Skills
Research and audit search work with ak:seo
Use ak:seo for SEO and GEO audits, keyword research, and pSEO artifacts while separating provider queries, OAuth, bulk generation, and live search changes.
Use ak:seo to audit a page or domain, check GEO and AI discoverability,
research and cluster keywords, or design programmatic SEO templates. The Skill also packages scripts for Search Console,
Core Web Vitals, sitemaps, schema, and bulk page generation. Those scripts have
different network, credential, file, and account effects and are not one
automatic workflow.
Choose ak:seo for evidence and implementation artifacts
Use ak:seo when
- You need a technical or on-page audit of a public URL or local content.
- You need keyword ideas, intent clusters, difficulty data, or competitor gaps with the source and freshness recorded.
- You need a pSEO template, data schema, internal-linking plan, and sample pages.
- You need a reviewed sitemap or JSON-LD artifact before implementation.
- You need a GEO and AI-discoverability contract or audit:
robots.txt,sitemap.xml,llms.txt,llms-full.txt, per-page.mdvariants, social metadata and cards, and Copy, Share, or Open in AI page actions.
Choose another workflow when
- You need broader content strategy without search evidence. Use
ak:content-marketing. - You need final copy from an approved search brief. Use
ak:writeafter the keyword and claim inputs are accepted. - You need to implement those GEO surfaces in a framework. Implementation
belongs to that framework's Skill; for Next.js, use the Engineer Kit
ak:web-frameworksSEO and GEO reference when that Kit is installed. - You want guaranteed rankings, traffic, rich results, indexing, or revenue. No Skill or provider output can establish those outcomes.
Prepare data, access, and mutation boundaries
Confirm Marketing Kit is installed for the current runtime and scope. Provide the target URL, domain, topic, country or market, audience, business relevance, current analytics, competitor set, and output boundary. Identify which data may be sent to external services.
ReviewWeb-backed keyword and audit data requires REVIEWWEB_API_KEY. Core Web
Vitals requests use Google PageSpeed Insights and can use PAGESPEED_API_KEY.
Google Search Console requires a verified property, Google Cloud OAuth client,
the Search Console API, googleapis, and
.claude/secrets/google_client_secret.json or the corresponding user-level
secret path. OAuth tokens are saved locally; protect them as credentials.
| Runtime | Invocation | Availability boundary |
|---|---|---|
| Claude Code | /ak:seo ... | Native delivery is the default; network and account operations still require explicit access and approval. |
| Cursor | /ak:seo ... | The Skill is projected to Cursor and slash invocation is user-verified; provider tooling can still differ. |
| Codex | $ak:seo ... | The Skill uses native Codex discovery; available web, shell, credential, and provider capabilities govern each stage. |
Run a routed SEO task
/ak:seo audit https://example.com/pricing "Use public page evidence, disclose every provider request, write a report only, and make no site or Search Console changes"/ak:seo keywords "privacy-first analytics" "Cluster by intent for the US market, distinguish measured API data from estimates, and cite the source date"$ak:seo pseo comparison "Design one template, its data schema, and three sample pages from the supplied approved dataset. Do not publish or bulk-generate production pages"The canonical argument hint lists
[audit|geo|keywords|pseo|optimize|schema] [target]. In this release, only
audit, geo, keywords, and pseo have routed reference workflows, and a
full-site audit also runs the geo checklist. Treat optimize and
schema as unproven direct routes; use a clearly scoped request or the bundled
schema tools only after inspecting their effect.
Audit AI discoverability with geo
The geo route treats search crawlers, AI answer engines, and people who paste
pages into chat assistants as one contract. It checks only what the site's
framework and hosting can serve and reports missing surfaces as gaps.
| Surface | What the audit expects |
|---|---|
robots.txt | Public pages allowed, private and search-result paths blocked, an absolute Sitemap: line, and an AI crawler policy the site owner chose explicitly. |
sitemap.xml | Canonical, indexable URLs only, real lastmod, XML-escaped <loc>, no .md variants. |
llms.txt and llms-full.txt | llmstxt.org format with ## Optional last, links to each page's .md variant, and a full file generated from the same source with no private content. |
<page>.md | A Markdown twin at the page URL plus .md (index.html.md for a URL ending in /), served as text/markdown; charset=utf-8, linked from the HTML head with rel=alternate, and sent X-Robots-Tag: noindex. |
| Metadata and social cards | rel=canonical, valid JSON-LD matching visible content, complete Open Graph and Twitter tags, and a per-page social image at 1200×630. |
| Page actions | Copy as Markdown, Copy URL, Open in ChatGPT or Claude, and Share controls that work without login and degrade gracefully. |
Blocking AI crawlers is never a default. The robots reference separates
training, search, and user-initiated fetch agents per operator, notes that
retired names such as anthropic-ai and Claude-Web are unused, and asks you
to confirm current agent names with each operator before shipping a policy.
The audit reports each item as pass, fail, or not applicable with the observed
URL and response.
Understand the stages
- Confirm the question and data source. Define URL, topic, market, timeframe, evidence source, and whether the task is analysis or mutation.
- Collect evidence. Read local content or fetch public pages. With approval, query ReviewWeb, PageSpeed Insights, or Search Console and record the source.
- Analyze the evidence. Audit technical and on-page elements, classify intent and clusters, or design a pSEO data and template architecture.
- Create artifacts. Produce reports, templates, schema, sitemaps, sample pages, or implementation recommendations inside the approved output boundary.
- Validate locally. Check generated JSON-LD structure, sitemap content,
template escaping, duplicates, claims, canonical intent, and sample quality.
The bundled sitemap generator XML-escapes each URL and omits
lastmodfor a page without a supplied date instead of stamping today's date. - Review before live action. A person approves code changes, bulk files, deployment, indexing controls, sitemap submission or deletion, outreach, and any provider account mutation.
Separate queries, file generation, and account changes
SEO analysis does not authorize live search changes
Search Console query actions can read private property data, while separate bundled flags can submit or delete a sitemap. The pSEO generator writes many HTML files and a checkpoint. Sitemap and schema generators write output files. Approve each effect explicitly; never infer it from a request for an audit.
Review URLs and customer data before sending them to ReviewWeb, Google, browser tools, or other providers. Do not place API keys or OAuth tokens in arguments, reports, or commits. Link building and outreach remain separate human-approved work; avoid spam, deceptive placements, and fabricated relationships.
Verify outputs and evidence
The top-level Skill summary names report paths under assets/reports/seo/,
while routed audit and keywords references name assets/seo/audits/ and
assets/seo/keywords/. Confirm the actual path returned by the run before
building automation. The pSEO route writes under
assets/attraction/pseo-templates/; schema output is described under
assets/seo/schemas/.
Require the report to identify URLs, query dates, country or market, provider and endpoint, measured fields versus estimates, missing data, issues by priority, and recommended next checks. For generated files, inspect the exact file list and diff. A local schema validation pass does not guarantee a search engine rich result, and a submitted sitemap does not guarantee indexing.
Troubleshoot and know the limits
| Symptom | Safe next step |
|---|---|
| ReviewWeb data is unavailable | Continue with clearly labeled public or local evidence, or configure REVIEWWEB_API_KEY; do not invent volume, CPC, difficulty, traffic, or backlinks. |
| Search Console authentication fails | Verify the property, OAuth client, scopes, secret path, local callback, dependency, and token status without exposing credentials. |
| The report path differs from the overview | Use the path actually returned and record the source-path inconsistency; do not move or overwrite files silently. |
optimize or schema does not route | Use audit, geo, keywords, or pseo, or request a bounded schema artifact and inspect the bundled generator directly. |
| Bulk pSEO output is low quality or duplicated | Stop generation, inspect the template and dataset, delete nothing automatically, and require unique value plus human sampling before publication. |
| Runtime discovery fails | Follow Runtime cannot find a Skill or Agent. |
Search data, provider APIs, platform rules, and ranking systems change. The Skill cannot guarantee data completeness, ranking, indexing, traffic, rich results, conversions, or revenue. Continue with the Marketing Kit overview and keep every live mutation separately approved.
Draft and review marketing copy with ak:write
Route a bounded writing task through ak:write, preserve brand and evidence inputs, and distinguish file readiness from actual publication.
Design email copy and sequences with ak:email
Create email drafts, flows, and sequence artifacts with ak:email while keeping consent, compliance, deliverability, and sending authority separate.