This repository offers workshop materials for developing AI agents using Anthropic's Claude Managed Agents. It targets developers and researchers seeking practical, hands-on experience with advanced agentic patterns, including multi-agent systems, eval-driven development, and production deployment scenarios, providing a clear path to understanding and implementing sophisticated LLM-powered applications.

How It Works

The workshops showcase Anthropic's methodology for constructing advanced AI agents, emphasizing the use of Claude Managed Agents, skills, and the Message Communication Protocol (MCP). Key architectural patterns demonstrated include decomposing complex problems into multi-agent systems, integrating memory primitives for state persistence across sessions, and employing rigorous eval-driven development loops. These loops utilize programmatic metrics and LLM-as-judge evaluations to systematically iterate and enhance agent performance and reliability.

Quick Start & Requirements

No explicit installation or setup instructions are provided within the README. The workshops inherently require familiarity with and access to Anthropic's Claude Managed Agents API and a suitable development environment for interacting with these services.

Highlighted Details

• Demonstrates optimizing LLM inference configurations for achieving the best quality-per-dollar and quality-per-second trade-offs.
• Features the composition of multi-agent systems using skills, code execution capabilities, and MCP for intricate task decomposition.
• Illustrates an AI-assisted product development workflow, spanning from initial interviews to generating a Vite+React application with runtime verification mechanisms.
• Includes examples of deploying production-ready agents, such as an incident dashboard powered by an SRE agent capable of log analysis and commit identification.
• Showcases the implementation of memory primitives for agents, progressing from basic amnesia to robust cross-session persistence and knowledge consolidation.
• Details eval-driven development for agents, employing multi-layered grading systems (combining programmatic metrics with LLM-as-judge assessments) to guide prompt iteration and refinement.

Maintenance & Community

The repository is explicitly marked as "Not maintained and not accepting contributions." No community links, roadmaps, or contributor information are provided.

Licensing & Compatibility

The project is licensed under the Apache License 2.0. This license generally permits commercial use, modification, and distribution, provided the terms of the license are followed.

Limitations & Caveats

The primary limitation is the project's status as "Not maintained and not accepting contributions," indicating potential for outdated information or unaddressed issues. The materials are specifically tied to Anthropic's Claude Managed Agents, necessitating access to and familiarity with that particular platform.