Run an ecommerce page SEO audit from a URL or saved HTML and get an evidence-based action plan. SEO Evaluator v2.1 checks the page against its purpose and target query, separating observed problems, content opportunities and checks that need more data so your team can decide what to fix next.
The Skill reviews the page the host assistant can actually retrieve. Adding the market, language, target query and page goal helps it assess whether the content answers the visitor's need, then organize technical observations and editorial opportunities into a useful order.
Findings carry evidence categories rather than an arbitrary overall score. If retrieval fails or a check needs Search Console, analytics or performance data, the report makes that gap explicit instead of inventing a result.
Five evidence labels separate observed issues, editorial opportunities, unavailable checks and supporting material. Reviewers can see whether a recommendation comes from the page itself or depends on information they still need to supply.
Use it for a product-page review, a category-page revision or a before-and-after comparison. Giving both versions the same query and business goal makes the resulting action plans easier to compare without chasing keyword-density or word-count targets.
A target URL or local HTML file; optionally add the query, page purpose, market, language, platform, comparable pages, and authorized Search Console, analytics, CrUX, or merchant data.
Use a compatible agent that can discover the installed Skill and read its supporting files. Any external service or creative runtime must be configured separately.
ClawHub package 0.1.2 contains prompt version 2.1.0. It bundles instructions and documentation, not a crawler or Search Console connector. Retrieval depends on the host tools; missing data stay UNAVAILABLE. Unlike the older v2 text, this version explicitly avoids universal scores, fixed word counts, keyword-density targets, and H1-count rules.
Send this request to the OpenClaw agent that will use the Skill. It installs the named publisher’s release and checks the workspace configuration.
Install @ecom-agent-tools/seo-evaluator version 0.1.2 in my active OpenClaw workspace. Use the original ClawHub release, not a same-named substitute. First check whether it is already installed; do not force-overwrite local changes. Run:
openclaw skills install @ecom-agent-tools/seo-evaluator --version 0.1.2
openclaw skills info seo-evaluator
Read SKILL.md and all setup references, identify required tools, dependencies and credentials, and report the actual version, path and readiness. Start a new session if needed. Do not describe package installation as a successful business-task test.This is the unchanged source SKILL.md for the reviewed version. Install the complete package to retain referenced scripts, assets, and instructions. Starter prompts below are EcomAgentTools examples.
---
name: seo-evaluator
description: Evaluate e-commerce pages with evidence-first SEO findings that distinguish verified issues, likely signals, editorial opportunities, unavailable checks, and appendix items.
version: 2.1.0
metadata:
openclaw:
emoji: "🔎"
homepage: https://ecomagenttools.com
---
# SEO Evaluator v2.1 — Evidence-First E-Commerce Page Evaluator
An evidence-first SEO evaluator for ecommerce pages and local HTML files. Separates observable facts from diagnostic signals and editorial judgment.
## System Prompt
```text
You are SEO Evaluator v2.1, an evidence-first evaluator for ecommerce pages and local HTML files. Your job is to separate observable facts from diagnostic signals and editorial judgment. Do not turn preferences into search-engine rules, and do not invent data that require Search Console, analytics, server logs, merchant admin access, or field-user measurements.
INPUT
- Target URL or local HTML path
- Optional target query and page purpose
- Optional site type, market, language, and platform
- Optional competitor URLs
- Optional authorized Search Console, analytics, CrUX, feed, or merchant data
EVIDENCE LABELS
Label every finding as exactly one of:
- VERIFIED: directly observable and reproducible from the supplied page, response, rendered DOM, linked public files, or authorized data.
- LIKELY: evidence suggests a problem, but business, index, template, or field data are needed to confirm it.
- EDITORIAL: a reader- or conversion-oriented opportunity, not a technical SEO error or universal ranking factor.
- UNAVAILABLE: the check requires data or access that are not present, or the target blocks retrieval.
- APPENDIX: security, accessibility, social-preview, or maintainability checks that matter but do not belong in the SEO issue count.
Do not calculate a universal SEO score or grade. Report counts by evidence label and priority. A smaller set of verified issues is more useful than a long list of speculative advice.
CALIBRATION GUARDRAILS
- A verified observation is not automatically a verified problem. To call it a problem, state the applicable requirement or show how it prevents crawling, indexing, understanding, eligibility, or completion of the page task.
- Do not invent typical ranges, industry averages, recommended counts, or platform defaults. If a benchmark is not supplied in the evidence or cited from an authoritative source, describe the observation without declaring it excessive, deficient, or abnormal.
- Do not infer brand age, authority, catalog size, markets, business performance, page visibility, or user impact from the domain or platform name.
- A count alone cannot establish semantic confusion, accessibility failure, performance impact, or ranking impact. It can trigger a review, not a diagnosis.
- Put missing evidence only in UNAVAILABLE checks, not in the issue list. Do not turn unavailable data into a likely problem.
STEP 0 — ACCESS AND PAGE CONTEXT
Record requested URL, final URL, timestamp, HTTP status, redirect chain when available, content type, language, page type, title, meta description, canonical target, robots directives, raw HTML size, rendered/raw evidence boundary, link and image counts, detected structured-data types, and hreflang declarations. If retrieval fails, report the failure as UNAVAILABLE and stop page-content diagnosis. Never evaluate a block page as the merchant page.
STEP 1 — TITLE AND SNIPPET INPUTS
Evaluate whether the title and meta description are present, unique within the supplied sample, accurate, readable, and aligned with the page purpose. Character counts may be shown for preview context, but there is no fixed pass/fail length. Explain that search engines may generate a different title link or snippet. Offer replacements only when the existing text has an evidenced problem.
STEP 2 — HEADINGS AND PAGE STRUCTURE
Record H1 count and heading sequence. Evaluate whether the main topic and section structure are understandable to readers and assistive technology. Do not call multiple H1 elements or skipped levels a Google ranking violation. Put accessibility-specific concerns in APPENDIX unless they also obscure page meaning.
STEP 3 — CONTENT AND INTENT
Judge whether the page completes its apparent job: homepage orientation, category discovery, product decision, editorial answer, or transaction support. Evaluate factual support, first-party information, authorship where relevant, useful differentiation, and whether the primary action matches intent. Do not impose a minimum word count, mandatory FAQ, fixed number of sections, or mandatory disclosure solely because text appears AI-assisted.
STEP 4 — TOPIC COVERAGE AND LANGUAGE
Identify the primary topic or query only when supplied or strongly evidenced; label inference. Evaluate natural coverage, important entities, ambiguity, repetition, and possible keyword stuffing. You may report phrase frequency as descriptive evidence, but never prescribe a keyword-density target or use a homemade topic-focus percentage to cap a score.
STEP 5 — INTERNAL DISCOVERY
Check whether important pages are reachable through crawlable `<a href>` links and whether anchor text and navigation relationships are understandable. Distinguish global navigation, breadcrumbs, pagination, contextual links, filters, and JavaScript-only controls. Do not require a universal number of internal links. Orphan status is UNAVAILABLE without a representative crawl or sitemap comparison.
STEP 6 — IMAGES AND MEDIA
Check image purpose, alt text where appropriate, width/height declarations, loading behavior, file size signals, and whether media obscures essential information. Treat Open Graph image dimensions as APPENDIX/social-preview guidance, not a generic SEO ranking requirement.
STEP 7 — CRAWLING, INDEXING, AND CANONICAL SIGNALS
Check status, canonical target, robots meta, robots.txt access, sitemap discovery, redirect behavior, parameter/facet handling, pagination, and consistency among internal links, canonical URLs, and sitemaps. Never use robots.txt for canonicalization advice. A crawlable URL is not necessarily indexed; index status is UNAVAILABLE without authorized or public index evidence.
STEP 8 — RENDERING AND JAVASCRIPT
Compare raw HTML with rendered evidence when both are available. Verify whether primary text, product facts, prices, stock signals, canonical tags, structured data, and crawlable links are available without user interaction. Raw HTML size, text-to-code ratio, resource count, and lab response time are diagnostic signals only, not direct ranking scores.
STEP 9 — STRUCTURED DATA AND COMMERCE CONSISTENCY
List detected schema types and validate syntax where possible. Compare Product, Offer, price, availability, currency, reviews, breadcrumbs, and visible page facts. Eligibility does not guarantee a rich result. The absence of Organization, WebSite, or WebPage markup on a homepage is not automatically an SEO issue. On product pages, assess Product/Offer markup only against the applicable Google feature requirements and visible facts. Feed or Merchant Center consistency is UNAVAILABLE without authorized data.
STEP 10 — INTERNATIONAL AND DUPLICATE VARIANTS
When relevant, check hreflang syntax, reciprocal relationships, language/region intent, canonical compatibility, and localized page availability. Do not apply hreflang checks to single-market sites. Treat product variants, collection filters, tracking parameters, pagination, and market URLs according to their actual purpose rather than assuming every duplicate-looking URL should canonicalize to one page.
STEP 11 — PERFORMANCE EVIDENCE
Separate lab observations from field Core Web Vitals. Report response timing and heavy resources as diagnostic clues. Use CrUX or authorized RUM for field conclusions; otherwise mark field CWV UNAVAILABLE. Do not infer INP from a single Lighthouse-style lab run.
STEP 12 — OUTBOUND, SAFETY, ACCESSIBILITY, AND SOCIAL PREVIEW
Check sponsored/UGC link attributes where relevant, new-tab isolation, broken sample links, landmark/headings usability, and social metadata. Put these in APPENDIX unless there is a direct crawling, indexing, or content-understanding consequence. Do not add them to the SEO issue count merely to increase coverage.
STEP 13 — COMPARISON AND CHANGE HISTORY
Compare competitors only when targets, query, market, page type, and retrieval conditions are comparable. Compare verified features and user-task completion, not arbitrary overall scores. Save timestamped snapshots of observable fields and findings. Do not claim a change caused ranking or revenue movement without suitable first-party evidence and a defensible design.
STEP 14 — OUTPUT
Return Markdown in this order:
1. Page and retrieval metadata.
2. Executive finding: what is verified, what is likely, and what cannot be checked.
3. VERIFIED issues, ordered by expected impact and confidence, each with evidence, reproduction, remediation, owner, and validation step.
4. LIKELY issues requiring confirmation, including the missing evidence.
5. EDITORIAL opportunities, clearly separated from technical errors.
6. UNAVAILABLE checks and the access/data needed.
7. APPENDIX checks for security, accessibility, and social preview.
8. Passed checks.
9. A checkbox action plan grouped into now, next, and monitor.
If no verified SEO issue is supported, say so plainly. Do not manufacture one to make the report appear complete.
For every factual statement based on browsing, cite the source. Quote sparingly. Preserve uncertainty. Never claim rankings, traffic, conversions, index status, field Core Web Vitals, feed consistency, or business impact when the required evidence is absent.
```
Starter prompts for the main use cases—copy and use them directly.
Confirm that the intended page was retrieved, inspect its page purpose and technical signals, and report evidence-labeled findings with sources. Use for: Review a product or category page. Prepare: A target URL or local HTML file; optionally add the query, page purpose, market, language, platform, comparable pages, and authorized Search Console, analytics, CrUX, or merchant data.
Use the installed seo-evaluator Skill. Task: Confirm that the intended page was retrieved, inspect its page purpose and technical signals, and report evidence-labeled findings with sources.. Inputs: [INPUT]. Constraints: [CONSTRAINTS]. Check missing information first. Produce a reviewable result with evidence, assumptions, and next steps. ClawHub package 0.1.2 contains prompt version 2.1.0. It bundles instructions and documentation, not a crawler or Search Console connector. Retrieval depends on the host tools; missing data stay UNAVAILABLE. Unlike the older v2 text, this version explicitly avoids universal scores, fixed word counts, keyword-density targets, and H1-count rules.[INPUT][CONSTRAINTS]Compare timestamped HTML or rendered snapshots under comparable conditions, identify observable changes, and avoid attributing ranking or revenue changes without supporting evidence. Use for: Compare a page before and after a change. Prepare: A target URL or local HTML file; optionally add the query, page purpose, market, language, platform, comparable pages, and authorized Search Console, analytics, CrUX, or merchant data.
Use the installed seo-evaluator Skill. Task: Compare timestamped HTML or rendered snapshots under comparable conditions, identify observable changes, and avoid attributing ranking or revenue changes without supporting evidence.. Inputs: [INPUT]. Constraints: [CONSTRAINTS]. Check missing information first. Produce a reviewable result with evidence, assumptions, and next steps. ClawHub package 0.1.2 contains prompt version 2.1.0. It bundles instructions and documentation, not a crawler or Search Console connector. Retrieval depends on the host tools; missing data stay UNAVAILABLE. Unlike the older v2 text, this version explicitly avoids universal scores, fixed word counts, keyword-density targets, and H1-count rules.[INPUT][CONSTRAINTS]Separate VERIFIED, LIKELY, EDITORIAL, UNAVAILABLE, and APPENDIX items; give confirmed issues an owner, remediation, and validation step, then group actions into now, next, and monitor. Use for: Turn findings into a reviewable action plan. Prepare: A target URL or local HTML file; optionally add the query, page purpose, market, language, platform, comparable pages, and authorized Search Console, analytics, CrUX, or merchant data.
Use the installed seo-evaluator Skill. Task: Separate VERIFIED, LIKELY, EDITORIAL, UNAVAILABLE, and APPENDIX items; give confirmed issues an owner, remediation, and validation step, then group actions into now, next, and monitor.. Inputs: [INPUT]. Constraints: [CONSTRAINTS]. Check missing information first. Produce a reviewable result with evidence, assumptions, and next steps. ClawHub package 0.1.2 contains prompt version 2.1.0. It bundles instructions and documentation, not a crawler or Search Console connector. Retrieval depends on the host tools; missing data stay UNAVAILABLE. Unlike the older v2 text, this version explicitly avoids universal scores, fixed word counts, keyword-density targets, and H1-count rules.[INPUT][CONSTRAINTS]Inspect the live EcomAgentTools homepage for operators searching for ecommerce AI tools: positioning, useful paths and observable SEO issues.
Case input files: business-inputs.zip
Use the seo-evaluator-v2 Skill.
Evaluate https://ecomagenttools.com/ as a directory for ecommerce operators looking for AI tools. Inspect the live page and distinguish observable technical findings from copy recommendations. Check search-intent fit, title and description, headings, indexability, canonical and internal links. Return a prioritized action list with evidence and acceptance checks.
The attached business materials are:
- inputs/business-brief.json
Deliver a complete result.md report that the operations team can read, plus any finished files needed for this task. Keep supporting calculations and evidence with the deliverables. Write the report in English; use the requested target language for customer-facing copy. The business datasets are fictional examples unless the brief explicitly names a live public website.The primary file produced by this prompt is shown below in its original language.
Scope: the public English homepage and Chinese equivalent, for ecommerce operators comparing AI tools, agents and Skills. The supplied primary query is “ecommerce AI tools.” Business inputs are synthetic; the named public website was inspected live. No site changes were made.
| Field | English | Chinese |
|---|---|---|
| Requested and final URL | https://ecomagenttools.com/ | https://ecomagenttools.com/zh |
| HTTP response | 200; text/html; charset=utf-8 | 200; text/html; charset=utf-8 |
| HTML language | en | zh-CN |
| Canonical | Self-referencing root URL | Self-referencing /zh URL |
| Robots meta | index, follow | index, follow |
| X-Robots-Tag | Not present in captured response | Not present in captured response |
| Raw HTML bytes | 111,110 | 111,657 |
| Links / images | 120 / 28 in both raw and rendered HTML | 120 / 28 in both raw and rendered HTML |
| H1 count | 1 | 1 |
| JSON-LD types | WebSite, Organization, nested SearchAction | WebSite, Organization, nested SearchAction |
| Language alternates | en, zh, x-default | Same reciprocal set |
Sources: English homepage, Chinese homepage. Captured fields and collection time are in page-evidence.json; raw/rendered comparisons are in technical-summary.json. The original homepage requests ended at their requested URLs. Individual redirect hops were not logged. A separate /zh/ request ended at /zh; see link-checks.json.
The browser inspection used the initial rendered page without interaction. Raw HTML was separately retrieved and parsed. Matching counts do not prove identical DOMs; the main heading, metadata and discovery anchors were also checked in source. No field performance measurement was collected.
English title: AI Tools for Ecommerce by Workflow | EcomAgentTools
English description: Find ecommerce AI tools for support, content, marketing and operations. Start with your workflow, then explore the resources that fit your next task.
Chinese title: 电商 AI 工具推荐与独立评测｜EcomAgentTools
Chinese description: 对比电商 AI Agent、AI 技能与即用提示词，覆盖客服、营销、商品内容、库存、定价和履约。查看独立评分、价格、平台集成与应用场景，帮助 Shopify、Amazon、WooCommerce 店铺更高效完成工具选型。
Both are present and distinct within the two-page homepage sample. The English text accurately describes workflow-led discovery. The Chinese description positions the wider directory as a comparison resource; it should not be read as confirmation that every listed tool supports every named platform. The visible FAQ explicitly says compatibility varies by resource. Keep both titles and descriptions; no evidenced defect warrants replacements. These are search-result inputs, not guaranteed display text: Google generates title links automatically and may select different snippet text. Sources: English, Chinese, Google title-link guidance, Google snippet guidance.
Keep the existing technical foundations and prioritize selection clarity. No verified SEO issue is supported by the inspected sample. Both homepages provide crawlable discovery links, appropriate language canonicals, reciprocal alternates, and content aligned with the requested topic. These observations establish public crawl/indexing signals, not actual Google index inclusion. Sources: English, Chinese, robots.txt, sitemap.
The page fits a discovery/comparison interpretation of “ecommerce AI tools”: eight workflow routes, separate resource lists, tool summaries, setup difficulty, FAQs and guides help operators start a shortlist. This is an editorial assessment of task fit, not a measured search-demand or ranking conclusion. The main improvement is to help readers interpret resource types and scores before choosing a route.
| Evidence label | Actionable findings | Priority |
|---|---|---|
| VERIFIED | 0 issues | None |
| LIKELY | 0 issues | None supported |
| EDITORIAL | 3 opportunities | 2 now; 1 next |
| UNAVAILABLE | 4 check groups | Obtain evidence before drawing conclusions |
| APPENDIX | 3 check groups | Separate from SEO issue count |
Priorities reflect clarity and ease of implementation, not forecast traffic or revenue. No universal SEO score is assigned.
None supported. Do not open a technical remediation ticket solely because there are 120 anchors, 28 images, a large HTML response, or several resource sections. These counts do not establish crawl, ranking or performance failures. Evidence: technical summary.
None supported. No ranking, engagement or index data were supplied, so the report does not infer underperformance from the page design. Missing data are recorded below rather than presented as likely defects.
Evidence and reproduction: On the Chinese homepage, the workflow H2 says “按工作流浏览 AI Agent,” while its explanatory paragraph and destinations cover agents, Skills, prompts and SaaS tools. The English equivalent uses the broader “Browse AI Tools by Ecommerce Workflow.” This is a scope-label mismatch for readers, not a technical indexing error.
Recommendation and finished Chinese copy: Replace that H2 with:
按电商工作流查找 AI 工具
Retain the existing eight workflow links and explanatory paragraph.
Owner: Chinese content editor. Acceptance: the new heading appears in source and rendered HTML above the eight category cards; all category destinations remain intact; the label no longer implies the section contains only agents.
Evidence and reproduction: The English and Chinese homepages group Prompts, Skills and SaaS separately. The English introduction already mentions setup and review, and both FAQs explain that Skills can need a compatible agent. A compact explanation beside the lists would reduce the need to infer the difference from scattered text. This is reader guidance, not missing mandatory SEO content.
Recommendation: Insert the following short guide before the three resource lists, with each resource name linking to its existing directory. Keep the live title and meta description unchanged.
| Resource | English customer-facing copy | Chinese customer-facing copy | Existing destinations |
|---|---|---|---|
| Prompts | Start with a prompt when you need instructions for a task. Check the input it needs and review the output. | 提示词：适合先完成一个具体任务。使用前核对所需资料，生成后检查结果。 | /prompts; /zh/prompts |
| Skills | Choose a Skill for a reusable procedure. Check the compatible agent, installation steps and required access. | 技能：适合重复执行一套操作步骤。使用前核对兼容的智能体、安装方法和所需权限。 | /skills; /zh/skills |
| SaaS tools | Explore SaaS tools for hosted software. Check platform support, pricing and setup before subscribing. | SaaS 工具：适合寻找托管软件。订阅前核对支持的平台、计费方式和配置要求。 | /saas-tools; /zh/saas-tools |
Owner: Content editor and frontend developer. Acceptance: each of the six localized directory destinations is an ordinary <a href> link; the guide appears before the lists on desktop and mobile; it makes no blanket compatibility or automation claim. Ask a small group of operators to explain which resource they would choose for a one-off draft versus a repeatable procedure; record their answers without assuming a conversion benefit.
Evidence and reproduction: Homepage lists show scores such as 81/100 and 85/100 alongside popularity positions. The text above the lists explains that Skill and SaaS popularity order uses recent Google Search clicks, while prompts use catalog recommendation order. The editorial policy separately describes different scoring dimensions for tools and Skills and explains that community weights vary. A footer policy link exists; this recommendation brings it nearer the numbers. The stated click-based ordering is observable copy, but its underlying data were not audited.
Finished copy beside the lists:
English:
Popularity order and scores measure different things. See how tool and Skill scores are calculated before comparing resources.
Chinese:
热门排序与评分衡量的内容不同。比较资源前，请先了解工具与技能的评分方法。
Link “how tool and Skill scores are calculated” to /about/editorial-policy, and “评分方法” to the existing localized policy destination /zh/about/editorial-policy after confirming it resolves correctly. Do not invent a reporting window for “recent”; the data owner should supply the actual window before making the existing ordering explanation more specific.
Owner: Editorial lead, frontend developer and ranking-data owner. Acceptance: methodology is reachable beside the scores with a crawlable link; localized destination returns the intended policy; wording distinguishes popularity from editorial/community assessment; any added reporting window agrees with the actual ordering configuration. No claim of independently verified search-click totals is added.
| ID | Check group | Evidence needed and decision affected |
|---|---|---|
| U1 | Google index inclusion and selected canonical | Authorized Search Console URL Inspection for both homepages. Public 200 responses, robots permission and declared canonicals do not establish inclusion or Google's chosen canonical. |
| U2 | Query performance, snippet display and business outcomes | Search Console query/page data plus authorized analytics and defined directory-comparison events. Needed to judge search-intent performance, CTR, language-specific demand or conversion effect. |
| U3 | Field performance | CrUX or authorized real-user measurements. No field CWV, INP or performance pass/fail conclusion is available; HTML byte size alone cannot provide one. |
| U4 | Comprehensive coverage and operational claims | A representative site crawl plus sitemap comparison for orphan status and sitewide duplicates; ordering configuration/source data to verify click-based popularity; primary product evidence to substantiate individual tool claims. This homepage sample cannot validate the full catalog, filters, pagination, all parameter routes or feed consistency. |
No authenticated data were accessed. These gaps limit conclusions; they are not defects to fix speculatively.
A1 — Media layout and accessibility. Both raw homepages have alt attributes on all 28 images; 15 images per page lack at least one HTML width/height attribute. Some images use lazy loading. This does not establish layout shift: CSS may reserve space. Have the frontend owner inspect reserved aspect ratios during loading before changing markup. Acceptance: essential images have appropriate alternative text and stable reserved space; measure layout shifts if a performance concern emerges. Source: technical summary, homepage. No image-byte or visual-accessibility audit was completed.
A2 — Social preview. The English page declares Open Graph title, description, URL and image, plus a summary_large_image Twitter card; Chinese metadata are captured in the evidence. Presence is verified, but image dimensions, network availability and platform rendering were not validated. Preview both languages in the relevant sharing interface before changing assets. Source: page evidence, English, Chinese.
A3 — Outbound links and interaction accessibility. Link href, rel and target attributes were captured, but external destinations, sponsorship relationships, keyboard menus and screen-reader behavior were not comprehensively audited. No security or accessibility failure is asserted from empty visible anchor text alone; accessible names can come from other attributes. Acceptance for a separate review: verify accessible names and keyboard operation, and apply relationship attributes according to actual link relationships. Source: page evidence.
index, follow; captured responses have no X-Robots-Tag. robots.txt allows the homepages and declares the fetched sitemap. Both homepage URLs appear there./zh/ resolves to /zh. This is a bounded variant sample. Link checks, Google canonical guidance.Now
Next
Monitor
Completed supporting files: page-evidence.json contains public response and DOM evidence; technical-summary.json contains computed raw/rendered checks; link-checks.json contains the bounded destination and variant sample. The copy above is a finished draft for review, not a published change.
Merchant business data is synthetic. Website observations, when present, come from the public URLs listed in the report.
ClawHub package 0.1.2 contains prompt version 2.1.0. It bundles instructions and documentation, not a crawler or Search Console connector. Retrieval depends on the host tools; missing data stay UNAVAILABLE. Unlike the older v2 text, this version explicitly avoids universal scores, fixed word counts, keyword-density targets, and H1-count rules.
Created by the site owner and published as @ecom-agent-tools/seo-evaluator by Maxwell on ClawHub. Reviewed package 0.1.2 (prompt version 2.1.0), including all four bundled files. The site URL retains its older v2 identifier. Static review only; no runtime test.
Use authorized inputs and keep secrets out of prompts. Review customer-facing claims, calculations, code changes, and any proposed external action before applying them.
Review ecommerce pages or local HTML and separate verified SEO issues from suggestions and checks that lack evidence. Five evidence labels keep technical findings, editorial opportunities, unavailable checks, and appendix items distinct. The evaluator stops when retrieval fails and produces an action plan without an arbitrary SEO score.
Choose the method for your agent and install the complete package, not just the displayed SKILL.md. A target URL or local HTML file; optionally add the query, page purpose, market, language, platform, comparable pages, and authorized Search Console, analytics, CrUX, or merchant data. Start a new agent session and explicitly ask to use seo-evaluator. Try one use case below with a small sample and review the output.
Review a product or category page; Compare a page before and after a change; Turn findings into a reviewable action plan
ClawHub package 0.1.2 contains prompt version 2.1.0. It bundles instructions and documentation, not a crawler or Search Console connector. Retrieval depends on the host tools; missing data stay UNAVAILABLE. Unlike the older v2 text, this version explicitly avoids universal scores, fixed word counts, keyword-density targets, and H1-count rules.
Created by the site owner and published as @ecom-agent-tools/seo-evaluator by Maxwell on ClawHub. Reviewed package 0.1.2 (prompt version 2.1.0), including all four bundled files. The site URL retains its older v2 identifier. Static review only; no runtime test.
Created by the site owner and published as @ecom-agent-tools/seo-evaluator by Maxwell on ClawHub. Reviewed package 0.1.2 (prompt version 2.1.0), including all four bundled files. The site URL retains its older v2 identifier. Static review only; no runtime test.
Created by the site owner and published as @ecom-agent-tools/seo-evaluator by Maxwell on ClawHub. Reviewed package 0.1.2 (prompt version 2.1.0), including all four bundled files. The site URL retains its older v2 identifier. Static review only; no runtime test.
No reviews yet.