Getting Started with Human Compiler
This guide walks you through everything you need to go from a fresh machine to running Human Compiler and putting its generated skills to work.
Step 1: Install Claude Code
Human Compiler runs on top of Claude Code, Anthropic's agentic coding assistant for the terminal. You'll need a paid Anthropic account (Claude Pro, Max, Team, Enterprise, or API Console) — the free plan does not include Claude Code access.
macOS / Linux (Recommended)
The native installer is the simplest path. It requires no Node.js and auto-updates in the background:
curl -fsSL https://claude.ai/install.sh | bash
Windows
Open PowerShell (not CMD) and run:
irm https://claude.ai/install.ps1 | iex
Claude Code on Windows requires Git for Windows. Install it first from git-scm.com if you don't already have it.
macOS via Homebrew
brew install --cask claude-code
Homebrew installations do not auto-update. Run brew upgrade claude-code periodically to stay current.
Verify the installation
claude --version
You should see the current version number printed in your terminal. If you see command not found, restart your terminal and try again.
Authenticate
On first launch, Claude Code will prompt you to log in via your browser:
claude
Follow the browser prompts to connect your Anthropic account.
Step 2: Download Human Compiler from Git
Clone the Human Compiler repository to your local machine:
git clone https://github.com/your-org/human-compiler.git
cd human-compiler
Download and install Git from git-scm.com, then re-run the command above.
If the repository is private, make sure you have access and are authenticated with GitHub. You can use SSH:
git clone git@github.com:your-org/human-compiler.git
Step 3: Launch Human Compiler
From inside the cloned directory, start Human Compiler through Claude Code:
cd human-compiler
claude
Claude Code will load the project context automatically. Once the session starts, kick off the Human Compiler workflow by typing:
/run human-compiler
Human Compiler will initialize and begin its interactive interview session.
Step 4: Follow the Questions
Human Compiler works by asking you a series of focused questions to understand the skill it should generate. Here's what to expect:
What the interview covers
| Phase | What it asks |
|---|---|
| Goal | What should this skill accomplish? |
| Context | What kind of project or codebase will use it? |
| Constraints | Any limitations, libraries, or patterns to follow? |
| Examples | Sample inputs and expected outputs (optional but helpful) |
| Edge cases | Unusual scenarios the skill should handle gracefully |
Tips for better results
- Be specific. Instead of "parse a file," say "parse a CSV file with headers and return a list of dicts."
- Mention your stack. If you're using Python 3.12 with Pydantic, say so — the generated skill will match your environment.
- Provide examples when prompted. Even one concrete input/output pair significantly improves skill quality.
- Review intermediate output. Human Compiler may show you a draft and ask for feedback before finalizing.
Answer each question and press Enter to continue. Type skip to skip an optional question, or restart to begin the interview from the top.
Step 5: Use Your Generated Skills
Once the interview is complete, Human Compiler writes your skill to the /skills directory in your project.
Skill file structure
skills/
├── your-skill-name/
│ ├── SKILL.md ← Human-readable documentation
│ ├── skill.py ← Implementation (or .ts, .js, etc.)
│ └── examples/
│ └── example_1.md
Using a skill with Claude Code
You can invoke any generated skill directly in a Claude Code session. Navigate to your project and start a session:
cd your-project
claude
Then ask Claude to apply the skill:
Use the skill in skills/your-skill-name to process the data in /src/data.csv
Claude Code reads SKILL.md automatically for context and applies the implementation to your task.
Using skills programmatically
Import and call the generated implementation directly in your own code:
from skills.your_skill_name.skill import run
result = run(input_data)
print(result)
Regenerating or refining a skill
If a generated skill doesn't quite fit, re-run the interview with additional context:
/run human-compiler --refine skills/your-skill-name
Human Compiler will reload your previous answers and let you adjust any response before regenerating.
Quick Reference
| Task | Command |
|---|---|
| Install Claude Code (macOS/Linux) | curl -fsSL https://claude.ai/install.sh | bash |
| Install Claude Code (Windows) | irm https://claude.ai/install.ps1 | iex |
| Clone Human Compiler | git clone https://github.com/your-org/human-compiler.git |
| Launch Human Compiler | claude → /run human-compiler |
| Refine an existing skill | /run human-compiler --refine skills/<skill-name> |
| Update Claude Code (Homebrew) | brew upgrade claude-code |
Next Steps
- Browse the Skill Gallery to see examples of what Human Compiler can produce.
- Read the Skill Authoring Guide if you want to write or edit skills by hand.
- Check the Troubleshooting page if something isn't working as expected.