seo

name	seo
description	Claude-SEO engine root — the orchestrator behind `/mk:seo audit|keywords|ai|programmatic|schema`. Detects business type (SaaS, local, e-commerce, publisher, agency) from homepage signals, dispatches the 24 sibling seo-* skills/agents in parallel, and synthesizes raw findings into a prioritized action plan via a 10-principle framework. Owns the SEO Health Score, quality gates, and the Critical/High/Medium/Low priority model shared by every audit-side sub-skill.
allowed-tools	Read, Write, Glob, Grep, WebSearch, WebFetch

SEO — Claude-SEO Orchestrator

A pile of findings isn't a strategy. This skill collects what the sibling skills observe, walks it through PERCEIVE → ANALYZE → VALIDATE → ACT, and ships a dependency-sequenced action plan where every recommendation carries its own falsifiability check.

When this skill activates

Implicit: "audit my site's SEO", "how's my SEO health", "full SEO audit", "what's wrong with my site's SEO", "SEO for [business type]". Explicit: "Use the seo skill to [task]." Routed from: /mk:seo audit|keywords|ai|programmatic|schema (.claude/commands/mk/seo.md) dispatches here for anything audit-shaped; /mk:seo plan|write|campaign instead route to seo-writing + the seo-writer agent, which this skill does not own. This is the parent skill 24 sibling seo-* skills cross-reference back to.

Scope

Covers:

• Business-type detection from homepage signals (SaaS / local service / e-commerce / publisher / agency).
• Deciding which sibling skills a given audit needs, and in what order.
• The 10-principle synthesis framework that turns raw per-skill findings into a coherent, dependency-sequenced action plan.
• The SEO Health Score (0–100, weighted) and the Critical/High/Medium/Low priority model.
• Cross-cutting quality gates: content thresholds, doorway-page limits, deprecated-schema guardrails — the rules every sibling skill must not violate.

Does NOT cover:

• Any single domain's actual analysis logic — that lives in the sibling skill: technical crawl/CWV → [[seo-technical]]; E-E-A-T/content → [[seo-content]]; schema generation → [[seo-schema]]; AI-search/GEO → [[seo-geo]]; topic clusters → [[seo-cluster]]; content briefs → [[seo-content-brief]]; images → [[seo-images]]; full-site audit assembly → [[seo-audit]]; single-page deep dive → [[seo-page]]; backlinks → [[seo-backlinks]]; local/GBP → [[seo-local]]; maps/geo-grid → [[seo-maps]]; hreflang/i18n → [[seo-hreflang]]; programmatic pages → [[seo-programmatic]]; e-commerce → [[seo-ecommerce]]; search experience → [[seo-sxo]]; competitor pages → [[seo-competitor-pages]]; drift monitoring → [[seo-drift]]; DataForSEO data → [[seo-dataforseo]]; Google APIs → [[seo-google]]; AI image generation → [[seo-image-gen]]; strategic planning templates → [[seo-plan]]; FLOW prompts → [[seo-flow]].
• Writing/publishing articles → seo-writing pipeline + seo-writer agent (a separate ClauKit-native pipeline this skill does not orchestrate; see /mk:seo plan|write|campaign).

Orchestration logic

When /mk:seo audit (or an equivalent implicit request) fires:

1. Detect business type from homepage signals (see table below).
2. Select sibling skills to activate based on type + detected capability:
   • Always: [[seo-technical]], [[seo-content]], [[seo-schema]], [[seo-sxo]] (search experience applies to every site).
   • If a Google Search Console / PageSpeed / GA4 connection is available: also [[seo-google]].
   • If local-service signals detected: also [[seo-local]]; add [[seo-maps]] too when a DataForSEO connection is available.
   • If backlink data sources are reachable (free: Moz/Bing/Common Crawl, or DataForSEO): also [[seo-backlinks]].
   • If content-strategy signals (blog, pillar pages, topic clusters) are present: also [[seo-cluster]].
   • If e-commerce signals detected: also [[seo-ecommerce]].
   • If a stored drift baseline exists for this URL: also [[seo-drift]].
   • GEO / AI-search readiness ([[seo-geo]]) runs on every full audit.
3. Collect each sibling's findings without bucketing them yet.
4. Synthesize via the 10-principle framework (below) — this is the step that turns findings into a strategy, not optional polish.
5. Score: compute the weighted SEO Health Score and bucket recommendations into Critical/High/Medium/Low.
6. Sequence: order the action plan as a dependency graph (what unblocks what), not a flat list.
7. Ship the report to plans/marketing/<target>/seo-audit-report.md.

Narrower requests ("just check my schema", "how's my E-E-A-T") load the single relevant sibling skill directly and skip steps 2–3 — but per step 4's discipline, still pass their recommendations through at least THINK + ACCEPT before emitting.

Business-type detection

Signal	Type	Follow-up
Pricing page, /features, /integrations, /docs, "free trial", "sign up"	SaaS	—
Phone number, address, service area, "serving [city]", Maps embed	Local Service	activate [[seo-local]] (+ [[seo-maps]] if available)
/products, /collections, /cart, "add to cart", product schema	E-commerce	activate [[seo-ecommerce]]
/blog, /articles, /topics, article schema, author pages, publication dates	Publisher	activate [[seo-cluster]]
/case-studies, /portfolio, /industries, "our work", client logos	Agency	—

If detection is ambiguous, present the top two candidate types with their supporting signals and ask the user to confirm before applying type-specific recommendations.

Synthesis: the 10-principle framework

Full source: references/thinking-framework.md. Four phases, ten principles — a recommendation that hasn't passed through all four is a finding, not a recommendation.

Phase	Principle	Discipline
PERCEIVE	Observe (external)	Collect signals — HTML, schema, SERP visibility, backlinks, CWV field data, AI-citation patterns, competitor pages — without scoring or classifying yet.
	Observe (internal)	Audit your own assumptions before recommending: is the homepage representative? Is "low traffic" actually "low value"? Is a CMS limitation really unfixable?
	Listen	Read the site's existing copy, the SERP for target keywords, and user-review language before prescribing a rewrite — they're data, not noise.
ANALYZE	Think	Reduce to first principles: page type + intent match, AI-feature eligibility floor (must be indexed first), the single highest-leverage constraint gating everything else.
	Connect (lateral)	Combine findings sibling skills wouldn't naturally pair — e.g. thin-content × SERP-overlap → merge three weak pages into one cluster hub; low AI-citation × weak brand-mentions → reframe link-building budget toward PR/Reddit/YouTube.
	Connect (system)	Wire validated recommendations into a dependency graph: what unblocks the most other work goes first; flag anything that needs a tool/data source not yet available.
VALIDATE	Feel	Pressure-test against UX, brand voice, and operator capacity — would this recommendation make the page worse for a human, or ask a 2-person team to ship 30 location pages?
	Accept	Every recommendation states "how would we know this failed?" — a measurable, falsifiable check. Retract stale v1 findings explicitly when the guidance has moved on.
ACT	Create	Ship the artifact: the markdown report, the JSON-LD block, the content brief — not another round of strategizing.
	Grow	Name one or two leading indicators to monitor, and what the next audit should look for.

Full audits walk all four phases before bucketing. Narrow single-skill requests still owe THINK + ACCEPT at minimum.

SEO Health Score (0–100)

Weighted aggregate across categories:

Category	Weight
Technical SEO	22%
Content Quality	23%
On-Page SEO	20%
Schema / Structured Data	10%
Performance (Core Web Vitals)	10%
AI Search Readiness	10%
Images	5%

Priority levels

• Critical — blocks indexing or risks a penalty; fix immediately.
• High — significantly impacts rankings; fix within a week.
• Medium — optimization opportunity; fix within a month.
• Low — nice to have; backlog.

Quality gates (hard rules — apply across every sibling skill)

Full thresholds: references/quality-gates.md.

• Content minimums — homepage 500 words, service page 800, blog post 1,500, product page 400+ (80% unique), location page 500–600 depending on tier. See the reference for the full table.
• Location-page scaling — WARNING at 30+ location pages (enforce 60%+ unique content per page); HARD STOP at 50+ (require explicit user justification — doorway-page risk). Doorway tells: only the city name changes, no local specifics, no local business signals, keyword-stuffed URLs.
• Safe-at-scale vs. penalty-risk programmatic pages — integration pages, template/tool pages, glossary entries, and real product pages scale safely; "best X for [industry]" pages, thin competitor-alternative pages, and mass AI-generated pages carry real penalty risk (E-E-A-T failure, thin/duplicate content).
• Schema: never recommend HowTo (Google removed its rich result Sept 2023). FAQPage rich results are fully retired for all sites (May 7, 2026) — flag existing FAQPage at Info, not Critical; don't claim AI-citation benefit; don't recommend removal or new FAQPage for SERP benefit. Use QAPage for genuine user Q&A instead. Full type table (active / deprecated / recent additions): [[seo-schema]] references/schema-types.md.
• Core Web Vitals: always INP, never FID (FID was removed from CrUX/PSI Sept 9, 2024). Thresholds and diagnostic breakdown: references/cwv-thresholds.md.
• E-E-A-T: score content 0–100 against Experience/Expertise/Authoritativeness/Trust before recommending a rewrite; Trust is Google's own stated "most important member of the family". Full rubric + weights: references/eeat-framework.md.

Key concepts

• SEO Health Score — the 0–100 weighted aggregate above; the single number a full audit reports, always accompanied by the underlying category breakdown.
• Falsifiable finding — a recommendation is not done until it states "how would we know this failed?"; this is the ACCEPT step, non-negotiable for every emitted item.
• Doorway page — a programmatic page whose only variation is a swapped city/keyword with no unique local or business content; Google's spam classifiers target these directly.
• Business-type detection — the homepage-signal classification (SaaS/local/e-commerce/publisher/agency) that decides which sibling skills activate and which industry-specific thresholds apply.
• Dependency-sequenced action plan — the ACT-phase output: not a flat priority list but a graph of what unblocks what, so the user knows what to do first.

Output

• plans/marketing/<target>/seo-audit-report.md — full audit: SEO Health Score, category breakdown, dependency-sequenced action plan, one leading indicator per major recommendation.
• plans/marketing/<target>/seo-<action>-report.md — narrower single-skill requests (schema-only, GEO-only, etc.), per .claude/commands/mk/seo.md.
• Inline recommendations in the conversation for quick checks that don't warrant a written report.

Cross-references

• plans/marketing-context.md — required hub (every /mk: command hard-fails without it).
• .claude/commands/mk/seo.md — the command surface that routes audit|keywords|ai|programmatic|schema here.
• .claude/workflows/marketing-rules.md — content quality rules shared across the marketing kit.
• .claude/workflows/seo-workflow.md — the 7-phase closed-loop campaign that wraps this skill's audit output with the seo-writing pipeline.
• skills/marketing/README.md — full kit overview, the 25-skill claude-seo roster, and the 3-layer architecture diagram.
• Sibling skills: see "Does NOT cover" above for the full list with wikilinks.

Provenance

Imported from AgriciDaniel/claude-seo (root seo skill, v2.2.4) and adapted for KitForge. Dropped: the bundled-Python-runtime layer (claude-seo run, /seo setup/doctor, isolated launcher) and the extension-installer mechanics (Firecrawl/DataForSEO/image-gen install.sh) — ClauKit skills read/write/search directly and have no equivalent runtime to bootstrap; the PDF-report offer and the "Optional Extensions" section were dropped for the same reason. The community footer (Skool community links) was dropped as promotional content not relevant to ClauKit. Subagent-count language was adjusted: the source lists 18 conditionally-spawned subagents, ClauKit currently ships 5 real agents for this domain (seo-content, seo-geo, seo-schema, seo-technical, seo-writer) plus the remaining coverage as skills Claude activates directly — the dispatch logic (who gets included and when) is preserved even though the execution mechanism differs.