Release v0.1.0

First tagged release of the skillme catalog — 132 skills, each with an SKILL.md and a live-testable evals/evals.json, run by this repo's own eval runner (smeval).

New this release

• skill-inspector — reviews a third-party AI agent skill for safety before you install it, using NVIDIA's SkillSpector static scanner plus a source-aware semantic review pass (score and verdict are different axes; text addressing the reviewer is itself the finding, not reassurance).
• 6 web-quality skills — web-quality-audit, core-web-vitals, performance, seo, accessibility, best-practices — covering the same four categories Lighthouse audits, each with its own live-verified eval cases.
• skill-docs pages for all 7 new skills, following this catalog's human-facing docs template (what it does, when to reach for it, common questions, where it fits).
• README: documented installing a single skill via skills.sh --skill <name> instead of the whole catalog; the Claude Code plugin route stays all-or-nothing.

Fixed this release

• Completed a full live-test sweep of the original 125-skill catalog: every partial failure was individually diagnosed against the model's actual saved output. 41 turned out to be eval-authoring bugs (narrow phrasing, markdown/formatting breaking a literal match, grading a chat response when the model wrote the real answer to a file) — none were genuine skill-content defects.
• Found and fixed a real bug in smeval itself: a case could escape its isolated per-run workspace and write files into this actual repo. Root-caused and fixed in internal/engine and cmd/smeval, verified closed by re-running the case live.
• Consolidated all smeval run output under smeval-workspace/runs/ instead of scattering a <skill>-workspace directory next to every skill in skills/.
• Normalized every skill's metadata.version to 0.1.0 (7 had drifted).