Build and evaluate AI agent systems using real workshop code
For developers who want to build complex AI agent systems, this repository provides the code and scenarios used in actual workshops. It is a direct release of the educational materials from the 'Code with Claude' workshops run by Anthropic. Each folder contains the agent configuration and evaluation methods for solving a specific problem.
Practical examples by workshop topic
The repository includes workshop materials covering various topics, from model selection to production deployment.
- Agent competition and memory
agent-battle/covers a competition to configure an agent that controls a local game bot for 45 minutes, whileagents-that-remember/details the process of gradually applying a memory store and dream service to an agent that loses memory between sessions. - Evaluation-driven development
eval-driven-agent-development/iterates on a PPTX generation agent through six variations, using a two-layer evaluation that combines programmatic XML metrics and LLM judgment to measure the effect of each prompt change. - Production-grade systems
production-ready-agent/andresearch-desk/build a multi-agent M&A research team and an SEC document research desk, showing how to apply memory stores, MCP servers, and version-controlled agent updates.
Maintenance status and contribution policy
This repository is currently unmaintained and does not accept new contributions. The materials are records of workshops run by Anthropic and reflect the technology stack and API usage at a specific point in time. Therefore, they may not reflect the latest Claude API changes or new features. While they can be used for learning purposes, you need to verify them against current API documentation before applying them directly to a production environment. The license is Apache License 2.0, which permits commercial use and modification.
