Context Engineering Enablement

Every model has a context window, but capacity isn't quality.

Hands-on exercises for working with a coding agent without letting its context get away from you. The repo is a small toy Python project, a wheel bakery, with a noisy test suite and forty build logs: enough to feel the meter. You'll watch the agent's context fill, watch it break, and learn the levers that control it.

Written for Claude Code. Codex CLI notes are inline where commands differ.

Setup

git clone https://github.com/TorchedHat/context-engineering-enablement.git
cd context-engineering-enablement
claude

Type hello and get a reply. That's it. Nothing to install.

Exercises

Seven exercises in managing context and memory inside a coding session, and three standalone follow-ons. Start at exercises/.

Repository layout

.
├── README.md                 this file
├── CLAUDE.md                 standing rules for Claude Code, loaded at session start
├── AGENTS.md                 the same rules, for Codex
├── pyproject.toml            project metadata
├── .claude/
│   ├── settings.json         model pin and status line for these exercises
│   └── statusline.sh         status bar: model, context used, session cost
├── bakery/                   the toy app
│   ├── config.py             oven and wheel configuration
│   └── oven.py               the bake loop
├── tests/
│   └── test_oven.py          test suite; prints enough output to matter
├── data/
│   ├── catalog.json          wheel catalog
│   └── logs/                 40 pre-generated build logs, bake-0000 to bake-0039
├── tools/
│   └── gen_data.py           regenerates data/
└── exercises/                one file per exercise, plus an index