v2.0.0 — a skill that runs

2.0.0 — 2026-08-15

The release that stops this being only a prompt.

It runs now

• tools/seo-audit.mjs — a real audit against a live origin or a local
  build. Zero dependencies, Node 18+. Checks robots.txt against RFC 9309 group
  semantics, sitemap traversal and limits, lastmod credibility, host
  consolidation, per-template soft 404s, canonical self-reference, hreflang
  reciprocity across pages, duplicate titles, streamed-metadata visibility,
  raw-HTML content volume, JSON-LD validity, and image dimensions. Emits a
  computed score, Markdown and JSON, and exits 1 on any P1 so it can gate a
  deploy.
• tools/seo-smoke.sh — the ten-second deploy tripwire: production
  noindex, robots.txt sanity, canonical and JSON-LD still rendering, and 404s
  per template via SEO_SMOKE_404_PATHS.
• tools/README.md — what each one covers, and the three things they
  deliberately cannot see.

Both were tested against five production sites before release. The audit tool
found two live P1s (a Disallow: /_next/ blocking render, and a template
soft-404ing an entire section) that a source-only review had not surfaced.

It installs as a skill

install.sh drops the knowledge base into a target project and writes the entry
point each agent actually reads:

Agent	File
Claude Code	.claude/skills/seo-audit/SKILL.md
Codex, Amp, Jules	AGENTS.md
Cursor	.cursor/rules/seo-prompt-master.mdc
Gemini CLI	GEMINI.md
GitHub Copilot	.github/copilot-instructions.md

Existing AGENTS.md / GEMINI.md files are appended to rather than replaced,
and re-running is safe.

The knowledge base grew where it was thinnest

Six new documents, taking it from 11 to 17. The previous docs covered what to
put on a page; these cover the mechanisms around it:

• docs/12 Crawling, robots.txt & crawl budget — RFC 9309 semantics,
  response-code behaviour, the crawler table, capacity vs demand, log analysis.
• docs/13 Indexing, canonicals & duplicates — which instrument for which
  job, the duplicate taxonomy, faceted navigation, soft 404s, reading GSC index
  states, index bloat.
• docs/14 Content quality, E-E-A-T & spam policies — all 16 spam policies,
  the line on AI-assisted content, the directory/UGC threshold rule, manual
  action recovery.
• docs/15 Measurement & verification — Search Console reports, BigQuery,
  the KPI set, and the traffic-drop diagnosis flow.
• docs/16 Migrations & incident response — the five-source URL inventory,
  redirect rules, rollback criteria, hack response.
• docs/17 Off-page & entity authority — link quality, digital PR, disavow
  as a last resort, entity building.

Fixed

• The <title> and canonical of any page whose framework streams metadata
  (Next.js 15+ and friends) were previously invisible to head-only parsing. The
  audit tool now reads the whole document and reports the streaming itself as a
  finding, since bots that stop at </head> genuinely do see an untitled page.
• hreflang validation is case-insensitive, per Google. zh-tw is not an error.