Four commands. No key, no Codex, no risk.
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python3 evals/score.py
python3 scripts/propose_improvement.pyExpected first run:
score=0.75
changed config/scoring.yaml
implementation_page_visit: 4 -> 6
score=1.00
open PR for human review
That is the whole idea. The system found a rule that was wrong, proved the fix against a test, and stopped to wait for a human.
Then install Codex and hand the same job to a model:
npm install -g @openai/codex
codex login
scripts/run_codex_step.sh improve_scoringCodex runs on your normal ChatGPT plan. No OpenAI API key is required.
This repo never sends messages and never merges its own changes.
The managed version is max.