Skip to content

cosmix/claude-code-setup

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Claude Code Agent & Skills Library

A comprehensive collection of hierarchical AI agents and reusable skills for Claude Code CLI, designed to optimize performance across software engineering, security, machine learning, infrastructure, product design, analytics, QA, data engineering, and technical writing domains.

Overview

This project provides:

  • 19 Specialized Agents - Senior (opus) and Standard (sonnet) pairs across 8 domains, plus Security Engineer, Tech Lead, and Doc Editor
  • 48 Reusable Skills - Modular capabilities that agents can leverage

Why Agents + Skills?

Claude Code's power comes from combining agents (specialized subagents with focused expertise) and skills (reusable knowledge modules). This architecture provides:

Context Efficiency

Each subagent runs in its own context window. Instead of one massive conversation that hits context limits, work is distributed across multiple focused agents. This means:

  • Larger projects: Break down 50-file refactors into parallel subagent tasks
  • Preserved context: The main conversation stays clean while subagents handle details
  • Better results: Each agent focuses on its specialty without context pollution

Versatility Through Composition

Skills are loaded dynamically based on the task. A software-engineer agent working on a Python API can automatically load python, api-design, and testing skills—getting specialized knowledge without bloating every conversation.

Parallel Execution

Subagents can run in parallel (up to 10 concurrent). A single prompt like "refactor authentication across all services" can spawn multiple agents working simultaneously on different files, dramatically reducing total time.

Learn more: Subagents Documentation · Agent Skills Blog Post

Installation

Prerequisites

Quick Install

git clone https://github.com/YOUR_USERNAME/claude-code-setup.git
cd claude-code-setup
bash install.sh

The installer will:

  • Copy agents and skills to ~/.claude/
  • Append orchestration rules to ~/.claude/CLAUDE.md
  • Back up any existing files (as *.bak.<timestamp>)

Manual Install

If you prefer to install manually or to a specific project:

# User-level (available in all projects)
cp -r agents ~/.claude/
cp -r skills ~/.claude/
cat CLAUDE.template.md >> ~/.claude/CLAUDE.md

# Project-level (team-shared via git)
cp -r agents /path/to/your/project/.claude/
cp -r skills /path/to/your/project/.claude/
cat CLAUDE.template.md >> /path/to/your/project/CLAUDE.md

What Gets Installed

The CLAUDE.template.md configuration includes:

  • Agent orchestration - When to use senior (opus) vs standard (sonnet) agents
  • Parallel/sequential execution - Guidelines for spawning agents efficiently
  • Context passing - What information to provide subagents
  • Development standards - Implementation, planning, documentation, code quality
  • Dependency management - Always use package managers, never edit manifests manually
  • Progress tracking - How to record and clean up task progress

Verify Installation

Start a Claude Code session and check that agents and skills are loaded:

claude

# Inside Claude Code, ask:
> What agents are available?
> What skills do you have access to?

Official Documentation:

How the CLAUDE.md Configuration Works

The CLAUDE.md instructions you add to ~/.claude/CLAUDE.md become part of Claude's system context. They guide Claude to:

  1. Recognize delegation opportunities - When you ask for a complex task, Claude checks if specialist agents can handle parts of it

  2. Choose the right agent tier - Senior agents (opus) for architecture/debugging, standard agents (sonnet) for implementation

  3. Parallelize independent work - Claude spawns multiple subagents simultaneously when tasks don't depend on each other

  4. Manage context efficiently - Instead of trying to do everything in one context window, Claude distributes work across focused subagents

Example: Multi-File Refactoring

Without orchestration rules, Claude might try to refactor 20 files sequentially in one context, eventually hitting limits.

With orchestration rules, Claude will:

1. Use tech-lead to analyze the refactoring scope
2. Spawn senior-software-engineer to design the approach
3. Spawn multiple software-engineer agents IN PARALLEL to refactor different files
4. Each agent loads relevant skills (python, refactoring, testing)
5. Results merge back without exhausting main context

Agent Hierarchy

Each domain has two agents with distinct responsibilities:

Domain Senior (opus) Standard (sonnet)
Software Engineering senior-software-engineer software-engineer
Security security-engineer (single agent)
Machine Learning senior-ml-engineer ml-engineer
Infrastructure senior-infrastructure-engineer infrastructure-engineer
Product Design senior-product-designer product-designer
Analytics senior-data-analyst data-analyst
Quality Assurance senior-qa-engineer qa-engineer
Data Engineering senior-data-engineer data-engineer
Technical Writing senior-technical-writer technical-writer

Special Agents

Agent Model Purpose
tech-lead opus Cross-functional coordination, project planning, work distribution
doc-editor haiku Markdown linting, formatting fixes, documentation consistency

When to Use Senior Agents (opus)

Use senior agents for higher-level thinking and complex work:

  • Planning - System design, project architecture, implementation strategies
  • Architecture - Component design, API contracts, data modeling decisions
  • Difficult Algorithms - Complex logic, optimization problems, novel solutions
  • Design Patterns - Selecting and applying appropriate patterns
  • Debugging - Root cause analysis of complex issues
  • Code Review - Evaluating design decisions and code quality
  • Strategic Decisions - Technology selection, trade-off analysis

When to Use Standard Agents (sonnet)

Use standard agents for implementation and routine work:

  • Boilerplate Code - Standard implementations, CRUD operations
  • Well-Defined Components - Fleshing out specs that are already designed
  • Routine Tasks - Following established patterns and conventions
  • Standard Configurations - Writing configs, manifests, pipelines
  • Data Processing - ETL, preprocessing, standard transformations
  • Documentation - Writing docs for implemented features

Skills Library

Skills provide modular capabilities that agents can invoke. They are loaded dynamically based on the task context.

Language Expertise

Skill Description
python Pythonic idioms, type hints, async patterns, pytest
golang Go idioms, error handling, concurrency, modules
rust Ownership, lifetimes, error handling, cargo, traits
typescript Type system, generics, utility types, strict mode

Code Quality

Skill Description
code-review Comprehensive code reviews for correctness and maintainability
refactoring Restructure code without changing behavior
testing Create unit, integration, and e2e test suites
documentation Generate technical docs, API references, READMEs

Development

Skill Description
api-design Design RESTful APIs, GraphQL schemas, RPC interfaces
database-design Design schemas, relationships, indexes, migrations
git-workflow Branching strategies, commits, conflict resolution
debugging Systematic bug diagnosis and resolution

Documentation

Skill Description
technical-writing Clear prose, audience-aware docs, structure
diagramming Mermaid diagrams, architecture, sequences, ERDs
api-documentation OpenAPI specs, endpoint docs, SDK documentation
md-tables Markdown table alignment and spacing fixes

QA & Testing

Skill Description
test-strategy Test pyramid, coverage goals, what/when to test
e2e-testing Playwright/Cypress patterns, fixtures, selectors
performance-testing Load testing, benchmarking, profiling

Infrastructure

Skill Description
docker Dockerfiles and docker-compose optimization
kubernetes K8s deployments, services, configurations
terraform Infrastructure as Code for cloud resources
ci-cd CI/CD pipeline design and implementation

Security

Skill Description
security-audit Comprehensive vulnerability assessment
security-scan Quick routine checks (secrets, deps, SAST)
threat-model STRIDE/DREAD analysis, secure architecture
dependency-scan CVE scanning and license compliance
auth OAuth2, JWT, RBAC/ABAC, session management

Reliability & Operations

Skill Description
error-handling Error types, recovery strategies, propagation
logging-observability Structured logging, tracing, metrics, alerts
concurrency Async patterns, parallelism, race conditions
caching Cache strategies, invalidation, Redis patterns
code-migration Version upgrades, framework migrations
rate-limiting Throttling, backpressure, API quotas

Architecture Patterns

Skill Description
event-driven Message queues, pub/sub, event sourcing, CQRS
feature-flags Rollouts, A/B testing, kill switches
background-jobs Job queues, schedulers, workers, idempotency
webhooks Design, verification, retry logic, idempotency
serialization JSON/protobuf/msgpack, schema evolution

Data

Skill Description
sql-optimization Query analysis and performance tuning
data-visualization Charts, dashboards, visual analytics
data-validation Schema validation, sanitization, contracts
search Elasticsearch, full-text search, indexing

AI/ML

Skill Description
prompt-engineering LLM prompt design and optimization
model-evaluation ML model performance and fairness testing

Frontend

Skill Description
accessibility WCAG compliance, a11y testing, screen readers
i18n Internationalization, translations, RTL
react React patterns, hooks, state management

Usage

Using Agents

Agents are automatically invoked by Claude Code when tasks match their descriptions. You can also explicitly request them:

Use the senior-software-engineer agent to design the architecture
Have the ml-engineer preprocess this dataset

Using Skills

Skills are model-invoked based on context. Claude will automatically use relevant skills when appropriate:

Use the python skill for this implementation
Apply the e2e-testing skill to write Playwright tests

Customization

Adding New Agents

Create a new .md file in agents/:

---
name: my-agent
description: What this agent does. Use PROACTIVELY when relevant.
tools: Read, Edit, Write, Glob, Grep, Bash
model: sonnet
---

Your agent's system prompt here.

Adding New Skills

Create a new directory in skills/ with a SKILL.md:

---
name: my-skill
description: What this skill does and when to use it.
allowed-tools: Read, Grep, Glob
---

# My Skill

## Instructions

Step-by-step guidance.

## Best Practices

Key principles.

## Examples

Concrete examples.

Further Reading

License

MIT

About

A curated list of agents, skills and a CLAUDE.md starter for your agentic sessions!

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published