AI Agents Collection for Claude Code & OpenCode

🙏 This repository is forked from wshobson/agents, the original creator of this comprehensive subagent collection. Thank you for building this amazing resource for the AI coding assistant community!

A comprehensive collection of specialized AI agents for both Claude Code and OpenCode, providing domain-specific expertise across software development, infrastructure, and business operations.

Overview

This repository provides production-ready agents that extend AI coding assistants' capabilities with specialized knowledge. The collection includes:

• 81 Claude Code agents - Optimized for Claude's subagent system
• 81 OpenCode skills - Mirrored from the Claude collection in OpenCode format

Each agent incorporates:

• Current industry best practices and standards (2024/2025)
• Production-ready patterns and enterprise architectures
• Deep domain expertise with 8-12 capability areas per agent
• Modern technology stacks and frameworks
• Optimized model selection based on task complexity

Repository Structure

agents/
├── claude/          # Claude Code agents (flat files, e.g. claude/python-pro.md)
├── opencode/        # OpenCode skills (per-skill dirs, e.g. opencode/python-pro/SKILL.md)
├── scripts/         # Conversion, linting, and utility scripts
├── docs/            # Usage and migration docs
├── CLAUDE.md        # Claude Code specific documentation
├── OPENCODE.md      # OpenCode specific documentation
└── README.md        # This file

Platform-Specific Documentation

• Claude Code agents: See CLAUDE.md for setup and usage
• OpenCode agents: See OPENCODE.md for setup and usage

Claude Code Agents

Claude agents live as flat markdown files in claude/.

Quick examples:

• claude/code-reviewer.md
• claude/security-auditor.md
• claude/python-pro.md
• claude/cloud-architect.md

To list all available Claude agents:

ls claude/*.md

Model Configuration

Claude agents specify model choice in each file's frontmatter (haiku, sonnet, opus, or provider-specific model IDs). Model assignment is based on task complexity and expected reasoning depth.

To inspect model usage in the current repo:

rg '^model:' claude/*.md

OpenCode Agents

OpenCode skills are organized as one directory per skill under opencode/, each with a SKILL.md file. The current repository mirrors 81 skills from the Claude set.

Quick examples:

• opencode/code-reviewer/SKILL.md
• opencode/security-auditor/SKILL.md
• opencode/python-pro/SKILL.md
• opencode/cloud-architect/SKILL.md

To list all available OpenCode skills:

ls opencode/*/SKILL.md

OpenCode Features

• Open-source focused: Prioritizes FOSS tools and libraries
• Privacy by default: Local-first processing and data minimization
• Transparent operation: Includes reasoning traces and decision explanations
• Community oriented: Supports collaborative development patterns
• Self-hostable: Designed for on-premise and air-gapped deployments

Installation

For Claude Code

Note: Claude Code expects agents directly in ~/.claude/agents/, not in subdirectories. See INSTALLATION.md for detailed setup options.

Quick setup with symlink:

git clone https://github.com/brianluby/agents.git ~/agents-repo
ln -s ~/agents-repo/claude ~/.claude/agents

For OpenCode

# Clone the repository
git clone https://github.com/brianluby/agents.git

# Link OpenCode agents globally
ln -s /path/to/agents/opencode/* ~/.config/opencode/agent/

# Or link to a project
ln -s /path/to/agents/opencode/* .opencode/agent/

See OPENCODE.md for detailed OpenCode setup instructions.

Linting Agents

Validate agent metadata before committing:

python scripts/lint_agents.py --roots opencode claude --require-model

Auto-fill missing model values:

python3 scripts/lint_agents.py --roots opencode --fix-missing-model zai/glm-4.6

Auto-fix canonical key ordering & validate allowed tools:

python scripts/lint_agents.py --roots opencode claude --check-order --require-model

Soft (non-failing) report mode:

python scripts/lint_agents.py --roots opencode --require-model --warn-only

Install pre-commit hook:

ln -s ../../.githooks/agent-lint .git/hooks/pre-commit

See CHANGELOG.md for normalization history.

Tooling Utilities (Phase 1)

The repo ships lightweight tool scripts (under scripts/tools/) used by some OpenCode agents or local workflows:

• glob_tool.py – Glob file listing (no content read)
• grep_tool.py – Regex line search with match cap (default 500)
• diff_tool.py – Unified diff (file↔file / dir↔dir) with byte limit
• format_tool.py – Minimal trailing-space + final newline normalizer (dry-run by default)
• webfetch_tool.py – Disabled-by-default HTTP(S) fetch (text-only, 100 KB cap, requires --allow or OPENCODE_ALLOW_WEB=1)

List allowed agent tool names:

python scripts/lint_agents.py --list-tools

Enable only what an agent strictly needs; omit webfetch unless remote context is essential.

Usage

Automatic Delegation

Claude Code automatically selects the appropriate subagent based on task context and requirements. The system analyzes your request and delegates to the most suitable specialist.

Explicit Invocation

Specify a subagent by name to use a particular specialist:

"Use code-reviewer to analyze the recent changes"
"Have security-auditor scan for vulnerabilities"
"Get performance-engineer to optimize this bottleneck"

Usage Examples

Code Quality & Security

code-reviewer: Analyze component for best practices
security-auditor: Check for OWASP compliance
tdd-orchestrator: Implement feature with test-first approach
performance-engineer: Profile and optimize bottlenecks

Development & Architecture

backend-architect: Design authentication API
frontend-developer: Create responsive dashboard
graphql-architect: Design federated GraphQL schema
mobile-developer: Build cross-platform mobile app

Infrastructure & Operations

devops-troubleshooter: Analyze production logs
cloud-architect: Design scalable AWS architecture
network-engineer: Debug SSL certificate issues
database-admin: Configure backup and replication
terraform-specialist: Write infrastructure modules

Data & Machine Learning

data-scientist: Analyze customer behavior dataset
ai-engineer: Build RAG system for document search
mlops-engineer: Set up experiment tracking
ml-engineer: Deploy model to production

Business & Documentation

business-analyst: Create metrics dashboard
documentation-expert: Generate technical documentation
api-documenter: Write OpenAPI specifications
content-marketer: Create SEO-optimized content

Multi-Agent Workflows

Subagents coordinate automatically for complex tasks. The system intelligently sequences multiple specialists based on task requirements.

Common Workflow Patterns

Feature Development

"Implement user authentication"
→ backend-architect → frontend-developer → test-automator → security-auditor

Performance Optimization

"Optimize checkout process"
→ performance-engineer → database-optimizer → frontend-developer

Production Incidents

"Debug high memory usage"
→ incident-responder → devops-troubleshooter → debugger → performance-engineer

Infrastructure Setup

"Set up disaster recovery"
→ database-admin → database-optimizer → terraform-specialist

ML Pipeline Development

"Build ML pipeline with monitoring"
→ mlops-engineer → ml-engineer → data-engineer → performance-engineer

Integration with Claude Code Commands

For sophisticated multi-agent orchestration, use the Claude Code Commands collection which provides 52 pre-built slash commands:

/full-stack-feature   # Coordinates 8+ agents for complete feature development
/incident-response    # Activates incident management workflow
/ml-pipeline         # Sets up end-to-end ML infrastructure
/security-hardening  # Implements security best practices across stack

Subagent Format

Each subagent is defined as a Markdown file with frontmatter:

---
name: subagent-name
description: Activation criteria for this subagent
model: haiku|sonnet|opus  # Optional: Model selection
tools: tool1, tool2       # Optional: Tool restrictions
---

System prompt defining the subagent's expertise and behavior

Model Selection Criteria

• haiku: Simple, deterministic tasks with minimal reasoning
• sonnet: Standard development and engineering tasks
• opus: Complex analysis, architecture, and critical operations

Agent Orchestration Patterns

Sequential Processing

Agents execute in sequence, passing context forward:

backend-architect → frontend-developer → test-automator → security-auditor

Parallel Execution

Multiple agents work simultaneously on different aspects:

performance-engineer + database-optimizer → Merged analysis

Conditional Routing

Dynamic agent selection based on analysis:

debugger → [backend-architect | frontend-developer | devops-troubleshooter]

Validation Pipeline

Primary work followed by specialized review:

payment-integration → security-auditor → Validated implementation

Agent Selection Guide

Architecture & Planning

Task	Recommended Agent	Key Capabilities
API Design	backend-architect	RESTful APIs, microservices, database schemas
Cloud Infrastructure	cloud-architect	AWS/Azure/GCP design, scalability planning
UI/UX Design	ui-ux-designer	Interface design, wireframes, design systems
System Architecture	architect-review	Pattern validation, consistency analysis

Development by Language

Language Category	Agents	Primary Use Cases
Systems Programming	c-pro, cpp-pro, rust-pro, golang-pro	OS interfaces, embedded systems, high performance
Web Development	javascript-pro, typescript-pro, python-pro, ruby-pro, php-pro	Full-stack web applications, APIs, scripting
Web Frameworks	django-pro, fastapi-pro	Django and FastAPI development
Enterprise	java-pro, csharp-pro, scala-pro	Large-scale applications, enterprise systems
Mobile	ios-developer, flutter-expert, mobile-developer	Native and cross-platform mobile apps
Specialized	elixir-pro, unity-developer, minecraft-bukkit-pro	Domain-specific development

Operations & Infrastructure

Task	Recommended Agent	Key Capabilities
Production Issues	devops-troubleshooter	Log analysis, deployment debugging
Critical Incidents	incident-responder	Outage response, immediate mitigation
Database Performance	database-optimizer	Query optimization, indexing strategies
Database Operations	database-admin	Backup, replication, disaster recovery
Infrastructure as Code	terraform-specialist	Terraform modules, state management
Network Issues	network-engineer	Network debugging, load balancing
Kubernetes	kubernetes-engineer	K8s configurations, cluster management
Platform Engineering	platform-engineer	Internal developer platforms, service mesh
Site Reliability	sre-engineer	SLIs/SLOs, error budgets, reliability practices

Quality & Security

Task	Recommended Agent	Key Capabilities
Code Review	code-reviewer	Security focus, best practices
Security Audit	security-auditor	Vulnerability scanning, OWASP compliance
Test Creation	test-automator	Unit, integration, E2E test suites
Performance Issues	performance-engineer	Profiling, optimization
Bug Investigation	debugger	Error resolution, root cause analysis
QA Engineering	qa-engineer	Manual testing strategies, test plans
Config Security	config-security-auditor	Configuration security, secrets management
Supply Chain Security	supply-chain-security	Dependency scanning, SBOM generation

Data & Machine Learning

Task	Recommended Agent	Key Capabilities
Data Analysis	data-scientist	SQL queries, statistical analysis
LLM Applications	ai-engineer	RAG systems, prompt pipelines
ML Development	ml-engineer	Model training, feature engineering
ML Operations	mlops-engineer	ML infrastructure, experiment tracking

Documentation & Business

Task	Recommended Agent	Key Capabilities
Technical Docs	documentation-expert	Comprehensive documentation generation
API Documentation	api-documenter	OpenAPI/Swagger specifications
Technical Writing	documentation-expert	User guides, tutorials, documentation
Tutorial Creation	documentation-expert	Step-by-step educational content
Reference Building	reference-builder	Technical references, API docs
Product Management	product-manager	Feature prioritization, roadmaps
Business Metrics	business-analyst	KPI tracking, reporting
Legal Compliance	legal-advisor	Privacy policies, terms of service

Best Practices

Task Delegation

1. Automatic selection - Let Claude Code analyze context and select optimal agents
2. Clear requirements - Specify constraints, tech stack, and quality standards
3. Trust specialization - Each agent is optimized for their specific domain

Multi-Agent Workflows

1. High-level requests - Allow agents to coordinate complex multi-step tasks
2. Context preservation - Ensure agents have necessary background information
3. Integration review - Verify how different agents' outputs work together

Explicit Control

1. Direct invocation - Specify agents when you need particular expertise
2. Strategic combination - Use multiple specialists for validation
3. Review patterns - Request specific review workflows (e.g., "security-auditor reviews API design")

Performance Optimization

1. Monitor effectiveness - Track which agents work best for your use cases
2. Iterative refinement - Use agent feedback to improve requirements
3. Complexity matching - Align task complexity with agent capabilities

Contributing

To add a new subagent:

1. Create a new .md file with appropriate frontmatter
2. Use lowercase, hyphen-separated naming convention
3. Write clear activation criteria in the description
4. Define comprehensive system prompt with expertise areas

Troubleshooting

Agent Not Activating

• Ensure request clearly indicates the domain
• Be specific about task type and requirements
• Use explicit invocation if automatic selection fails

Unexpected Agent Selection

• Provide more context about tech stack
• Include specific requirements in request
• Use direct agent naming for precise control

Conflicting Recommendations

• Normal behavior - specialists have different priorities
• Request reconciliation between specific agents
• Consider trade-offs based on project requirements

Missing Context

• Include background information in requests
• Reference previous work or patterns
• Provide project-specific constraints

License

MIT License - see LICENSE file for details.

Resources

• Claude Code Documentation
• Subagents Documentation
• Claude Code GitHub
• Claude Code Commands