skill-creator

Automate OpenCode skill creation by orchestrating the lifecycle from skeleton SKILL.md generation to eval scaffolding, running an evaluation loop with train/test splits, live reporting, and benchmark aggregation to drive iterative improvements.

The evaluation loop works by scaffolding eval sets, applying train/test splits, aggregating benchmarks, and generating live reports to compare multiple iterations and converge on reliable skill activation and behavior.

The best way to scaffold evals is using the provided templates and scripts to design, run, and package OpenCode skills, enabling a structured workflow for discovering, activating, and evaluating skills with benchmarks and reports.

Yes, you need the pyyaml dependency to run the scripts and tooling that generate skeleton SKILL.md files, scaffold evals, and execute the evaluation loop for building and refining OpenCode skills.

Yes, you can compare multiple iterations by running the benchmark aggregation and live reporting features, which evaluate different versions against test sets to help you converge on consistent, repeatable skill behavior.

Limitations include its strict focus on the OpenCode platform ecosystem, meaning it orchestrates SKILL.md generation and evals specifically for OpenCode skills rather than providing general software testing or generic CLI automation capabilities.