v1.9.0 — SEO audit hardening and drift checks

Changelog

[Unreleased]

[1.9.0] - 2026-05-28

Added

• Shared URL safety layer — scripts/url_safety.py centralizes outbound URL validation for audit fetches, blocking private/internal IPs, loopback, link-local ranges, obfuscated IPv4 forms, URL credentials, and unsafe redirect targets.
• SPA-aware rendering — scripts/render_page.py adds optional Playwright rendering, and fetch_page.py / generate_report.py now support --render never|auto|always for JavaScript-heavy sites.
• Content quality scanner — scripts/content_quality.py detects deterministic E-E-A-T risks including filler phrasing, citation gaps around statistics/claims, missing author signals, and missing dates.
• SEO drift monitor — scripts/drift_monitor.py adds SQLite-backed baseline, compare, and history snapshots for SEO-critical elements like title, meta description, canonical, robots, H1, schema hash, headings hash, and internal link count.
• Regression tests — Added focused unittest coverage for URL safety, render detection, redirect safety, recommendation metadata, content-quality scoring, and drift snapshot extraction.

Changed

• scripts/crawl_adapter.py — Routes requests, Firecrawl, and Playwright fetching through the shared URL safety layer and preserves rendered-page metadata.
• scripts/generate_report.py — Adds content quality into the report pipeline and renders falsifiable recommendation metadata: dependency, failure check, and leading indicator.
• Plugin bundle — Root scripts and plugin-bundled scripts now include the new safety, rendering, content-quality, and drift-monitoring tools.

[1.8.5] - 2026-04-11

Added

• scripts/generate_report.py — --crawl-deep, --crawl-max-pages, --crawl-depth to run multi-page broken_links and canonical_checker crawls (capped; longer script timeouts).
• scripts/parse_html.py — og:description fallback with meta_description_source; duplicate <title>, duplicate meta name="description", and duplicate rel=canonical findings; multi-H1 info note; canonical rel parsing when rel is a list.
• HTML report — Raw-HTML disclaimer for Next.js/Nuxt/JS-heavy stacks; on-page canonical filled from canonical audit when HTML has no <link rel="canonical">; labels for og-description fallback and audit-filled canonical.

Changed

• scripts/canonical_checker.py — Treat canonical sent only via HTTP Link: ...; rel="canonical" as present (with warning to add HTML <link> for parity).
• AGENTS.md — GEO vs traditional robots.txt: scope “remove Disallow” to AI crawlers; do not recommend removing Googlebot facet/category disallows unless the user asks for crawl-budget review.
• SKILL.md — Routing row for content relevance + GEO (procedures 03 + 06, article_seo, readability, internal_links).
• README.md, .claude-plugin/marketplace.json, plugin.json — Audience, default vs deep crawl scope, marketplace descriptions.

[1.8.4] - 2026-04-06

Added

• references/optional-extensions-mcp.md — Optional Firecrawl / DataForSEO MCP setup for plugin installs that do not include the monorepo extensions/ tree; AGENTS.md Extensions row now points here.
• scripts/generate_report.py --format pdf — PDF export via optional WeasyPrint (pip install weasyprint; OS libraries may be required). Documented in references/procedures/21-script-toolbox.md and AGENTS.md with HTML + browser print-to-PDF fallback.
• Plugin bundle: agents/ — setup-plugin.sh now copies agents/*.md into plugins/.../skills/.../agents/ so agents/PARALLEL-AUDIT.md resolves in Claude Code installs. scripts/check-plugin-sync.py validates root vs plugin agents/ parity.
• requirements.txt — Comment block for optional weasyprint.

Changed

• extensions/README.md — Points to references/optional-extensions-mcp.md for the skill/plugin copy.
• .github/workflows/validate-plugin.yml — Triggers on agents/** changes.
• Marketplace / plugin.json — Descriptions and keywords (mcp, pdf, parallel-agents) updated for the gaps above.

[1.8.3] - 2026-04-03

Changed

• SKILL.md — Refactored into a routing shell (~230 lines): §0, global guardrails, and an index to references/procedures/*.md. Detailed procedures for former §1–§21 moved verbatim into references/procedures/ (22 files + README.md) for progressive disclosure and lower default context size when the host loads the skill.
• AGENTS.md, GEMINI.md, .github/copilot-instructions.md — Pointers updated to references/procedures/ instead of monolithic SKILL.md sections.
• agents/ — Replaced six per-agent markdown files with agents/PARALLEL-AUDIT.md + shorter agents/README.md.
• chatgpt/copy-knowledge-files.sh — Also copies references/procedures/*.md into chatgpt/knowledge/procedures/ for Custom GPT uploads.
• Version — 1.8.3 in SKILL.md, AGENTS.md, plugin.json, and marketplace catalog.

[1.8.2] - 2026-04-03

Fixed

• scripts/robots_checker.py — Added OAI-SearchBot (ChatGPT Search indexing crawler) to AI_CRAWLERS list with inline comment distinguishing it from GPTBot (training-only). Every GEO audit that checks robots.txt now correctly detects whether ChatGPT Search can index the site. Previously, a site blocking OAI-SearchBot would pass the AI crawler check undetected.
• scripts/pagespeed.py — Critical: the entire response-processing block (performance score, CrUX field data, opportunities, diagnostics — ~100 lines) was inside the for retry loop body after the break statement, making it completely unreachable. The script fetched the PSI API successfully but always returned empty metrics {}. Fixed by initializing data = None before the loop and moving extraction to function scope after the loop. CrUX distribution buckets (% good / needs-improvement / poor per metric) and a "source": "field"/"lab" tag are now also surfaced in the output.
• scripts/hreflang_checker.py — Stale internal path in docstring (resources/skills/seo-hreflang.md → references/international-seo.md).
• scripts/image_checker.py — Expanded from alt-text-only to full image SEO check: added fetchpriority="high" detection on first/LCP image (critical severity if missing on lazy-loaded hero), srcset/sizes coverage, width/height dimension attributes (CLS prevention), and WebP format detection. Also fixed missing-alt finding not being emitted (counter populated but issues.append dropped in prior refactor).
• scripts/run_individual_checks.sh — Added canonical_checker.py, site_mapper.py, and programmatic_seo_auditor.py which existed but were missing from the runner.
• references/audit-script-matrix.md — Added site_mapper.py, crawl_adapter.py, and backlink_analyzer.py to the Utilities table; all three are user-facing tools mentioned in AGENTS.md but absent from the matrix.
• scripts/backlink_analyzer.py — Documented CSV column mapping in docstring (field name variants for Ahrefs/Moz/Semrush exports were handled by normalize_backlinks but never documented).
• AGENTS.md — Version synced to match SKILL.md (was 1.8.0 while SKILL.md was already 1.8.1).

[1.8.1] - 2026-04-01

Added

• Functional page exemption — Sign up, sign in, log in, register, create account, forgot/reset password, membership enroll, checkout, cart, account dashboard, profile settings pages are now explicitly excluded from thin-content checks and word-count floors. Audit rule table added to SKILL.md § 6 and AGENTS.md § 6. Relevant checks for these pages: title accuracy, meta description, form labels, trust signals, schema.
• Eval 15 — Regression test: auditing /signup and /login pages must not flag thin content or recommend adding copy. 3 assertions.

Changed

• SKILL.md § 6 — Added "Functional Page Exemption" table before the Content Audit step-by-step. Classifies functional pages (task UI), landing/marketing pages, and content pages with clear audit rules per type.
• AGENTS.md § 6 — Prepended functional page exemption notice with page-type list and applicable checks.

[1.8.0] - 2026-04-01

Added

• scripts/backlink_analyzer.py — 7-section backlink analysis with CSV/API data adapter. Profile overview, anchor text distribution, referring domain quality, toxic link detection (22 patterns, 2 risk tiers), top pages, competitor gap analysis, and velocity placeholder. Backlink Health Score (0-100) with weighted composite. Supports --source csv|gsc|sample.
• scripts/crawl_adapter.py — Pluggable crawl backend abstraction. Supports requests (default, stdlib), firecrawl (optional, via API key), and playwright (optional, local install). Auto-detection selects best available backend. Used by other scripts via import.
• scripts/site_mapper.py — Fast URL discovery via sitemap parsing + BFS internal link crawl. Supports --max-pages, --depth, --include-status. Uses crawl_adapter.py for all fetching.
• references/backlink-quality.md — 30 toxic link patterns, anchor text benchmarks, health score formula, disavow file format, competitor gap analysis methodology.
• generate_report.py --format xlsx|all — Excel export with openpyxl. Sheets: Summary, Issues, Links, Technical. Navy header styling, frozen headers, auto-filter, auto-column-width. --format all generates HTML + XLSX together.
• extensions/ directory — Platform-neutral extension architecture. Each extension has extension.json manifest + install adapters for Claude Code, Cursor, and generic (env vars). Two extensions: Firecrawl (JS-rendered crawling) and DataForSEO (live SERP/backlink data).
• agents/ directory — 6 subagent capability definitions (seo-technical, seo-content, seo-schema, seo-geo, seo-performance, seo-links). Platform-neutral markdown format — each platform interprets natively.
• LLM-agnostic agent identity — AGENTS.md, SKILL.md, README.md updated to reflect multi-platform support. Platforms row added to header tables. Descriptions reframed from "skill" to "agent."

Changed

• User-Agent strings — All 20 scripts updated from SEOSkill/* and ClaudeSEO/* to UltimateSEO/* with project homepage URL.
• AGENTS.md routing index — Added backlinks, URL discovery, and extensions rows.
• AGENTS.md script table — Added backlink_analyzer.py, crawl_adapter.py, site_mapper.py. Updated generate_report.py description to mention XLSX.
• AGENTS.md section 21 — Added Excel export usage, extensions table, and subagent definitions table.
• requirements.txt — Added openpyxl>=3.1.0 as commented optional dependency.

[1.7.2] - 2026-03-31

Changed

• SKILL.md — Strengthen robots.txt High-Risk gate: withhold file/code until user confirms. Add context-budget awareness section to § 0 (graceful degradation for fast models). Add Quality Gate criterion 11 for High-Risk deliverable withholding check.
• Evals — 3 new discriminating GEO assertions (eval 12), 5 assertion fixes across evals (regex precision, subtype broadening, hyphenated deindex variant). Total: 14 prompts, 60 assertions.

Fixed

• README — Version badge updated to 1.7.2; eval assertion count corrected to 60.

[1.7.1] - 2026-03-30

Added

• Eval 13 — Execute mode risk gate — Negative test: user requests blocking all crawlers via robots.txt. Verifies the skill warns about deindexing, classifies the change as high-risk, and asks for explicit confirmation before producing the file. 4 assertions.
• Eval 14 — Evaluator-Optimizer fabrication check — User requests CWV scores for example.com with no pagespeed.py data available. Verifies the skill does NOT fabricate LCP/INP/CLS numbers and instead states data is unavailable. Tests §19 criterion 2. 4 assertions.
• Competitive mode output example (references/audit-output-example.md) — Full "External Observation Only" example with no /100 Health Score, showing correct labeling and scope constraints for external site observations.
• GEO-only audit output example (references/audit-output-example.md) — Complete GEO Quick Check table, GEO Score breakdown by dimension, AI citation findings, and before/after citation demonstration for a scoped GEO request.

Fixed

• README version badge — Updated from stale v1.2.1 to current version (was 5 major versions behind).
• README eval counts — Updated from "12 prompts, 49 assertions" to "14 prompts, 57 assertions" in Eval Results section, architecture file tree, and test scenario description.

[1.7.0] - 2026-03-30

Added

• § 3 GEO — RSL 1.0 — Added check for RSL 1.0 (Really Simple Licensing), the December 2025 standard backed by Reddit, Yahoo, Medium, Quora, Cloudflare, Akamai, and Creative Commons. GEO Audit step 2 now includes /rsl.txt and RSL meta tag verification alongside llms.txt.
• § 4 Technical — AI crawler distinctions — Added explicit guidance on Google-Extended (blocks Gemini training only, not Google Search or AI Overviews) and GPTBot vs ChatGPT-User (blocking GPTBot does not block ChatGPT Search citations). Common misconfiguration that silently removes sites from ChatGPT live search.
• § 6 Content — Word count caveat — Clarified that word count minimums are topical coverage floors, not targets. Google confirmed word count is not a direct ranking factor.
• § 6 Content — Google AI Mode — Added to Key Policy Updates table: AI Mode (May 2025, 180+ countries) delivers zero blue links; AI citation is the only visibility mechanism.
• § 8 Competitors — Comparison page title formulas — Added proven title formulas for X vs Y, alternatives, and roundup pages. Added ItemList schema recommendation for roundup pages.
• § 11 Crawl — Sitemap tag note — Added that <priority> and <changefreq> tags are ignored by Google and Bing; omit from new sitemaps.
• § 13 Images — decoding="async" — Added decoding="async" guidance for non-LCP images to prevent image decoding from blocking the main thread.
• § 13 Images — JPEG XL — Added note that Chrome reversed its 2022 removal decision in November 2025 (Rust-based decoder); not yet in stable, ~20% lossless savings over JPEG.
• § 14 Hreflang — 3-method comparison — Added implementation method comparison table (HTML link tags vs HTTP headers vs XML sitemap), sitemap hreflang XML format, and cross-domain setup guidance.
• § 15 Programmatic — Enforcement timeline — Added Scaled Content Abuse enforcement dates: November 2024 site reputation abuse, June 2025 manual actions wave, August 2025 SpamBrain update, 45% low-quality content reduction statistic, progressive rollout rule (50–100 page batches).
• § 21 Scripts — DataForSEO MCP — Added optional DataForSEO MCP integration table covering GEO visibility checks, on-page analysis, competitive intelligence, keyword research, and live SERP data.

[1.6.2] - 2026-03-30

Performance

• generate_report.py: parallel script execution — All analysis scripts (17–20 depending on HTML fetch) now run concurrently via ThreadPoolExecutor (max 8 workers) instead of sequentially. Wall-clock time for a full audit drops from sum-of-all-scripts to max-of-any-script — typically a 3–6× speedup on real sites where each script spends most of its time waiting on network I/O. Profiling showed startup overhead (requests + bs4 imports) was stacking at ~140–180ms per script × 20 scripts = ~3s of dead import time alone.
• entity_checker.py: parallel Wikidata + Wikipedia lookups — check_wikidata() and check_wikipedia() were called serially (each with an 8s timeout). They now run concurrently via ThreadPoolExecutor(max_workers=2), cutting entity check I/O wait time roughly in half.
• score_eval_transcript.py: pre-compiled regex patterns — Regex patterns in evals.json assertions are now compiled once at load time (_compile_assertions) and stored as _compiled on each assertion object. check_assertion uses the pre-compiled object directly, eliminating per-call compilation overhead. This matters most in long-running eval sessions scoring many transcripts.

[1.6.1] - 2026-03-30

Fixed

• CI: check_version_sync.py excluded from plugin bundle sync check — Added check_version_sync.py to SCRIPT_EXCLUDE in scripts/check-plugin-sync.py and to SCRIPT_EXCLUDE_LIST in setup-plugin.sh. This maintainer-only CI script was incorrectly being compared against the plugin bundle, causing the "scripts/*.py list mismatch" CI failure introduced in v1.6.0.

[1.5.6] - 2026-03-27

Added

• Evaluator-Optimizer self-review pass (§19) — Mandatory 7-criterion internal evaluation table Claude runs after generating any Mode 1 audit output, before delivering it: checks Evidence presence on Critical/High findings, no fabricated metrics, Health Score justification, structured format, no duplicates, scope label, and actionable fix directives. Adapted from Anthropic's Evaluator-Optimizer pattern (claude-cookbooks/patterns/agents).
• Progressive Disclosure hard rule (§0) — Explicit "load at most 3 reference files per response" rule with reasoning, linked to Anthropic's Skills progressive disclosure architecture. Prevents unnecessary context bloat on single-topic requests.
• Orchestrator-Workers pattern (§21) — Formal ASCII diagram showing how to delegate independent scripts as worker nodes and synthesize in the main thread. Includes hard constraint: never run generate_report.py + individual scripts for the same URL simultaneously.
• Context compaction guidance (§21) — Step-by-step instructions for managing long audit sessions when context fills: compact findings to one-line format, checkpoint score, continue, merge at end. Adapted from Anthropic's session memory compaction pattern.
• Citation demonstration pattern (§3 GEO) — "Before/After" citation rewrite block added to GEO section. Audits now produce a concrete example of what an AI-quotable passage from the user's content would look like, not just a score. Adapted from Anthropic's Citations cookbook.
• Attribution row for anthropics/claude-cookbooks added to §21 Attribution table.

[1.5.5] - 2026-03-27

Added

• scripts/check_github_release.py — pre-deployment check that verifies the current plugin version has a published GitHub Release on the Marketplace. Reads version from plugin.json, queries GitHub public API, exits 1 if the release is missing or a draft. Prints the exact gh release create fix command. Run after git push to confirm the Marketplace is live.
• CI Marketplace check — validate-plugin.yml now includes a "Check GitHub Marketplace Release" step that runs check_github_release.py --warn on every push. Warns without blocking CI so you're alerted if a GitHub Release was never published.
• RELEASE.md step 6b — new required verification step after git push, plus updated step 6c with the gh release create one-liner.

Changed

• setup-plugin.sh — added check_github_release.py to the maintainer-script exclusion list so it is never bundled into the plugin for end users.
• scripts/check-plugin-sync.py — SCRIPT_EXCLUDE updated to include check_github_release.py.

[1.5.4] - 2026-03-27

Added

• Browser caching / Expires header coverage — technical-checklist.md now includes a dedicated "Browser Caching checklist" block under §6 Core Web Vitals (4 pass/fail items), a new row in the Critical Technical Issues table with detect + fix instructions, and a full ### Fixing Missing Browser Cache Headers section with working Apache .htaccess, NGINX, and WordPress plugin fix code (including the safety note on 1-year cache with versioned filenames).
• Image-level caching audit — image-seo.md gains a ### Browser Caching for Images section (3-row audit table, DevTools quick-check, fix pointers) and a Browser Caching dimension (5pts) in the Image SEO Score audit template.

Changed

• technical-checklist.md — added missing cache headers to the Common LCP Issues list; tightened Cache-Control reference in the LCP fix steps.
• image-seo.md — updated Contents header and Alt Text score weight (30 → 25) to accommodate new Browser Caching dimension while keeping total at 100.

[1.5.3] - 2026-03-26

Changed

• Skill description made "pushier" for better triggering — added explicit "Make sure to use this skill whenever..." phrasing and expanded trigger keywords (site speed, Core Web Vitals, structured data, rich results, indexing issues, search engine visibility). Follows Anthropic's skill-creator guidance that Claude tends to "undertrigger" skills.
• Mode routing converted from table to decision tree — ASCII tree format is easier for the model to follow branching logic vs. scanning table rows. Sourced from Anthropic's webapp-testing and claude-api skill patterns.
• Quality Gates (§19) rewritten with reasoning — every hard rule now explains why it exists (e.g., deprecated schema wastes effort, blocking GPTBot loses live search visibility). Follows Anthropic's skill-creator principle: "explain the why" instead of rigid MUSTs.
• Featured Snippet advice (§7c) softened with reasoning — "Never exceed 60 words" → explains Google truncation; "Never defer the answer" → "Lead with the direct answer" with explanation.

Added

• Script black-box rule (§21) — "Run scripts as black boxes. Always try --help first. Do not read script source unless --help is insufficient." Prevents wasting context tokens on large script files. Pattern from Anthropic's webapp-testing skill.
• Reference Reading Guide table (§0) — consolidated task→file→script mapping near the top of SKILL.md. Allows the model to identify which reference file to load without scanning every section. Pattern from Anthropic's claude-api skill.

Meta

• All 12 evals pass (0 regressions). SKILL.md: 1,070 lines (+38 from structural additions, net from 1,032).

[1.5.2] - 2026-03-26

Changed

• SKILL.md optimized per Anthropic Skill best practices — reduced from 1,167 to 1,031 lines (-136 lines, ~2,300 fewer tokens per request).
  • Removed 7 content blocks duplicated in reference files (On-Page SEO Checklist, audit example, AI Search Landscape table, Brand Signals playbook, AI Crawler table, llms.txt template, Local review benchmarks, Script table + usage blocks). Replaced with 1-2 line pointers to the existing reference files.
  • On-Page SEO Checklist moved to references/technical-checklist.md; audit output example moved to new references/audit-output-example.md.
  • All 18 ambiguous script references now have explicit execution intent verbs ("Run scripts/..." or "See references/...").
  • Standardized "Core Web Vitals" / "CWV" terminology — full name on first use per section, abbreviation after.

Added

• Table of Contents added to 13 reference files over 100 lines (previously 0/15 had TOCs). Ensures Claude sees full scope of available content even on partial file reads.
• New references/audit-output-example.md — 3-finding Greenleaf.io excerpt extracted from SKILL.md.
• Quora, Reddit, and Influencer/Newsletter tactical playbooks added to references/ai-search-geo.md → Tactical Playbooks per Channel.

[1.5.1] - 2026-03-26

Changed

• Routing table (§ 1) — improved coverage and precision.
  • Added disambiguation rule above the table: most specific match wins; first match breaks ties; no-match falls back to § 0 Intake Checklist.
  • Added cross-reference linking § 1 (topic routing) to § 0 Mode Routing to clarify the two-level dispatch.
  • Added section labels to every "Go To" cell (e.g., "§ 3 GEO" instead of bare "§ 3") for self-documenting routing.
  • New Traffic Drop / Rankings Lost row — triggers on "traffic dropped", "lost rankings", "rankings fell", "core update", "algorithm update", "rankings dropped"; routes to § 10 Analytics first, then § 4 / § 6.
  • Expanded GEO trigger keywords: added "SearchGPT", "Gemini", "llms.txt", "AI search".
  • Expanded Technical SEO trigger keywords: added "mobile", "mobile-friendly", "HTTPS", "security headers", "redirect chain".
  • Expanded Analytics / Reporting trigger keywords: added "penalty", "manual action" (removed "traffic drop" — now has its own dedicated row).
  • Expanded Crawl & Indexation trigger keywords: added "duplicate content", "content cannibalization".
  • New No clear match fallback row at bottom of main table.

Added

• Eval 11 — traffic drop routing test: verifies analytics-first response to core update traffic loss; asserts no schema-first or migration routing.
• Eval 12 — GEO platform routing test: verifies GEO-focused response to Gemini/SearchGPT citation request; asserts no health score or full crawl audit.
• All 12 evals pass (0 regressions).

[1.5.0] - 2026-03-26

Added

• Schema Detection Caveat (§2 Evidence Integrity + §5 Schema Audit) — new Evidence Integrity row for "Schema not found on a CMS site" requires confirmation via Rich Results Test or browser JS console before reporting missing schema. Schema Audit step 1 expanded with caveat that web_fetch/curl/raw HTML cannot detect JS-injected schema from plugins like Yoast, RankMath, and AIOSEO.
  WHY: Many CMS sites inject JSON-LD via client-side JavaScript. Static source fetch shows no schema, leading to false "no schema found" findings. This prevents misdiagnosis on WordPress/CMS sites.

• Expanded Competitor & Alternatives Page Playbook (references/link-building.md) — four detailed page-structure templates: [Competitor] Alternative (Singular), [Competitor] Alternatives (Plural), You vs [Competitor], and [Competitor A] vs [Competitor B] (Third-Party). Each template includes URL patterns, target keywords, search intent, and 7-point content outlines. Added centralized competitor data architecture with update cadences and a 5-step competitor research process.
  WHY: The existing playbook listed page types but lacked actionable structure. Practitioners needed concrete outlines for each format to produce consistently structured comparison/alternatives pages at scale.

Changed

• SKILL.md version → 1.5.0; updated date → 2026-03-26.

[1.4.1] - 2026-03-26

Fixed

• Remove "skills", "license", "homepage", and "repository" fields from plugin.json that blocked skill discovery in Claude.ai web app and Claude Code.
• Add version field to SKILL.md YAML frontmatter for Claude Code skill detection.
• Fix Attribution table rendering in Claude.ai web app (plain-text URLs instead of object references).

[1.4.0] - 2026-03-26

Added

• GEO Brand Signal Tactical Playbook (§3) — new "Building Brand Signals" section placed after the GEO Score Components table, directly below the line that surfaces Brand Signals as 20% of the GEO Score.
  WHY: The skill already audited whether Reddit, YouTube, and Quora signals were present, but had no fix directive for building them. This closes the gap between "Finding" and "Fix" — without the playbook, a practitioner seeing a brand-signal gap had nowhere to go next.

• Competitor GEO Stance (§8 Step 2 + Assessment Table) — Step 2 now instructs fetching [competitor-url]/robots.txt and [competitor-url]/llms.txt alongside top-ranking pages. Two new rows added to the Competitor Assessment Dimensions table: "AI crawler configuration" and "llms.txt presence."
  WHY: A competitor blocking OAI-SearchBot or lacking llms.txt is invisible to ChatGPT Search and gives the audited site an immediate GEO first-mover advantage. Previously §8 checked citation presence but not the configuration-level reasons behind it.

• Competitor Sitemap Gap Analysis (§8 Step 2 + Assessment Table) — Step 2 extended to run sitemap_checker.py [competitor-url] for reachability, then fetch raw sitemap XML and read <loc> URL path patterns. New "Topic coverage gap (sitemap)" row added to the Assessment Dimensions table.
  WHY: Topic clusters present in a competitor's sitemap but absent from the audited site are the most reliable, evidence-backed content calendar input. Previously gap identification was inferred from fetched pages only — sitemap analysis makes it systematic.

Changed

• SKILL.md version → 1.4.0; updated date → 2026-03-26.

[1.3.0] - 2026-03-25

Added

• Internal vs. Competitive Mode (§0) — two-mode split enforced before routing. Internal Mode (user's own site) enables full scored audit, all scripts, Execute mode, and /100 Health Score. Competitive Mode (external URL) limits to surface crawl (homepage + up to 20 pages), disables Execute mode, and labels all output "External Observation Only."
  WHY: Prevents confident /100 scores on sites the model has never fully seen; enforces evidence integrity by architecture, not by rule.

• Evidence Integrity Table (§2) — replaces the single CWV-only evidence rule with a six-row table covering LCP/INP/CLS, backlinks, traffic, Health Score, thin content, and schema validation. Each claim now has an explicit data-source precondition. When data is absent the output reads [metric] not measured — run [script] for actual data.
  WHY: The old rule guarded only performance scores. Backlink counts, traffic numbers, and thin content findings were unguarded and could be hallucinated.

• Execute Mode Risk Gate (§2 Mode 3) — classifies every Execute output as Safe (output directly) or High-Risk (state change, ask for confirmation before outputting). High-Risk covers robots.txt, canonical tags, redirect maps, noindex, hreflang, and bulk CMS template changes.
  WHY: One bad robots.txt change can deindex a site. The confirmation step is asymmetric — costs 5 seconds, prevents weeks of recovery.

Changed

• SKILL.md version → 1.3.0; updated date → 2026-03-25.

[1.2.1] - 2026-03-25

Changed

• SKILL.md — Flat YAML frontmatter (name, license, version, updated, description only). Author, links, and upstream credits moved to readable markdown tables so previews no longer nest YAML as awkward “tables in cells.”
• check-plugin-sync.py — Reads top-level version: in skill frontmatter (still accepts legacy indented version under metadata:).
• README — Clearer update instructions (cache git-pull one-liner, full reinstall commands); added Cursor IDE and Claude Desktop install methods; replaced .skill file reference; added Troubleshooting table covering known cache/skill-loading bugs.

Fixed

• plugin.json — Skills auto-discovery now works correctly (removed invalid "skills" field that caused "Plugin not found" errors in some Claude Code versions).
• .gitignore — Added .claude/ to prevent accidental commits of local Claude Code settings.

[1.2.0] - 2026-03-24

Added

• scripts/score_eval_transcript.py — score saved model transcripts against evals/evals.json (--eval-id, --all-fixtures).
• scripts/meta_lengths_checker.py — title, meta description, and H1 length/presence from local HTML or --url.
• evals/fixtures/eval{1–10}_pass.txt — golden transcripts; CI runs score_eval_transcript.py --all-fixtures.
• evals/evals.json — four new scenarios (7–10): news publisher, scoped technical-only, international, pre-launch strategy.
• references/finding-verifier-example.json + finding-verifier-context-example.json — sample input for finding_verifier.py.

Changed

• SKILL.md — §0 routing index; “when not to run Mode 1” table; §1 routing notes; §2 lab/PSI evidence rule + industry preset table; §21 script list + evidence integrity + pip install -r requirements.txt; version 1.2.0.
• README — script/eval counts, score_eval_transcript / meta_lengths_checker, Updating the Claude Code plugin (cache refresh).
• references/audit-script-matrix.md — meta lengths row, eval/QA section, score_eval_transcript, finding_verifier CLI + examples.
• scripts/run_individual_checks.sh — runs meta_lengths_checker when HTML fetch succeeds.
• CI (.github/workflows/validate-plugin.yml) — py_compile all scripts/*.py + eval fixture regression.
• Marketplace + plugin.json descriptions and version 1.2.0.

Fixed

• score_eval_transcript.py — clear stderr + exit 2 when --text-file is missing (no traceback); docstring points at evals/fixtures/eval1_pass.txt.
• meta_lengths_checker.py — JSON error includes hint to run pip install -r requirements.txt / requirements-check.py.

[1.1.2] - 2026-03-24

Added

• sitemap_checker.py, local_signals_checker.py, image_checker.py — URL/HTML checks for crawl/sitemap, local surface signals, and image alt coverage.
• references/audit-script-matrix.md — maps each audit step to its script and example CLI (plus reference-only rows where no script exists by design).
• scripts/run_individual_checks.sh — runs each diagnostic sequentially (JSON samples); bundled beside audit scripts in the plugin tree.
• requirements.txt — requests + beautifulsoup4 for fetch/HTML scripts.

Changed

• RELEASE.md §3 — points to the audit matrix and run_individual_checks.sh smoke path.
• generate_report.py wired to schema JSON-LD validation, image alt, sitemap, local, and IndexNow probe (keyless) sections + scoring and dashboard blocks.
• validate_schema.py supports --json for tooling; indexnow_checker.py supports --probe without --key.
• Skill §21 + README updated for 23 audit scripts; README documents requirements.txt and PEP 668 venv use.

Fixed

• generate_report.py — On-Page table no longer crashes when canonical is JSON null.

[1.1.1] - 2026-03-24

Changed

• Skill YAML description tightened to stay within common 1024-character limits while preserving trigger terms.
• Plugin bundle (setup-plugin.sh) now copies scripts/ (all audit .py files except check-plugin-sync.py) and evals/ into plugins/.../skills/ultimate-seo-geo/ so Claude Code plugin installs can run python scripts/... as documented.

[1.1.0] - 2026-03-24

Added

• Reference library expanded: core-eeat-framework.md, cite-domain-rating.md, entity-optimization.md (CORE-EEAT, CITE domain authority, entity / Knowledge Graph signals).
• Credits updated for CORE-EEAT benchmark, CITE Domain Rating, Entity Optimizer, and AI SEO / GEO content optimizer sources.
• Skill routing extended for CORE-EEAT / CITE scoring, entity optimization, and related triggers.

Changed

• Documentation alignment: generate_report.py wording reflects the bundled pipeline (not a literal “19 scripts” count); §21 clarifies orchestration vs optional parallel subagents.

[1.0.0] - 2026-03-23

Initial Public Release

Core Capabilities

• Full-site SEO audits with SEO Health Score (0–100) and prioritized findings
• Generative Engine Optimization (GEO) for AI Overviews, ChatGPT, Perplexity
• Technical SEO, on-page, content/E-E-A-T, schema, links, local, international, programmatic SEO
• Site migration playbooks and analytics alignment
• Bundled Python audit scripts and eval scenarios