a5c-ai/research-first-dev

Research-First Development

Overview

Research-first development methodology adapted from the Everything Claude Code project. Mandates investigation of existing solutions and alternatives before writing any code.

Research Process

1. Problem Analysis

• Parse the request into specific technical requirements
• Identify the domain and relevant technology stack
• List known constraints (time, resources, compatibility)
• Define success criteria

2. Existing Solution Search

• Search GitHub for similar implementations
• Check package registries (npm, PyPI, crates.io, etc.)
• Review documentation for framework-specific solutions
• Identify relevant design patterns
• Check for known anti-patterns to avoid

3. Alternative Brainstorming

• Generate at least 3 alternative approaches
• Include a "build" option and at least one "buy/reuse" option
• Consider unconventional approaches

4. Trade-Off Evaluation

• Complexity: implementation effort, learning curve
• Time: development timeline, time-to-value
• Risk: failure modes, dependency risks, maintenance burden
• Scalability: growth limits, performance under load
• Score each alternative on all 4 axes

5. Recommendation

• Rank alternatives by composite score
• Provide clear recommendation with justification
• Include risk mitigation plan for chosen approach
• Define go/no-go criteria

Iterative Retrieval

• Start broad, narrow based on findings
• Use confidence scoring to decide when to stop
• Maximum 3 retrieval rounds per topic
• Cache findings for reuse in subsequent phases

When to Use

• New feature development (always)
• Architecture changes
• Technology selection
• Dependency evaluation
• Performance optimization strategy

Agents Used

• planner (primary consumer)
• architect (architecture-specific research)