A curated collection of specialized agents for Claude Code, designed to enhance your AI-assisted development workflow with focused, expert capabilities.
Custom agents are specialized AI assistants in Claude Code that focus on specific tasks. Each agent is optimized with:
- Tailored instructions for their domain
- Specific tool permissions
- Optimized model selection
- Clear operational guidelines
- code-explorer - Intelligently locate and read files by path or role description using LSP for semantic navigation; extract specific code sections with surgical precision
- code-analyst - Deeply understand how systems work by examining documentation, code, and related materials; trace implementation flows and clarify ambiguous specifications
- forensics-analyst - Reverse-engineer unfamiliar codebases and understand the structure, behavior, and intent of code you didn't write; analyze only provided files without autonomous exploration
- feature-builder - Translate technical discussions and specs into production-ready code that aligns with existing architecture
- code-refactorer - Refactor code with surgical precision while maintaining exact functionality and minimizing disruption
- test-builder - Create comprehensive, well-reasoned test suites with clear traceability to requirements
- test-designer - Design and implement tests that validate functionality, match existing patterns, and provide meaningful coverage
- test-fixer - Update broken tests after code changes to match new implementations
- feature-reviewer - Evaluate proposed changes and provide advisory recommendations for approval decisions
- docs-maintainer - Keep documentation synchronized with code changes and architectural updates
- solution-architect - Evaluate different implementation approaches and recommend optimal solutions for technical problems
- design-reviewer - Identify unnecessary complexity and over-engineered patterns in specific code components or architectural decisions
- task-planner - Analyze project state and develop strategic roadmaps for future development
- task-completer - Systematically recover from failed tasks by learning from past attempts and identifying root causes
Skills (slash commands) provide quick shortcuts for common workflows:
- init-agents - Read
AGENTS.mdand understand the project context and related protocols - hint-commit - Generate a Conventional Commits–style commit message for the last N changes (pass count as argument)
- hint-commit-desc - Like
hint-commitbut outputs only the subject line — no body text - fix-test - Use the test-fixer agent to understand why a given test is failing and then fix it (pass the test name as an argument)
- Clone this repository:
git clone https://github.com/gsscoder/claude-coding-agents.git- Copy the agent files you want to use into your Claude Code agents directory:
# Linux/macOS
cp ./agents/*.md ~/.claude/agents/
# Windows
copy .\agents\*.md %USERPROFILE%\.claude\agents\- Copy the skill directories you want to use into your Claude Code skills directory:
# Linux/macOS
cp -r ./skills/* ~/.claude/skills/
# Windows
xcopy /e /i .\skills %USERPROFILE%\.claude\skills\- The agents and skills will be available in Claude Code immediately
Agents are automatically invoked by Claude Code based on your requests and the task context. You can also explicitly request a specific agent:
"Use the code-explorer agent to find all authentication files"
"I need the feature-builder agent to implement this spec"
Each agent is defined in a markdown file with frontmatter. You can customize:
description- When the agent should be usedtools- Which Claude Code tools the agent can accessmodel- Which Claude model to use (sonnet, opus, haiku)color- UI color coding
Edit the agent files to adapt them to your workflow and coding style.
Contributions are welcome! Please feel free to submit a Pull Request.
MIT License - feel free to use and modify these agents for your projects.