Skip to content

AGENTS.md

Latest commit

 

History

History
31 lines (23 loc) · 1.22 KB

AGENTS.md

File metadata and controls

31 lines (23 loc) · 1.22 KB
Raw

Build Your Own Claude Code Workshop

Learn how modern AI coding agents work by building one from scratch: from a 17-line agent to a full production-grade system with teams and worktree isolation.

This is teaching material, not a product. The bar is that a learner can clone it and run it. Optimise for a working first command and a clear error when something is missing, not for architecture.

Running it

make help
make install
make run
make test

All targets: help, install, run, test, clean.

Python project (pyproject.toml). Use the repo's own virtualenv; never install into the system interpreter.

Rules

  • Never break the first-run path. A learner hitting an error in step one abandons the workshop. If you change setup, run it from a clean clone before calling it done.
  • Explanations belong in the README and in notebook prose, not in long code comments. The code is the lesson; it should read as the clearest version of the idea.
  • No em dashes in prose, per the house rules in /Users/param/.claude/skills/lwp-shared/house-rules.md.
  • Pin what you can. A workshop that worked last term and breaks this term because an upstream package moved is the most common failure here.