Ask a coding agent to "improve the SEO" or "make a launch video" and you get the average of the internet: a generic checklist, a purple gradient, a video that could be about any product. That's AI slop, and it bloats your work instead of moving it. You spend the afternoon cleaning it up.

A skill fixes that. It's a written method the agent follows: what to check, in what order, what good looks like, what to skip. The agent reads your code, your pages and your copy, and the output is about your product, not anyone's.

These are the skills we run on every SaaS we ship. Each one comes with a real result, so you can see what you get before you install it.

seo-audit crawls your site and reads your source, then ranks what's wrong by impact: crawlability, indexing, canonicals, sitemaps, meta tags, page speed. Every finding comes with the evidence and the line of code that causes it.

We ran it on paralect.com. It checked all 675 sitemap URLs and found that every page sent crawlers an empty , the homepage was a temporary redirect, 61 URLs were soft 404s and 3 returned HTTP 500. Some of these had been on our list for months, some for years. Nobody owned them, and "fix SEO" never beat a feature in planning. With the audit's plan in hand we shipped server rendering the next day, then canonicals, a clean sitemap and a real homepage within two weeks.

An audit tells you what's broken once. seo-autopilot is a daily agent that keeps SEO going end to end: site health checks and fixes, a keyword map, articles, comparison pages, free tools, link drafts and tracking of how AI assistants answer about you. Run it once the audit's critical items are fixed.

vibe-security audits the holes AI assistants leave in fast-built apps: API keys in the client bundle, Supabase tables without row-level security, prices and user IDs trusted from the browser, unsigned webhooks, missing rate limits. You get a self-contained HTML report with a fix for every finding.

Run it again with every new model. When Opus 5.5 came out we ran it on a codebase we had already audited, and it found a critical issue within minutes. A stronger model reads the same code and sees more, so a clean report from last quarter's model doesn't mean the app is clean. Switch models between runs too: each one catches different things.

Impeccable gives the agent a designer's vocabulary: typography, color, layout, motion, and a library of anti-patterns to avoid. Commands like /audit, /polish and /typeset turn the default "AI landing page" into a design that looks like your product. Above: the Control Panel landing page before and after. More on it in our stack .

Hyperframes renders video from HTML, so the agent can build it from your real interface instead of stock footage. Its product-launch-video skill turns a URL, a brief or a script into a storyboarded launch video with music, captions and motion. Above is our launch video for Paralect Agent, built with Hyperframes from the app's own screens: tasks running, Claude replying, a test going green.

When you don't need a storyboard, type /brag. The skill reads your project's code, picks the hero, the product and the one line worth saying, and renders a short shareable clip with Hyperframes. No live URL, no screenshots, no script. Good for a feature you shipped today and want on X tonight.

Every page you publish gets shared as a link, and most links show a logo on a flat color or nothing at all. og-image builds the share card in your own stack from your real layout, components and copy, captures it at 2x and wires it into the meta tags. Run it for every new page, not just the homepage. If you want the card to move, og-image-live makes the product come alive inside it.

A screenshot shows the hero. A scroll shows the page. website-scroll records a page top to bottom the way a person reads it, a flick, a pause, another flick, with cookie banners and chat widgets out of the way. We use it to present every new page: in the changelog, in a post, as a hover preview on a link card.

New features need a demo, and screen recordings come out shaky with a lost cursor. product-demo signs into your app and records the feature in a rounded window on a soft wallpaper, with a cursor that glides, hovers and clicks like a hand would. Describe the flow in a sentence ("open the Run modal, click each category, end on All") and you get a square MP4 ready for X or LinkedIn.

Install the ones you need with npx skills add, or tick them as tasks in Paralect Agent : Claude Code runs the skill on your project and reports back.