/score — Article scorer (A–G, dual verdict)

Target: $ARGUMENTS

The canonical article-level scorer of this plugin. /optimize uses it for before/after, /write-seo uses it as the final self-check. Weights, verdict tables, and every numeric target come from ${CLAUDE_PLUGIN_ROOT}/skills/seo-fundamentals/references/scoring-spec.md; per-criterion anchors from ${CLAUDE_SKILL_DIR}/references/criteria-details.md. Read both before scoring. Follow the ground-truth rules in ${CLAUDE_PLUGIN_ROOT}/skills/seo-fundamentals/SKILL.md.

1. Inputs

Ask only for what is missing and cannot be inferred:

1. Article (URL, file, or paste).
2. Primary keyword — infer from title/H1/URL if not given and state the assumption.
3. Language/locale — infer from the text.
4. Domain Rating (if known; affects A5) and whether other language versions exist (affects C7/F1).

2. Measure before judging

• URL: run python3 ${CLAUDE_PLUGIN_ROOT}/scripts/fetch_page.py "<url>" --json → title, meta, canonical, hreflang, OG, JSON-LD blocks, headings, links, images, body text. Then run python3 ${CLAUDE_PLUGIN_ROOT}/scripts/text_metrics.py --from-json <output> --keyword "<kw>".
• File or paste: save to a temp file and run text_metrics.py <file> --keyword "<kw>".
• The scripts return word count, sentences, Flesch / FK grade / Fog, passive ratio, keyword density and placements, words-per-heading, and citable-passage candidates per 1,000 words. Use those numbers; do not estimate them.
• If a script cannot run (no network, no Python), say so and mark the affected criteria "unverified".
• For A2/B2/G3 you need the SERP: WebSearch the primary keyword, note the top 5–10 content types and depth; for G3 build or reuse an /entity-coverage matrix when the user wants a rigorous information-gain check (otherwise mark G3 as an estimate).

3. Score

Score all seven categories A–G using the anchors. Be strict: 2 means genuinely met. When a criterion cannot be verified (no GSC, no CWV data), score what is visible and mark the rest "unverified — verify via {tool}". Never invent SERP or DR data.

Additional analysis modules to report alongside the score (they feed A, B, C but are not extra points):

• Intent alignment — target keyword, intent class, does the content type match the top-5 SERP, confidence.
• Length vs SERP — word count vs estimated top-10 median; sections to expand if below.
• Readability — script output vs targets; overall grade.
• Entity salience — G1 proxy checks (title, H1, first 50 words, sentence-subject count).

4. Output (in the article's language)

Use ${CLAUDE_SKILL_DIR}/references/report-template.md. Sections, in order:

1. Header (title, URL, language, date, primary keyword, DR).
2. Category tables A–G (#, criterion, max, score, evidence/notes) with subtotals.
3. Overall scorecard (A–G, max, score, %).
4. Dual verdict — overall (✅🟡🟠🔴⛔) and AI Overviews readiness (G/15 normalized: AI-READY / PARTIAL / WEAK / OPAQUE). Say explicitly when they diverge.
5. Priority fixes — categories under 60% of max; top 3–5 criteria with concrete action, effort, and re-check date. Put G fixes first whenever AI Overviews readiness is below PARTIAL.
6. Pre-publish checklists — classical TOP 10 and entity/AI TOP 7 (in the template), ✅/❌ each.
7. Locale notes when relevant (${CLAUDE_SKILL_DIR}/references/locale-notes.md).

5. Follow-ups to offer

• /citable-passage — passage-level deep dive when G6 < 2.
• /entity-coverage — build the matrix behind G2/G3.
• /alignment-audit — when G4 < 2 or schema was detected.
• /optimize — apply the fixes and re-score.