Claude Code
Specialized Agents
Open-sourcing our AI agent workflow for better software development
Why We Built This
At VAMFI, we believe AI should enhance human work, not replace it. That's why we've developed and open-sourced a specialized agent system for Claude Code that follows a structured research → plan → implement workflow.
This approach ensures every line of code is backed by current documentation, thoroughly planned, and precisely implemented—reducing errors and improving code quality.
The Three Agents
Purpose: Fetches current, version-accurate documentation from official sources
Use when: You need to implement features with specific libraries, update to new versions, or debug library-specific issues
Output: ResearchPack with authoritative API references, version info, and usage examples
Purpose: Creates detailed, actionable implementation plans based on research findings
Use when: You have research/requirements ready and need a concrete plan mapping to actual codebase files
Output: Step-by-step implementation plan with specific file changes, test criteria, and rollback strategies
Purpose: Executes surgical code edits following a predefined plan and research pack
Use when: You have both ResearchPack and Implementation Plan ready for execution
Output: Minimal, precise code changes with verification results
The Workflow
docs-researcher fetches official documentation
→ Produces: ResearchPack
implementation-planner creates actionable plan
→ Produces: Implementation Plan
code-implementer executes the plan
→ Produces: Working Code
Real-World Example
Adding Stripe Payment Integration
# Step 1: Research
User: "I need to add Stripe payment integration"
→ Claude uses docs-researcher to fetch latest Stripe API docs
# Step 2: Planning
User: "Create a plan for this integration"
→ Claude uses implementation-planner to design the implementation
# Step 3: Implementation
User: "Implement the payment system"
→ Claude uses code-implementer to execute the plan
Key Benefits
- Documentation-Backed: Every implementation is based on current, official documentation
- Predictable: Structured workflow reduces surprises and improves consistency
- Reversible: All changes include rollback strategies
- Testable: Every change comes with verification criteria
- Minimal: Surgical edits reduce risk and improve code review
Get Started
Ready to improve your AI-assisted development workflow?
GitHub Repository: github.com/irenicj/claude-code-specialized-agents
The setup is simple—just add our CLAUDE.md configuration to your global or project directory, and Claude Code will automatically use these specialized agents.
# Clone the repository
git clone https://github.com/irenicj/claude-code-specialized-agents.git
# Copy CLAUDE.md to your home directory
cp CLAUDE.md ~/.claude/
# Or add to a specific project
cp CLAUDE.md /path/to/your/project/
Join the Community
We believe in building tools that enhance human capabilities. If you share this vision:
- Star the repository on GitHub
- Report issues or suggest improvements
- Share your experience using the agents
- Contribute to the project
Together, we can make AI-assisted development more reliable, predictable, and human-centered.