iCog-Labs-Dev/subrep ? reverse-engineered prompt

Build me a Python project that demonstrates a certificate driven skill library for reinforcement learning.

I want a demo where the system tries a bunch of candidate skills, checks whether each one is safe to add using two admission tests, then only stores the skills that pass. It should keep an audit trail of what got accepted or rejected, save the certificates in a simple library format, and show the results in a small web app so I can explore the pipeline from start to finish.

Please make it include a full runnable demo for a LunarLander style environment, plus a way to run tests and regenerate the demo outputs. If there’s a trained model checkpoint missing, fall back gracefully so the project still runs. I’d also like the code to support a basic replay or reuse check after the environment changes mid run, so I can see that certified skills can still be reused safely. If you need current docs for any library, look them up online as you go.