seo-landing
SEO Landing: Give your AI coding agent the capabilities of a senior Technical SEO engineer. An agent skill for building high-performance, technically optimized SEO landing pages. Turn an AI coding agent into a technical SEO specialist. Build and improve landing pages with: β’ π 100/100 Google PageSpeed target β’ β‘ Core Web Vitals optimization
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/aleksandr-alhoff--seo-landing)HTML
<a href="/listings/aleksandr-alhoff--seo-landing"><img src="/api/listings/aleksandr-alhoff--seo-landing/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
The skill is hosted at [github.com/aleksandr-alhoff/seo-landing](https://github.com/aleksandr-alhoff/seo-landing). Clone the repository β the repo root is the skill folder itself β then run the commands below from the directory that contains `seo-landing/`:
git clone https://github.com/aleksandr-alhoff/seo-landing.git
Global (available in all projects)
Every recipe below is self-contained: it creates its destination directory first, then copies the skill into it. Each one must exit with status `0` and leave the layout `<skills-dir>/seo-landing/SKILL.md` in place.
Note: `cp -R` from a git clone also copies the clone's `.git` directory into the installation β harmless, but unnecessary. To keep installations lean, replace `cp -R seo-landing <dir>/` with `rsync -a --exclude=.git seo-landing <dir>/seo-landing/` (the same sync used for updates below).
# VS Code Copilot mkdir -p ~/.copilot/skills cp -R seo-landing ~/.copilot/skills/ # Claude Code mkdir -p ~/.claude/skills cp -R seo-landing ~/.claude/skills/ # OpenAI Codex CLI / ChatGPT desktop mkdir -p ~/.agents/skills cp -R seo-landing ~/.agents/skills/ # Cursor mkdir -p ~/.cursor/skills cp -R seo-landing ~/.cursor/skills/ # Gemini CLI mkdir -p ~/.gemini/skills cp -R seo-landing ~/.gemini/skills/ # Google Antigravity mkdir -p ~/.gemini/config/skills cp -R seo-landing ~/.gemini/config/skills/ # OpenCode mkdir -p ~/.config/opencode/skills cp -R seo-landing ~/.config/opencode/skills/ # OpenClaw (via CLI β installs into the shared ~/.openclaw/skills) openclaw skills install git:aleksandr-alhoff/seo-landing --global # or from a local clone: openclaw skills install ./seo-landing --global # or manually: mkdir -p ~/.openclaw/skills cp -R seo-landing ~/.openclaw/skills/ # Hermes mkdir -p ~/.hermes/skills cp -R seo-landing ~/.hermes/skills/
OpenClaw also picks up skills from `~/.agents/skills` (the Codex path above), and Hermes can scan it too if you add `~/.agents/skills` to `skills.external_dirs` in `~/.hermes/config.yaml`.
Gemini CLI can also install straight from a Git repository:
gemini skills install https://github.com/aleksandr-alhoff/seo-landing.git
Per project (workspace only)
# Shared .agents/skills β picked up by VS Code Copilot, Codex, Cursor, Antigravity, Gemini CLI, OpenCode, OpenClaw, Hermes mkdir -p .agents/skills cp -R seo-landing .agents/skills/ # Claude Code mkdir -p .claude/skills cp -R seo-landing .claude/skills/ # VS Code Copilot (GitHub-style location) mkdir -p .github/skills cp -R seo-landing .github/skills/ # OpenClaw (workspace skills β highest precedence) # via CLI: openclaw skills install ./seo-landing mkdir -p skills cp -R seo-landing skills/ # Hermes (project-local; then trust the repo once: hermes skills trust) mkdir -p .hermes/skills cp -R seo-landing .hermes/skills/
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md Β· Manifest Β· 128 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- ---
- name: seo-landing
- description: "Generates fast, SEO-optimized static HTML landing pages targeting 100/100 PageSpeed (LCP < 2.5s, INP < 100ms, CLS < 0.1), full schema.org JSON-LD, AVIF images, critical CSS, zero external dependencies. Use when: user asks to create/build/generate a landing page, one-pager, or static site with focus on SEO, speed, or PageSpeed; asks for an SEO-friendly page from a brief/Π’Π; or asks to audit/fix a landing against a performance checklist."
- metadata:
- argument-hint: "[topic/domain or brief]"
- ---
- # SEO Landing Generator
- Builds a static single-page HTML landing optimized for 100/100 PageSpeed and maximum SEO: critical CSS, AVIF images, full JSON-LD structured data, native-only interactivity, zero third-party requests on first load.
- ## When to Use
- - User asks for a landing page / one-pager focused on speed and SEO β **generate** mode.
- - User provides a brief (Π’Π) and wants a production-ready static page β **generate** mode.
- - User asks to audit an existing landing against the checklist without changes β **audit-only** mode (Β§0b, read-only, no project files).
- - User asks to fix/improve an existing landing β **fix-existing** mode (Β§0c).
- ## Procedure
- ### 0. Route the request to a mode first
- The skill serves three modes β pick one from the request (ask if ambiguous), because each mode collects different inputs and produces different output:
- - **generate** β build a new landing page from a brief ("build/create/generate a landingβ¦"). Runs the full procedure below.
- - **audit-only** β read-only inspection of an existing page ("audit/check/review this landing against the checklistβ¦", no changes requested). Runs Β§0b. Never writes project files.
- - **fix-existing** β apply targeted fixes to an existing page ("fix/improve/optimize this pageβ¦"). Runs Β§0c.
- Representative routing:
- - "Build a landing for a dental clinic from this brief" β **generate**.
- - "Audit https://example.com/ against the performance checklist and report what's wrong" β **audit-only**.
- - "This landing's hero image shifts on load β fix it" β **fix-existing**.
- Generation-only brief fields (target keywords, business type, CTA, media facts) are collected ONLY for generate and rebuild work. An audit or a targeted fix must not be blocked or delayed by missing generation inputs, and audit-only mode must not create or modify any project files.
- ### 0a. Collect the brief (generate mode; ask if missing)
- Required before generating anything:
- - Domain / final URL β for canonical, og:url, absolute paths, JSON-LD `@id`.
- - Site identity (for `WebSite` markup, only when the page is the domain/subdomain home page): preferred site name, optional alternate names, and the canonical home URL β collected separately from the landing URL.
- - Breadcrumb trail (for `BreadcrumbList` markup, only when a real site hierarchy exists): the visible breadcrumb trail and canonical parent URLs.
- - Page language, base direction, and Open Graph locale β three separate inputs, never one value copied across formats: a BCP-47 language tag for `<html lang>` (e.g. `en-US`, `ar-SA`), the base direction (`ltr`/`rtl` β RTL documents get `dir="rtl"` on `<html>`, and `lang` alone does not set directionality), and the Open Graph locale in `language_TERRITORY` format for `og:locale` (e.g. `en_GB`). Ask when direction is unknown for an RTL-capable language (tech-spec Β§2).
- - Topic + 1β3 target keywords β for H1, title, description.
- - Business type: Organization or LocalBusiness. For LocalBusiness collect the verified public/legal business name and the complete structured postal address (street, locality, region, postal code, country), plus phone and geo coordinates. Also collect the verified schema.org subtype(s) based on the actual business (e.g. `Restaurant`, `Dentist`, `HardwareStore`) β never chosen from target keywords. Never invent missing identity facts: fall back to `Organization` markup or omit entity markup until the facts are provided.
- - CTA and contacts (phone, form, messengers). When a form is requested, also collect its submission destination and method (a first-party endpoint or a documented form service β never invented), the consent/privacy text required for the collected personal data, and where submissions are stored and who owns them; with no destination, the form is omitted or explicitly stubbed (tech-spec Β§10 form submission contract).
- - A brand-approved favicon or explicit permission to create one β never invent a brand mark silently.
- - Approved source material and a claim owner for objective marketing facts (numbers, prices, qualifications, guarantees, comparisons, case studies) β without them such claims are omitted, never invented.
- - Whether images are provided; whether FAQ / reviews / video blocks are needed. For a video block collect source-backed facts: video URL/ID, title, description, accurate first-publication date/time with timezone, and a unique crawlable thumbnail (plus `contentUrl` when applicable). Never invent missing media facts. Also collect the video mode with its trade-off stated: click-only facade (default β privacy/performance; the page will not satisfy Google's video discovery requirements and no video-search benefit is claimed) or SEO-discoverable (self-hosted `<video>` or a documented direct embed β required when video search traffic matters) (tech-spec Β§9).
- If domain or keywords are missing β ask first, do not invent them.
- ### 0b. Audit-only mode (read-only)
- Audit an existing page without generating a replacement. No project files are created or modified in this mode β the deliverable is a report.
- 1. Identify the target: a deployed URL (preferred β lets every check run against reality) or local HTML files. If neither is supplied, ask; never guess the target.
- 2. Run the applicable checks from Β§5 against the target as-is: W3C validity, local asset existence (for local files), JSON-LD via a schema.org validator, Lighthouse against the served/deployed URL, crawlability contract (robots.txt + sitemap.xml at the deployed host), and the manual accessibility checks in tech-spec Β§8.
- 3. Report evidence per check: pass/fail with the measured value or observed markup, and the exact command/tool used. Where a check cannot run (no deployed URL for Lighthouse, no robots.txt on the host), report a blocker for that check β do not estimate, extrapolate, or omit it silently.
- 4. Distinguish syntax validity from Google feature eligibility (FAQPage, VideoObject, rich results): valid markup is reported as valid markup, never as an achieved search feature.
- 5. Optionally end with a prioritized fix list. Applying fixes is a separate **fix-existing** request β do not start editing without it.
- ### 0c. Fix-existing mode
- Apply targeted fixes to an existing page without a full rebuild.
- 1. Identify the page/files and the specific problems to fix; collect only the inputs those fixes need (never the full generation brief).
- 2. Apply each fix per the relevant tech-spec section, preserving unrelated markup and content.
- 3. STOP POINT (Β§4) applies: show the changed page before validation. If fixes change what the user approved earlier, obtain renewed approval before reporting.
- 4. Validate the changed page (Β§5) and report measured evidence only.
- ## Generation workflow (generate mode)
- ### 1. Create the project folder
- Every project lives in its own folder inside the workspace β **never write to the workspace root**. The output is a multi-file project: every local resource referenced by the HTML must exist as a real file.
- ```
- <workspace>/<project-slug>/
- index.html # the generated landing page
- styles.css # only when below-the-fold CSS is deferred (Β§1); absent when all CSS is inlined
- script.js # only when the page uses JS (Β§10); single file, defer
- images/ # every image variant referenced in src/srcset/preload/OG tags (AVIF/WebP/JPEG, all breakpoints)
- favicon.png # stable square brand icon, β₯48Γ48
- ASSETS.md # rights & provenance record for every asset
- robots.txt
- sitemap.xml
- SERVER-SETUP.md # hosting instructions
- ```
- Image branch:
- - Images provided in the brief β produce all required variants (AVIF/WebP/JPEG at every breakpoint named in `srcset`) from them.
- - No images available β request them from the user or omit the image/block. Never emit a successful-looking asset URL without producing the file or explicitly asking for it β a referenced-but-missing file is a generation failure, not a placeholder.
- ### 2. Generate the page
- Build `index.html` strictly following [references/tech-spec.md](./references/tech-spec.md) β 13 requirement sections (performance, HTML structure, SEO, security, CSS/fonts, forbidden list, testing, accessibility, embedded video, typical blocks, deferred widgets, content truthfulness & provenance, input sanitization & output encoding).
- Treat every brief value as untrusted: encode it for its exact output context (HTML text, attribute, URL, JSON-LD), allow-list URL schemes (reject `javascript:`/unexpected `data:`), escape `<` in serialized JSON-LD, and validate structured IDs (e.g. YouTube `^[A-Za-z0-9_-]{11}$`) before they reach any URL (tech-spec Β§13).
- For embedded YouTube video use the facade pattern by default; the SEO-discoverable mode (self-hosted `<video>` or a documented direct embed) is an explicit brief choice with a disclosed trade-off, never a silent switch β rules and the Google discovery requirements in tech-spec Β§9, facade reference implementation in [references/video-facade.md](./references/video-facade.md). Maps follow the facade rule only (tech-spec Β§10): a local screenshot in the initial DOM, the iframe inserted only on explicit activation β never a native `loading="lazy"` map iframe. Reference: [references/map-facade.md](./references/map-facade.md).
- ### 3. Generate companion files
- - `robots.txt` at the site root with a fully qualified `Sitemap:` line, never blocking the canonical page or required media.
- - `sitemap.xml` with XML-escaped absolute canonical `<loc>` URLs matching the HTML canonical; `lastmod` only from a verifiable significant-content-change timestamp (omit when unknown β never use generation time blindly).
- - Hosting instructions from [references/server-config.md](./references/server-config.md): caching, Brotli/gzip, security headers.
- ### 4. STOP POINT β user approval
- Show the generated page to the user and ask explicitly whether the HTML version is OK. **Do not proceed to validation and the final report until the user confirms.** If there are remarks β fix and ask again.
- ### 5. Validate
- Run the executable validation contract from tech-spec Β§7 β pinned commands against the served page, measured results only, explicit BLOCKER when a gate cannot run:
- - W3C HTML validity (Nu validator, JSON output; zero errors).
- - Local asset/link existence: extract every local URL referenced by the output (img `src`/`srcset`, `<source>` `srcset`, preload `href`/`imagesrcset`, favicon, OG/Twitter images, CSS `url()`, script `src`) and verify each file exists in the project folder. Any missing referenced local resource is a hard failure β produce the file or remove the reference; never ship HTML pointing at files that were never created.
- - JSON-LD syntax (JSON parse) β separate from Google rich-result eligibility, which is checked with Rich Results Test on the deployed page.
- - Responsive screenshots at 320/768/1280/1920px, inspected for overflow and reflow.
- - Lighthouse: pinned version/profile, 3 runs, median per category, threshold β₯ 90, artifacts kept in the project's `reports/` β lab evidence only, never field Core Web Vitals and never WCAG certification.
- - Manual accessibility checks (tech-spec Β§8) β no automated tool alone determines WCAG conformance: keyboard navigation, focus order/visibility, dialog focus flow, zoom/reflow, reduced motion, semantic name-role-value, alternative-text quality, and all interactive visual states. Record pass/fail evidence per applicable WCAG 2.1 AA criterion; report unresolved items instead of silently certifying them.
- - Crawlability contract: parse `sitemap.xml`, compare every `<loc>` with the HTML canonical, check the `Sitemap:` URL in `robots.txt`, and request both deployed files successfully (HTTP 200).
- Fix any violations found before reporting. Disclose evidence honestly: every reported number comes with the exact command and artifact path that produced it; a gate that could not run is reported as `BLOCKER: <reason>` instead of a number. Never output a PageSpeed/LCP score that was not actually measured.
- ### 6. Final report
- Briefly list:
- - LCP parameters
- - PageSpeed score
- - schema.org types used in the code
- ## Main pitfalls
- - Never use external JS/CSS libraries, external fonts, or SVG images (tech-spec Β§6).
- - Never reference a local asset that was never created: every `src`/`srcset`/preload/OG URL must resolve to a real file in the project folder; missing source images are requested from the user, not invented (OUTPUT contract).
- - Never emit a raw brief value into markup: context-encode everything, reject `javascript:`/unexpected `data:` URLs, and self-test generation with hostile values (quotes, `</script>`, event-handler payloads) (tech-spec Β§13).
- - Never load YouTube iframes, maps, chats, subscription popups, or cookie banners on first load (tech-spec Β§9, Β§10, Β§11) β the single documented exception is a brief-chosen SEO-discoverable video mode with a direct embed recorded as a first-load dependency (tech-spec Β§9 Mode S).
- - All content must exist in raw HTML β nothing rendered only by JS.
- - Absolute URLs in JSON-LD, canonical, and OG tags.
- - Total JS budget β€ 15 KB, one file, `defer` before `</body>`.
README.md Β· README Β· 315 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- # SEO Landing
- ## Give your AI coding agent the capabilities of a senior Technical SEO engineer.
- An agent skill for building high-performance, technically optimized SEO landing pages.
- Turn an AI coding agent into a technical SEO specialist.
- Build and improve landing pages with:
- - π 100/100 Google PageSpeed target
- - β‘ Core Web Vitals optimization
- - π Technical SEO
- - π§© Full Schema.org structured data
- - πΌοΈ AVIF image optimization
- - π¨ Critical CSS
- - π§Ή Zero third-party requests on first load
- - π± Mobile-first performance
- - π€ Semantic HTML
- π·πΊ [ΠΠΏΠΈΡΠ°Π½ΠΈΠ΅ ΡΠΊΠΈΠ»Π° Π½Π° ΡΡΡΡΠΊΠΎΠΌ ΡΠ·ΡΠΊΠ΅](./README.ru.md)
- ## Works with AI coding agents
- Designed for agentic coding workflows and compatible with Agent Skillsβstyle environments.
- ## What it does?
- The skill guides an AI coding agent through the complete landing-page workflow:
- 1. Analyze the existing page
- 2. Fix technical SEO issues
- 3. Optimize HTML structure
- 4. Improve Core Web Vitals
- 5. Optimize images and fonts
- 6. Add structured data
- 7. Remove unnecessary dependencies
- 8. Validate the final implementation
- ## Why?
- AI can generate a beautiful landing page in seconds.
- The problem is that generated pages often contain:
- - unnecessary JavaScript
- - oversized images
- - poor semantic structure
- - missing structured data
- - weak metadata
- - performance bottlenecks
- - technical SEO mistakes
- SEO Landing Skill gives the agent a repeatable technical SEO workflow instead of relying on generic prompting.
- An agent skill that builds and updates landing pages toward **100/100 lab PageSpeed** and Core Web Vitals-friendly performance, and gets the technical SEO right. Static HTML, critical CSS, AVIF images, full schema.org markup, zero third-party requests on first load (deferred widgets, when used, are consent-gated and documented in the dependency manifest). These are optimization targets, not guaranteed outcomes β results depend on content, hosting, devices, and real traffic; Core Web Vitals status itself is determined by Google from field (RUM) data, not by lab tools.
- ```text
- BEFORE
- Landing page
- β
- LCP: 4.2s
- Performance: 61
- Missing schema
- Large PNG
- Render-blocking CSS
- β SEO LANDING SKILL β
- AFTER
- Landing page
- β
- LCP: 1.1s
- Performance: 100
- Schema.org β
- AVIF β
- Critical CSS β
- Semantic HTML β
- ```
- Format β [Agent Skills](https://agentskills.io) (open standard originally developed by Anthropic): works in VS Code Copilot, Claude Code, OpenAI Codex, Google Antigravity, Cursor, Gemini CLI, OpenClaw, Hermes, and other compatible agents.
- ## What's inside
- ```
- seo-landing/
- βββ SKILL.md # Main workflow: mode routing β brief β generation β stop point β validation β report
- βββ references/
- β βββ tech-spec.md # Technical spec (13 requirement sections + executable validation contract)
- β βββ server-config.md # Server config: caching, Brotli/gzip, security headers (Nginx/Apache)
- β βββ video-facade.md # Reference implementation of the "facade" pattern for YouTube
- β βββ map-facade.md # Reference implementation of the "facade" pattern for map embeds
- βββ benchmark/ # Lab benchmark disclosure: fixture, raw Lighthouse reports, methodology
- βββ tests/
- βββ fixtures/broken-landing/ # Negative fixture: every validator gate must fail on it
- ```
- ## Key requirements enforced by the skill
- - **Performance**: LCP < 2.5s, INP < 100ms, CLS < 0.1; critical CSS inlined, the rest loaded async; the LCP image loads eagerly with `fetchpriority="high"` (a responsive `imagesrcset`/`imagesizes` preload only when measurement shows a benefit)
- - **Images**: AVIF β WebP β JPEG via `<picture>`, `srcset`/`sizes`, `width`/`height`, `loading="lazy"` for below-the-fold images only (never on the LCP image), breakpoints 320β1920
- - **SEO**: title/description/canonical/robots, Open Graph, Twitter Card, JSON-LD (`WebSite`, `Organization`, `BreadcrumbList`, `FAQPage`, `VideoObject` β emitted only when fact-backed and reported per the chosen video mode, never as a guaranteed search feature)
- - **Accessibility**: WCAG 2.1 AA, contrast β₯ 4.5:1, keyboard navigation, `prefers-reduced-motion`
- - **Security**: `X-Content-Type-Options`, `X-Frame-Options`, `Permissions-Policy`, `Referrer-Policy`, per-page CSP, staged HSTS, HTTPS enforcement, `rel="noopener noreferrer"`
- - **Fonts**: system fonts only β no external fonts, no Google Fonts
- - **JS budget β€ 15 KB** for the first load, one file with `defer`; deferred third-party widgets (when used) are consent-gated, excluded from the budget, and disclosed in the dependency manifest
- - **Forbidden**: external JS/CSS libraries, SVG images, synchronous scripts, iframes on first load
- - **Video & maps**: "facade" pattern by default β the cover is a local responsive image (eager when it is the LCP/above the fold, lazy below the fold), the iframe loads only on click. Click-only video trades Google video discovery for pre-activation privacy/performance; an opt-in SEO-discoverable mode (self-hosted `<video>` or a documented direct embed) exists when video search matters (tech-spec Β§9). Maps remain facade-only.
- - **Common blocks without JS**: FAQ via `<details>`, slider via `scroll-snap`, modal via `<dialog>`
- - **Stop point**: before validation and the final report, the skill always asks the user to confirm the HTML version
- ## Real-world result (lab measurements β not field Core Web Vitals)
- One reproducible lab benchmark: Lighthouse CLI 13.4.1, mobile emulation, simulated throttling. Original β an Angular SPA (102 requests, 1 run); the same page rebuilt as static HTML with this skill (17β18 requests, median of 5 runs). Full disclosure β fixture, exact flags, raw JSON reports, timestamps, and aggregation method β is in [benchmark/README.md](./benchmark/README.md).
- | Metric | Original | Rebuilt | Gain |
- |---|---:|---:|---:|
- | Performance | 75 | **100** | +25 |
- | Accessibility | 75 | **100** | +25 |
- | Best Practices | 73 | **100** | +27 |
- | LCP | 3.3 s | 1.6 s | β52% |
- | Total Blocking Time | 490 ms | 13 ms | β97% |
- | TTI | 9.4 s | 1.6 s | β83% |
- | Data transferred | 769 KiB | 110 KiB | β86% |
- | Requests | 102 | 17 | β83% |
- These are **lab** numbers for one before/after pair. They are not field Core Web Vitals (Google determines CWV status from CrUX/RUM field data at the 75th percentile, and Lighthouse cannot measure INP without real interaction), and they are not a guarantee that another page, host, device, or audience will reach the same results β the skill's targets are optimization goals, not promised outcomes.
- ## Installation
- The skill is hosted at [github.com/aleksandr-alhoff/seo-landing](https://github.com/aleksandr-alhoff/seo-landing). Clone the repository β the repo root is the skill folder itself β then run the commands below from the directory that contains `seo-landing/`:
- ```bash
- git clone https://github.com/aleksandr-alhoff/seo-landing.git
- ```
- ### Global (available in all projects)
- Every recipe below is self-contained: it creates its destination directory first, then copies the skill into it. Each one must exit with status `0` and leave the layout `<skills-dir>/seo-landing/SKILL.md` in place.
- Note: `cp -R` from a git clone also copies the clone's `.git` directory into the installation β harmless, but unnecessary. To keep installations lean, replace `cp -R seo-landing <dir>/` with `rsync -a --exclude=.git seo-landing <dir>/seo-landing/` (the same sync used for updates below).
- ```bash
- # VS Code Copilot
- mkdir -p ~/.copilot/skills
- cp -R seo-landing ~/.copilot/skills/
- # Claude Code
- mkdir -p ~/.claude/skills
- cp -R seo-landing ~/.claude/skills/
- # OpenAI Codex CLI / ChatGPT desktop
- mkdir -p ~/.agents/skills
- cp -R seo-landing ~/.agents/skills/
- # Cursor
- mkdir -p ~/.cursor/skills
- cp -R seo-landing ~/.cursor/skills/
- # Gemini CLI
- mkdir -p ~/.gemini/skills
- cp -R seo-landing ~/.gemini/skills/
- # Google Antigravity
- mkdir -p ~/.gemini/config/skills
- cp -R seo-landing ~/.gemini/config/skills/
- # OpenCode
- mkdir -p ~/.config/opencode/skills
- cp -R seo-landing ~/.config/opencode/skills/
- # OpenClaw (via CLI β installs into the shared ~/.openclaw/skills)
- openclaw skills install git:aleksandr-alhoff/seo-landing --global
- # or from a local clone: openclaw skills install ./seo-landing --global
- # or manually:
- mkdir -p ~/.openclaw/skills
- cp -R seo-landing ~/.openclaw/skills/
- # Hermes
- mkdir -p ~/.hermes/skills
- cp -R seo-landing ~/.hermes/skills/
- ```
- OpenClaw also picks up skills from `~/.agents/skills` (the Codex path above), and Hermes can scan it too if you add `~/.agents/skills` to `skills.external_dirs` in `~/.hermes/config.yaml`.
- Gemini CLI can also install straight from a Git repository:
- ```bash
- gemini skills install https://github.com/aleksandr-alhoff/seo-landing.git
- ```
- ### Per project (workspace only)
- ```bash
- # Shared .agents/skills β picked up by VS Code Copilot, Codex, Cursor, Antigravity, Gemini CLI, OpenCode, OpenClaw, Hermes
- mkdir -p .agents/skills
- cp -R seo-landing .agents/skills/
- # Claude Code
- mkdir -p .claude/skills
- cp -R seo-landing .claude/skills/
- # VS Code Copilot (GitHub-style location)
- mkdir -p .github/skills
- cp -R seo-landing .github/skills/
- # OpenClaw (workspace skills β highest precedence)
- # via CLI: openclaw skills install ./seo-landing
- mkdir -p skills
- cp -R seo-landing skills/
- # Hermes (project-local; then trust the repo once: hermes skills trust)
- mkdir -p .hermes/skills
- cp -R seo-landing .hermes/skills/
- ```
- ## Updating, verifying, and uninstalling installed copies
- Two facts drive everything below:
- 1. **`git pull` changes only the clone.** An installed copy made with `cp -R` is independent β pulling the source clone does NOT update any installation.
- 2. **`cp -R` copies, it does not synchronize.** GNU `cp` has no destination-sync/removal behavior: a file deleted upstream stays in the installed copy forever, producing a mixed release of old and new instructions.
- So an update is an explicit, bounded, idempotent sync into the resolved skill destination β never a blind re-copy.
- ### Destinations (one per client and scope)
- | Client / scope | Installed skill destination (`$DEST`) |
- |---|---|
- | VS Code Copilot β global | `~/.copilot/skills/seo-landing` |
- | Claude Code β global | `~/.claude/skills/seo-landing` |
- | OpenAI Codex CLI / ChatGPT desktop β global | `~/.agents/skills/seo-landing` |
- | Cursor β global | `~/.cursor/skills/seo-landing` |
- | Gemini CLI β global | `~/.gemini/skills/seo-landing` |
- | Google Antigravity β global | `~/.gemini/config/skills/seo-landing` |
- | OpenCode β global | `~/.config/opencode/skills/seo-landing` |
- | OpenClaw β global | `~/.openclaw/skills/seo-landing` |
- | Hermes β global | `~/.hermes/skills/seo-landing` |
- | Shared per-project (Copilot, Codex, Cursor, Antigravity, Gemini CLI, OpenCode, OpenClaw, Hermes) | `.agents/skills/seo-landing` |
- | Claude Code β per project | `.claude/skills/seo-landing` |
- | VS Code Copilot β per project (GitHub-style) | `.github/skills/seo-landing` |
- | OpenClaw β workspace (highest precedence) | `skills/seo-landing` |
- | Hermes β per project | `.hermes/skills/seo-landing` |
- ### Update (idempotent sync, removes upstream-deleted files)
- Run from the directory that contains the updated `seo-landing/` clone, with `$DEST` set to the destination from the table above:
- ```bash
- DEST=~/.copilot/skills/seo-landing # β substitute the right destination
- # 1. Recovery first: back up the current installation (the sync is destructive).
- cp -R "$DEST" "$DEST.backup-$(date +%Y%m%d)"
- # 2. Bounded idempotent sync: copy new/changed files AND delete files inside
- # $DEST that no longer exist upstream. --delete only ever touches $DEST.
- rsync -a --delete --exclude=.git seo-landing/ "$DEST/"
- # 3. Verify: zero differences (exit status 0) and the installed SKILL.md
- # matches the source revision byte-for-byte.
- diff -qr -x .git seo-landing "$DEST" && cmp "$DEST/SKILL.md" seo-landing/SKILL.md
- ```
- Notes:
- - `rsync -a --delete` is available out of the box on macOS (openrsync) and on typical Linux systems; `--exclude=.git` keeps the clone's history out of the installation.
- - The backup in step 1 is the recovery path: if the new version misbehaves, restore it with `rm -rf "$DEST" && cp -R "$DEST.backup-<date>" "$DEST"`. (macOS openrsync does not support `--backup`, so the explicit copy is the verified mechanism.)
- - The sync is safe to re-run at any time β running it twice in a row changes nothing the second time.
- - Tested upgrade path: an installation containing a file that the new revision removed loses that file after the sync, and the installed `SKILL.md` becomes byte-identical to the source revision.
- ### Verify an existing installation at any time
- ```bash
- diff -qr -x .git seo-landing "$DEST"
- ```
- Exit status `0` (no output) means the installation matches the clone exactly. Any printed difference is a stale, modified, or extra file β re-run the update sync to resolve it. To check which specification version is installed, read the `Version` line at the top of `$DEST/references/tech-spec.md` (the single source of truth for versions).
- ### Reload / restart after an update
- Agents read skills when a session starts. After updating, **start a new chat session** (or restart the agent CLI) before relying on the new version. Clients that gate skills behind trust/approval (e.g. Hermes: `hermes skills trust`) may require re-trusting the updated copy.
- ### Uninstall
- ```bash
- rm -rf "$DEST" # the destination from the table above
- ```
- If the copy was installed through a client CLI (`openclaw skills install`, `gemini skills install`), prefer that client's own uninstall command when it provides one; otherwise removing the destination directory is sufficient. Also remove any backups (`$DEST.backup-*`) you no longer need.
- ### Symlinked installations β optional/experimental
- Symlinking the clone into a skills directory (so `git pull` updates it in place) is possible in principle, but official symlink support, trust handling, and reload behavior differ per client and are NOT verified here β treat this as experimental. The rsync sync above is the supported, client-independent update path.
- Once installed, the skill is picked up automatically by its description β just ask your agent to "build a landing page from a brief with focus on SEO and PageSpeed".
- ## Usage
- 1. Give the agent a brief: domain, language (BCP-47 tag, base direction for RTL languages, and Open Graph locale β three separate inputs), topic and keywords, business type, CTA and contacts.
- 2. The skill creates a separate project folder (`<workspace>/<project-slug>/`) with `index.html` and every local asset it references (all image variants, favicon), plus β conditionally β `styles.css` (deferred CSS only), `script.js` (when JS is used), `robots.txt`, `sitemap.xml`, `ASSETS.md`, and `SERVER-SETUP.md`. Missing source images are requested from the user, never invented.
- 3. At the stop point, confirm the HTML version β the skill then runs validation (local asset/link existence, W3C, JSON-LD, Lighthouse as automated evidence, plus required manual accessibility checks) and produces a report: LCP, PageSpeed scores, and the schema.org types used.
- ## About the author
- **Aleksandr Alhov** β 14+ years of experience in SEO, content marketing, and GEO/AEO. Runs an SEO boutique for product teams working across EN, LATAM, MENA, and CIS markets: building in-house SEO departments turnkey, launching profitable corporate media, discovering growth hypotheses in the SEO channel, and providing hands-on SEO consulting.
- Co-founder of:
- - [Total Site Control](https://totalsitecontrol.com/)
- - [Udalenka.work](https://udalenka.work)
- - [QR Barcode Hub](https://qrbarcodehub.com)
- - [DealRocket](https://dealrocket.ru/)
- Made for the [t.me/sdelay_tam](https://t.me/sdelay_tam) channel β a cozy SEO channel for product teams. SEO questions: [t.me/alhov](https://t.me/alhov).
- ## License
- [MIT](./LICENSE) β free to use, modify, and distribute, including commercially. Just keep the copyright notice.
- Specification: see [references/tech-spec.md](./references/tech-spec.md) β the current version and change record are declared there.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If itβs helped, pass it on.