Public benchmark dashboard and reports for comparing agent review skills.
Dashboard: https://dk-code-factory.github.io/skill-bench-pages/
This repository contains generated benchmark exports. Tooling, source evidence, and skill tuning campaigns are maintained separately in the private Skill Bench repository. Tuning evidence is excluded from public reports, scores, answer-key sources, and embedded dashboard data.
- In the private Skill Bench checkout, run
make bench-publishandmake bench-publish-check. These render completed, classified benchmark snapshots without making model calls. - Run
make bench-export OUTPUT=/path/to/skill-bench-pages/docs. The exporter copies only its generated public inventory and removes superseded exports. - Commit and push to
main. GitHub Pages deploys frommainat/docs.
Cohort-local results remain distinct from the reviewed, sealed canonical answer key. Selecting a skill through tuning requires a separate benchmark evaluation; tuning trials are never promoted.