A recreation of the Claude Agent SDK demos built from Anthropic, but in Python.
It utilizes the patterns mentioned from the Building Effective Agents article, so it's not a 1:1 recreation.
This repo was made purely for educational purposes in order to get more practice building with the claude-agent-sdk along with applying the patterns mentioned from the article.
Each of the Agents include an iteration.md file that can be fed into claude code so that it can give you instructions on how to build each of the agents from the ground up.
| Agent | What It Does | Pattern Used |
|---|---|---|
hello-world |
Basic setup | Getting started |
email-agent |
Sort and write emails | Routing, Tools |
excel-agent |
Edit spreadsheets | Evaluator-Optimizer |
resume-generator |
Create resumes from web info | Tools |
content-pipeline-agent |
Pull out and summarize content | Prompt Chaining |
task-breakdown-agent |
Turn goals into task lists | Orchestrator-Workers |
- Copy the contents from
base-prompt.txtand pass that into claude-code - Pick an agent folder
- Deed in the
iteration.mdfile into claude-code - Follow the instructions given from claude-code
git clone <repository-url>
cd claude-agent-python
python -m venv venv
source venv/bin/activate
pip install claude-agent-sdk pydanticYou need Python 3.11 or newer and an Anthropic API key.
claude-agent-python/
├── hello-world/ # Start here
├── email-agent/
├── excel-agent/
├── resume-generator/
├── content-pipeline-agent/
├── task-breakdown-agent/
└── .claude/skills/ # Helper docs for Claude
A few supporting skills were created in .claude/skills/ to help guide how the agents should be created:
building-effective-agents— Patterns from Anthropic's articlestructured-outputs-reference— Get typed JSON responsessubagents-sdk-reference— Use worker agentsagent-skills-reference— Make custom skills