Loop Engineering Study

An executable, local-first course for Python developers learning how AI/Agent loops improve through decision, action, evaluation, feedback, and stopping policies.

The Chinese-first course README is the primary entry point. In about 45 minutes, its deterministic code-repair case lets you inspect a failing loop, run three controlled experiments, and verify one improvement with before/after artifacts. No API key, model service, or network is required.

Start here

python -m pip install -e ".[dev]"
python experiments/code_repair/baseline.py

Then follow the three lessons:

1. Baseline
2. Read the trace
3. Improve the loop

The small reusable framework exposes Policy, Action, Evaluator, StopPolicy, structured traces, and JSON artifacts. Follow-up material is organized in the reference index and advanced index.

Development

python -m pytest -q
python scripts/check_docs.py
python -m build --wheel

See CONTRIBUTING.md and the release checklist. Released under the MIT License.