Keep the evidence together
- corpus.json
- Repository selection and source commits.
- report.json / report.md
- Static scan results in structured and readable formats.
- jev-risk.json
- Judgments, probabilities, locations, and usage.
- Your review notes
- Human decisions with reasons and evidence links.
Explain the scope
State when the sample was collected, how repositories were selected, what was inspected, and what was excluded. This process reads text; it does not inspect binaries, images, or code fetched later at runtime. An unflagged row is not a safety certification.
How WhichSkills becomes a website
The site generator turns JSON artifacts into tables and links back to the recorded source commits.
The published build also consumes clones.json, a separate provenance dataset. This series does not generate that file or recreate the complete public website. Inspect the released snapshot for that part of the experiment.
Your finished workflow
- A local dashboard loads and displays your skills.
- A repeatable scan narrows a public sample.
- Structured judgments retain uncertainty and evidence.
- Human review adds an explained decision.
- The shared findings include their source versions and limits.
Explore the result
Open WhichSkills to see the published experiment: 18,041 skills in the static pass and 545 unique bodies read by Jev in the September 2026 snapshot.