docs: more web skills + score->fix->re-score case studies

docs/examples/skill-benchmark/README.md

@@ -52,13 +52,12 @@ clean**, the rest splitting into vendored-duplicate `SKILL.md` (e.g. [nextjs](ra
 
 ## From the web — real published skills
 
-Beyond installed plugins, **[from-the-web.md](from-the-web.md)** scores **69 community
-skills** pulled from three public GitHub collections (every one safety-scanned as inert
-text first — 0 tripped the gate, none executed). It surfaced a real bug class —
-**invalid YAML frontmatter** (an unquoted colon in `description`) — that used to *crash*
-skval; skval now scores it instead, and the
-[50 → 100 case study](from-the-web.md#case-study--invalid-frontmatter-50--100) shows the
-one-line fix.
+Beyond installed plugins, **[from-the-web.md](from-the-web.md)** scores **123 community
+skills** pulled from four public GitHub collections (every one safety-scanned as inert
+text first — 0 tripped the gate, none executed). It surfaced real, fixable classes —
+missing/invalid frontmatter and `<>` in the description — and several
+[score → fix → re-score case studies](from-the-web.md#case-studies--score--fix--re-score),
+the biggest being **`daydream` 38 → 100** (no frontmatter → Ship).
 
 ## Case study 1 — `docx`: 92 → 100 (real skill)
 

docs/examples/skill-benchmark/from-the-web.md

@@ -1,14 +1,14 @@
 # From the web — skval over real published skills
 
-A real `skval` **structural** run over **69 community skills** pulled from three public
+A real `skval` **structural** run over **123 community skills** pulled from four public
 GitHub collections — [alirezarezvani/claude-skills](https://github.com/alirezarezvani/claude-skills),
-[glebis/claude-skills](https://github.com/glebis/claude-skills), and
-[smartnews/claude-skills](https://github.com/smartnews/claude-skills) — deterministic, no
-model calls. Most are well-built; this page is about the ones skval flags, and what fixing
-them looks like.
+[glebis/claude-skills](https://github.com/glebis/claude-skills),
+[smartnews/claude-skills](https://github.com/smartnews/claude-skills), and
+[anthropics/skills](https://github.com/anthropics/skills) — deterministic, no model calls.
+Most are well-built; this page is about the ones skval flags, and what fixing them looks like.
 
 > **Safety first.** Every fetched skill was treated as **inert text** and run through skval's
-> D6 safety gate *before* anything else. **0 / 69** tripped it, and none were ever executed —
+> D6 safety gate *before* anything else. **0 / 123** tripped it, and none were ever executed —
 > only the `SKILL.md` text was read and statically analysed.
 >
 > *(Symlink/pointer entries — where the raw file is just a path to another location — were
@@ -33,20 +33,28 @@ finding and scores the skill. Pinned in
 [`tests/test_precision.py`](../../../tests/test_precision.py)
 (`test_invalid_yaml_frontmatter_scores_not_crashes`).
 
-## Case study — `annotate`: 50 → 100
+## Case studies — score → fix → re-score
 
-[`annotate`](https://github.com/glebis/claude-skills) (from glebis/claude-skills) — a real
-published skill whose `description` has an unquoted colon, scored verbatim:
+Real published skills, scored verbatim, then fixed against skval's ranked findings and
+re-scored. Every "after" is the actual re-score of the fixed skill (no remaining findings).
 
-| | Score | Grade | Verdict | Finding |
-|---|------:|-------|---------|---------|
-| **Before** ([scorecard](improved/web-frontmatter.before.md)) | 50 | D | Revise | invalid YAML frontmatter |
-| **After** ([scorecard](improved/web-frontmatter.after.md)) | 100 | A | Ship | — |
+| Skill | Source | Finding class | Before | After | Δ |
+|-------|--------|---------------|:------:|:-----:|:--:|
+| [`daydream`](https://github.com/glebis/claude-skills) | glebis | no frontmatter at all | [38 / F / Reject](improved/web-daydream.before.md) | [100 / A / Ship](improved/web-daydream.after.md) | **+62** |
+| [`disk-cleanup`](https://github.com/glebis/claude-skills) | glebis | invalid YAML frontmatter | [50 / D / Revise](improved/web-disk-cleanup.before.md) | [100 / A / Ship](improved/web-disk-cleanup.after.md) | **+50** |
+| [`annotate`](https://github.com/glebis/claude-skills) | glebis | invalid YAML frontmatter | [50 / D / Revise](improved/web-frontmatter.before.md) | [100 / A / Ship](improved/web-frontmatter.after.md) | **+50** |
+| [`rehydrate`](https://github.com/glebis/claude-skills) | glebis | `<` / `>` in description | [92 / A / Ship](improved/web-rehydrate.before.md) | [100 / A / Ship](improved/web-rehydrate.after.md) | **+8** |
 
-The fix skval points to — **quote the description** — is one line. skval's own diff
-([`improved/web-frontmatter.compare.json`](improved/web-frontmatter.compare.json)):
-`overall_delta: +50`, verdict `Revise → Ship`. What changed:
-[`improved/web-frontmatter.change.md`](improved/web-frontmatter.change.md).
+**The fixes are tiny, one each:**
+
+- **`daydream`** had *no YAML frontmatter* — no `name`, no `description`, so it can't trigger.
+  Adding a frontmatter block takes it from **38 / F / Reject → 100 / A / Ship** (skval's biggest
+  turnaround in the set, `overall_delta +62`). What changed:
+  [`web-frontmatter.change.md`](improved/web-frontmatter.change.md) shows the same class on `annotate`.
+- **`disk-cleanup` / `annotate`** ship an unquoted colon in `description` (`… Local-only: synthetic …`)
+  that breaks the YAML — quote the description and both go **50 → 100**.
+- **`rehydrate`** has `<` / `>` in `description` (breaks parsers / triggering) — drop the angle
+  brackets, **92 → 100**.
 
 ---
 

docs/examples/skill-benchmark/improved/web-daydream.after.json

@@ -0,0 +1,112 @@
+{
+  "score": 100,
+  "grade": "A",
+  "verdict": "Ship",
+  "safety_pass": true,
+  "dimensions": {
+    "D1": {
+      "score": 1.0,
+      "weight": 0.15,
+      "checks": [
+        {
+          "id": "frontmatter_present",
+          "passed": true,
+          "severity": "critical",
+          "detail": ""
+        },
+        {
+          "id": "frontmatter_valid_yaml",
+          "passed": true,
+          "severity": "critical",
+          "detail": ""
+        },
+        {
+          "id": "name_present",
+          "passed": true,
+          "severity": "critical",
+          "detail": ""
+        },
+        {
+          "id": "name_kebab_case",
+          "passed": true,
+          "severity": "major",
+          "detail": ""
+        },
+        {
+          "id": "description_present",
+          "passed": true,
+          "severity": "critical",
+          "detail": ""
+        },
+        {
+          "id": "description_len",
+          "passed": true,
+          "severity": "major",
+          "detail": ""
+        },
+        {
+          "id": "description_no_angle_brackets",
+          "passed": true,
+          "severity": "major",
+          "detail": ""
+        },
+        {
+          "id": "allowed_frontmatter_keys",
+          "passed": true,
+          "severity": "minor",
+          "detail": ""
+        },
+        {
+          "id": "single_skill_md",
+          "passed": true,
+          "severity": "critical",
+          "detail": ""
+        },
+        {
+          "id": "skill_md_line_budget",
+          "passed": true,
+          "severity": "minor",
+          "detail": ""
+        },
+        {
+          "id": "token_budget",
+          "passed": true,
+          "severity": "minor",
+          "detail": ""
+        },
+        {
+          "id": "no_broken_local_refs",
+          "passed": true,
+          "severity": "major",
+          "detail": ""
+        }
+      ]
+    }
+  },
+  "safety": {
+    "safety_pass": true,
+    "findings": []
+  },
+  "findings": [],
+  "provenance": {
+    "source": "github.com/glebis/claude-skills \u00b7 daydream",
+    "kind": "dir",
+    "resolved_path": "github.com/glebis/claude-skills \u00b7 daydream"
+  },
+  "metadata": {
+    "mode": "structural-only",
+    "skill_name": "daydream",
+    "classification": {
+      "type": "task",
+      "confidence": "medium",
+      "also": []
+    },
+    "timestamp": "2026-06-26T15:59:31Z",
+    "skipped_dimensions": [
+      "D2",
+      "D3",
+      "D4",
+      "D5"
+    ]
+  }
+}

docs/examples/skill-benchmark/improved/web-daydream.after.md

@@ -0,0 +1,14 @@
+# Skill Scorecard — `daydream`
+Source: github.com/glebis/claude-skills · daydream · mode: structural-only
+Type: task (confidence: medium)
+
+## ██████████████  100 / 100   Grade: A   Verdict: Ship
+
+| Dimension | Score | Weight | Detail |
+|-----------|-------|--------|--------|
+| D1 Structural | 1.00 | 0.15 | all checks pass |
+
+Safety gate: PASS
+Skipped dimensions (structural-only): D2, D3, D4, D5
+
+No findings — clean.

docs/examples/skill-benchmark/improved/web-daydream.before.json

@@ -0,0 +1,143 @@
+{
+  "score": 38,
+  "grade": "F",
+  "verdict": "Reject",
+  "safety_pass": true,
+  "dimensions": {
+    "D1": {
+      "score": 0.3846,
+      "weight": 0.15,
+      "checks": [
+        {
+          "id": "frontmatter_present",
+          "passed": false,
+          "severity": "critical",
+          "detail": "missing leading '---'"
+        },
+        {
+          "id": "frontmatter_valid_yaml",
+          "passed": false,
+          "severity": "critical",
+          "detail": "no YAML frontmatter found"
+        },
+        {
+          "id": "name_present",
+          "passed": false,
+          "severity": "critical",
+          "detail": "missing 'name'"
+        },
+        {
+          "id": "name_kebab_case",
+          "passed": false,
+          "severity": "major",
+          "detail": "name '' must be kebab-case, <=64 chars"
+        },
+        {
+          "id": "description_present",
+          "passed": false,
+          "severity": "critical",
+          "detail": "missing 'description'"
+        },
+        {
+          "id": "description_len",
+          "passed": false,
+          "severity": "major",
+          "detail": "description is 0 chars (max 1024)"
+        },
+        {
+          "id": "description_no_angle_brackets",
+          "passed": true,
+          "severity": "major",
+          "detail": ""
+        },
+        {
+          "id": "allowed_frontmatter_keys",
+          "passed": true,
+          "severity": "minor",
+          "detail": ""
+        },
+        {
+          "id": "single_skill_md",
+          "passed": true,
+          "severity": "critical",
+          "detail": ""
+        },
+        {
+          "id": "skill_md_line_budget",
+          "passed": true,
+          "severity": "minor",
+          "detail": ""
+        },
+        {
+          "id": "token_budget",
+          "passed": true,
+          "severity": "minor",
+          "detail": ""
+        },
+        {
+          "id": "no_broken_local_refs",
+          "passed": true,
+          "severity": "major",
+          "detail": ""
+        }
+      ]
+    }
+  },
+  "safety": {
+    "safety_pass": true,
+    "findings": []
+  },
+  "findings": [
+    {
+      "dimension": "D1",
+      "impact_estimate": 3,
+      "message": "missing leading '---'"
+    },
+    {
+      "dimension": "D1",
+      "impact_estimate": 3,
+      "message": "no YAML frontmatter found"
+    },
+    {
+      "dimension": "D1",
+      "impact_estimate": 3,
+      "message": "missing 'name'"
+    },
+    {
+      "dimension": "D1",
+      "impact_estimate": 3,
+      "message": "missing 'description'"
+    },
+    {
+      "dimension": "D1",
+      "impact_estimate": 2,
+      "message": "name '' must be kebab-case, <=64 chars"
+    },
+    {
+      "dimension": "D1",
+      "impact_estimate": 2,
+      "message": "description is 0 chars (max 1024)"
+    }
+  ],
+  "provenance": {
+    "source": "github.com/glebis/claude-skills \u00b7 daydream",
+    "kind": "dir",
+    "resolved_path": "github.com/glebis/claude-skills \u00b7 daydream"
+  },
+  "metadata": {
+    "mode": "structural-only",
+    "skill_name": null,
+    "classification": {
+      "type": "task",
+      "confidence": "medium",
+      "also": []
+    },
+    "timestamp": "2026-06-26T15:59:31Z",
+    "skipped_dimensions": [
+      "D2",
+      "D3",
+      "D4",
+      "D5"
+    ]
+  }
+}

docs/examples/skill-benchmark/improved/web-daydream.before.md

@@ -0,0 +1,20 @@
+# Skill Scorecard — `claude-skills · daydream`
+Source: github.com/glebis/claude-skills · daydream · mode: structural-only
+Type: task (confidence: medium)
+
+## █████░░░░░░░░░  38 / 100   Grade: F   Verdict: Reject
+
+| Dimension | Score | Weight | Detail |
+|-----------|-------|--------|--------|
+| D1 Structural | 0.38 | 0.15 | 6/12 checks failed |
+
+Safety gate: PASS
+Skipped dimensions (structural-only): D2, D3, D4, D5
+
+## Top findings (ranked by score impact)
+1. [D1] missing leading '---'
+2. [D1] no YAML frontmatter found
+3. [D1] missing 'name'
+4. [D1] missing 'description'
+5. [D1] name '' must be kebab-case, <=64 chars
+6. [D1] description is 0 chars (max 1024)

docs/examples/skill-benchmark/improved/web-daydream.compare.json

@@ -0,0 +1,11 @@
+{
+  "overall_delta": 62,
+  "dimensions": {
+    "D1": {
+      "a": 0.3846,
+      "b": 1.0,
+      "delta": 0.6154
+    }
+  },
+  "verdict_change": "Reject -> Ship"
+}