This project keeps a human-curated changelog so users and contributors can see how Snowl evolves over time.
Format:
Added: new features and capabilitiesChanged: behavior changes, refactors, and structural improvementsFixed: bug fixesDeprecated: old paths or APIs that will be removed laterRemoved: deleted featuresBreaking: upgrade notes for incompatible changes
- Plugin discovery for
BenchmarkRegistry— benchmarks fromsnowl-evalsare now automatically discovered viaimportlib.metadata.entry_points(group="snowl.benchmarks"). - Re-exported
JudgeClientFactoryfromsnowl.scorerpublic API. - Solver pipelines, framework adapters, and bridges subpackages.
- New benchmark adapters: cybench, cybergym, gaia, humaneval, math, swe_bench, tau_bench, webarena.
- CI workflows for model evaluation and PyPI publishing.
- CONTRIBUTING.md with development workflow and PR checklist.
- Tightened architecture boundary tests for core layer isolation.