Setup context
The first path is:
uv tool install skillroll
skillroll init --skills-path <skills-folder> --yes
skillroll validate --allThen configure an OpenAI-compatible endpoint and export the named API-key variable. Check the connection before running evals:
skillroll doctor
skillroll eval --allinit writes configuration and optional starter cases. It does not read a key,
contact a model, or spend inference. validate is also offline. Missing
skillroll.toml and a missing key are expected before these steps.
Interactive init can record any OpenAI-compatible endpoint and defaults to
leaving inference unset. --yes suppresses questions and keeps inference
unset unless settings are supplied. With no explicit --skills-path, it
accepts the detected folder. An explicit path selects that folder. For scripts, --openrouter-free can configure OpenRouter's changing
free route during init for a setup or pipeline sanity check. init only
writes configuration, and that route is not skill-quality or release evidence.
Use a named model for ongoing evals.
A generated GitHub workflow is optional and uses an explicit released Action reference.