reasonscape

Run `npx skills add the-crypt-keeper/reasonscape --all -g -y` in your terminal to install all skills in this suite globally.

It measures how reasoning models process information using parametric test generation, tracking accuracy, token cost, truncation, and reasoning quality instead of a single benchmark score.

Use the context-clip skill to create context-limited copies of an evaluation, such as 8k or 12k versions, then compare performance across them with the analysis commands.

Use the import skill, which finds your result folders, fetches model metadata, builds the cohort entry, and verifies the import before evaluation.

Yes. You can download the public r12 dataset covering 67 models and over 116,000 evaluation points, then explore it with the built-in leaderboard and analysis tools.