diff --git a/docs/index.html b/docs/index.html index 02f9b2a..6523d99 100644 --- a/docs/index.html +++ b/docs/index.html @@ -171,6 +171,8 @@

skval vs. skill-creator

Benchmark

skval's deterministic structural scan over 10 widely-used skills — 8 ship as-is; the two flagged hit the size budget. + Plus runs over 75 installed skills + and 69 from the web. See the full benchmark →

@@ -185,9 +187,13 @@

Benchmark

SkillTypeScoreFindings
skill-creatorfile_transform96 / ASKILL.md ~8246 tokens (>5000)
docxfile_transform92 / A599 lines & ~5142 tokens (over budget)
-
-

docx · 92 → 100

Applied skval's finding — moved the XML Reference section to references/ (progressive disclosure). skval's own compare.py: +8.

-

bad-skill · 73 → 100

Fixed four findings (kebab name, drop <>, broken ref, stray key). compare.py: +27, Revise → Ship.

+

The upgrade: score → fix → re-score

+

skval doesn't just grade — its ranked findings drive the fix. The biggest real-world turnarounds:

+
+

web skill · invalid frontmatter · 50100

+

A real published skill whose unquoted colon broke its YAML — it used to crash skval, and now scores. One-line fix: quote the description. compare.py: +50, Revise → Ship. case study →

+

bad-skill · 73100

+

Four findings fixed one-for-one — kebab name, drop <>, broken ref, stray key. compare.py: +27, Revise → Ship. case study →