Codex MCAF ML Delivery for Baselines and Repeatable Experiments
Frame a small ML project using data assumptions, a baseline, evaluation outputs, and explicit stop conditions.
中文Install in Codex
Codex with FeatureChecker or a project using MCAF guidance, synthetic data, and existing evaluation notes.
- Copy the entire pinned `.codex/skills/mcaf-ml-ai-delivery/` directory to `.agents/skills/mcaf-ml-ai-delivery/` in the target checkout.
- Open Codex in that checkout and invoke `$mcaf-ml-ai-delivery` with the example task and desired output.
Example task
Give `$mcaf-ml-ai-delivery` a fictional book-tagging task. Define a keyword baseline, fixed validation split, and repeatable comparison.What to check
The plan names the baseline, fixture fields, evaluation steps, and limitations.
Original SKILL.md on GitHub · Updated
Frequently asked question
Why retain a baseline?
Compare with the same inputs and metrics so any benefit can be separated from split or measurement changes.