We've been building AITasker with Claude Code as a daily collaborator — not just for writing code, but for design, copy, positioning, SEO, and editorial review. Along the way we ended up codifying a lot of the "how we actually do this" knowledge into skills — reusable instruction sets that teach Claude to perform a specific task the way we want it done.
Today we're open-sourcing seven of them.
👉 github.com/bleshinsky/aitasker-co/tree/main/claude-code-skills
What is a skill, exactly?
A Claude Code skill is a folder with a SKILL.md file in it. The file contains a methodology — rules, patterns, checklists, examples — that Claude loads when you ask it to do something matching the skill's description. That's it. No plugins, no installation, no build step. Just prompts, versioned in Git.
You copy the folder into your project's .claude/skills/ directory, and the next time you say something like "build a glassmorphism hero section" or "write positioning for my product," Claude picks up the relevant skill and follows it.
Think of them as expert playbooks. The magic is that they turn one-off prompts into reproducible output quality — every time, across every teammate, across every session.
Why we're open-sourcing them
A few reasons.
They're genuinely useful. These aren't toy demos. They're the same playbooks we use to ship AITasker itself — our marketing pages, our LinkedIn outreach, our landing pages, our editorial standards. If they work for us, they'll work for you.
Skills are prompts, not IP. The interesting work in AITasker is the marketplace, the prototype-as-bid pipeline, the evaluation engine — not the fact that we told Claude "don't write 'in today's fast-paced world.'" Keeping these private wouldn't protect anything; sharing them gives the community a head start.
Better skills, faster. The Claude Code ecosystem is young. The more teams publish battle-tested skills, the faster everyone gets to good defaults.
What's in the box
Design systems
glassmorphism-design-system — Frosted-glass UI components done right. Cards, navbars, modals, hero sections with blur, transparency, and depth. Includes design tokens, accessibility rules, and performance budgets so you don't accidentally ship a page that chokes on mobile Safari. Brand-configurable.
premium-saas-design-system — The one we lean on most. Awwwards-level marketing pages: typography as architecture, motion choreography, conversion patterns. It's built around a Seven Pillars framework and includes a 68-item quality checklist the skill runs against before handing work back. This is why AITasker's marketing pages don't look like every other SaaS template.
Marketing & strategy
saas-positioning — A full April Dunford positioning methodology, compressed into a skill. Competitive analysis, messaging hierarchy, sales narrative, and a 30/60/90 day rollout plan. Output is a 14-section positioning document you can actually act on, not a deck.
seo-landing-pages — SEO-optimized Next.js landing pages. Keyword strategy, JSON-LD structured data, a 9-section page template, CRO patterns, and a post-build audit step. We use this to ship our programmatic category and task-type pages.
website-redesign — Audit and redesign existing sites with a combined design-excellence + CRO lens. Seven-dimension scoring, an experiment bank, and recommendations that are specific enough to hand to an engineer.
Writing
linkedin-recommendation — Writes LinkedIn recommendations that actually sound like you wrote them. Five-beat structure, anti-cliché rules, tone matching via an optional writing-style template. No "hard worker and team player." No "it's rare to find someone who…".
Writing quality
stop-slop-evaluator — Our internal filter for generic AI filler. 45+ banned phrases, a specificity test, and a "slop score." We run it on evaluator justifications inside AITasker, but it works on any written output — blog drafts, emails, product copy.
How to use them
# From your project root
git clone https://github.com/bleshinsky/aitasker-co.git /tmp/ait
cp -r /tmp/ait/claude-code-skills/saas-positioning .claude/skills/
cp -r /tmp/ait/claude-code-skills/seo-landing-pages .claude/skills/
Then just ask Claude to do the thing:
- "Create positioning for a dev-tool SaaS targeting staff engineers at Series B startups."
- "Build an SEO landing page for the keyword 'AI meeting notes for law firms.'"
- "Write me a LinkedIn recommendation for Sarah, my former PM at Stripe."
Claude matches your request to the right skill automatically based on the skill's description metadata.
Some skills (the two design systems) read from a brand-config.md file in your project's .claude/ directory. There's a brand-config.example.md in each skill folder — copy it, fill in your colors, fonts, and voice, and the skill adapts to your brand.
What's not in there (yet)
We held back a few skills that are too specific to AITasker's internal evaluation pipeline to be generally useful — image scoring rubrics and rubric interpreters that depend on our internal dimensions. We'll keep iterating on those privately and may release cleaned-up versions later.
Everything in the repo is MIT-licensed. Fork it, edit it, build on top of it.
Contributing back
If you improve a skill — better rules, cleaner examples, a gap you plugged — send a pull request. We'd rather have ten contributors sharpening a stop-slop-evaluator than keep ours in isolation.
Repo: github.com/bleshinsky/aitasker-co/tree/main/claude-code-skills
If you end up using one of these on something cool, let us know — we're genuinely curious what they produce outside our four walls.