learn-claude-code: Learn to Build LLM Agent Harnesses

Overview

learn-claude-code is a progressive tutorial for building the software surrounding an LLM agent: the tools, context management, permissions, and coordination mechanisms that let a model operate in a coding environment. Its central teaching approach is to start with a small agent loop and add one harness mechanism at a time, rather than treating a large orchestration framework as the agent itself.

The repository is most useful as a learning resource, not as a ready-made coding agent. Its current course has 17 lessons with standalone Python examples, and its README distinguishes that track from a legacy 12-lesson version. It is aimed at developers who want to understand or prototype agent infrastructure and can work with an Anthropic API key.

Key Features

• Seventeen progressive lessons, each focused on a specific agent-harness mechanism.
• Runnable Python examples covering tool use, permissions, hooks, planning, subagents, and context compaction.
• Lessons on persistent memory, task graphs, background work, scheduling, and agent-team coordination.
• Examples of MCP tool integration, an integrated harness, workflow execution, and goal-based continuation.
• English, Chinese, and Japanese chapter materials, according to the README.
• A web app for exploring the course, alongside standalone chapter code.

Use Cases

• A developer new to LLM agents can follow the lessons to understand how a model calls tools and receives results.
• An engineer prototyping a coding agent can use the examples to study permissions, context handling, and task persistence before choosing a production architecture.
• An educator or team lead can use the course structure to teach harness concepts in small, runnable steps.
• A developer extending an agent with external tools can study the MCP lesson and how those tools fit into a shared tool pool.

Getting Started

git clone https://github.com/shareAI-lab/learn-claude-code
cd learn-claude-code
pip install -r requirements.txt
cp .env.example .env  # Configure ANTHROPIC_API_KEY
python s01_agent_loop/code.py

Start with the repository README for the course map, setup details, and guidance on the current and legacy tracks.

Alternatives

• ai-engineering-from-scratch: Covers a broad AI engineering curriculum, including agents and protocols, rather than focusing on building a Claude Code harness.

Considerations

• Running the examples requires Python dependencies and an Anthropic API key; the README's quick start uses pip and an .env file.
• The repository is a tutorial and set of examples, so it should not be assumed to provide a production-ready agent or deployment safeguards.
• The README identifies the root-level 17-lesson course as current and docs/ and agents/ as a legacy transition track. Readers should avoid mixing their lesson numbering.
• The concepts are centered on coding-agent harnesses. Applying them in other domains requires adapting tools, knowledge, permissions, and interfaces.