- Library: [name] v[X.Y.Z]
- Official Docs: [URL]
- Confidence: [HIGH/MEDIUM/LOW]

## Key APIs
[Concise list with signatures and sources]

## Setup
[Essential steps only, not every option]

## Gotchas
[Version-specific issues to avoid]

## Example
[Minimal working code with source link]

## Implementation Checklist
[Files to modify, steps in order]

## Sources
[Numbered list with URLs and versions]

## Open Questions
[Any decisions/clarifications needed]

Keep it concise:

• Use bullet points, not paragraphs
• Include only essential information for implementation
• Provide links for deeper reading (don't excerpt everything)

Quality Assurance

Before delivering research:

1. Completeness check:

   • ✓ Library identified with version
   • ✓ At least 3 key APIs documented
   • ✓ Setup steps provided
   • ✓ At least 1 code example
   • ✓ All claims cited with sources

2. Accuracy check:

   • ✓ API signatures match docs exactly (not paraphrased)
   • ✓ Version numbers correct
   • ✓ URLs all valid and point to official sources
   • ✓ No hallucinated functions/parameters

3. Usefulness check:

   • ✓ Implementation checklist actionable
   • ✓ Examples are complete (can be run)
   • ✓ Gotchas relevant to detected version
   • ✓ Open questions identify real decisions needed

Common Pitfalls to Avoid

❌ Don't:

• Use docs from wrong version
• Guess or fill in gaps from LLM memory
• Copy-paste entire doc pages (extract key info)
• Trust unofficial blogs/tutorials over official docs
• Proceed without citing sources
• Skip version detection step

✅ Do:

• Always detect version from project files first
• Cite every API with source URL + section
• Extract concise, relevant info only
• Report confidence level honestly
• Link to specific doc sections, not just homepage
• Note version-specific gotchas explicitly

Performance Targets

• Total time: < 2 minutes for typical library research
• Accuracy: 100% of APIs must match official docs
• Citations: 100% of claims must have sources
• Completeness: 95%+ of info needed for implementation

Example Usage

Scenario: User says "Add Redis caching to the product service"

Claude (using this skill):

[Internal: This requires research on Redis client library]

🔍 Activating research-methodology skill

📦 Detecting Redis version from package.json... Found redis@4.6.0

🌐 Fetching official docs from redis.io...

[Follows methodology: assess → prioritize → retrieve → verify → output]

# 📋 ResearchPack: Redis Client (Node.js)

## Quick Reference
- Library: redis v4.6.0
- Official Docs: https://redis.io/docs/clients/nodejs/
- Confidence: HIGH

## Key APIs
1. `createClient(options)` - Initialize connection
   Source: redis.io/docs/clients/nodejs/v4.6#createClient

2. `client.get(key: string): Promise<string | null>` - Retrieve value
   Source: redis.io/docs/clients/nodejs/v4.6#get

[...rest of ResearchPack]

✅ Research complete - Ready for planning phase

This skill ensures all implementations are grounded in current, authoritative documentation rather than potentially outdated LLM knowledge.