How do you install a skill from skills.sh?

Use the npx skills add command with the GitHub repository URL and the skill name, for example: npx skills add https://github.com/coreyhaines31/marketingskills --skill ai-seo. The exact command is shown on each listing page on skills.sh.

How is a skill different from a Cursor rule or a saved prompt?

Rules are persistent project constraints. Prompts are one-off instructions. Skills are portable packages you can add and update across projects—good when the workflow is large, maintained upstream, or shared by a community repository.

Can one founder orchestrate sub-agents across marketing, product, and engineering?

Yes in workflow terms: you route priorities while agents load specialized skills for repeatable procedures—growth copy and SEO, discovery and roadmaps, PRDs and triage-style workflows, backend (Convex), and UI shipping (Vercel). High-stakes finance and legal decisions still belong to qualified humans or counsel.

Which curated skills map to an AI engineer building agent products?

Use the AI engineer playbook on this page: it groups Anthropic generative-AI skills (Claude API, MCP builder, artifacts), LangChain and LangGraph (RAG, Deep Agents, persistence, human-in-the-loop), LangSmith (datasets, evaluators, traces), Convex and Vercel agent skills, scrape and automation add-ons, plus engineering orchestration skills for tests, triage, and PRDs. Marketing and pure product-strategy skills are separated into the GTM playbook.

SOLO OPERATOR · AGENTS · HUMAN IN THE LOOP · SKILLS.SH

Agent skills hub

Q: What is an agent skill?

An agent skill is a versioned package of procedural instructions (usually SKILL.md plus optional scripts and references) that you install into a coding agent so it follows a consistent workflow—SEO audits, schema work, or framework setup—instead of re-explaining the playbook every session.

Quick answer: Curated skills.sh packages for one person running a software business with agents—you stay the approver on strategy, risk, and narrative. This hub lists 176 maintained rows and breaks them into playbooks (operator loop, AI engineer stack, HITL and quality, GTM judgment) so you are not staring at a single flat table.

You are the human in the loop. Agents load repeatable procedures from skills.sh (npx skills add … --skill) so Cursor and Claude Code ship the same audits, PRDs, migrations, and deploys across sessions—while you route what matters. The ambition is outcome density: marketing and product skills cover acquisition and bets; engineering orchestration covers how code moves; Convex and Vercel cover data and what users see; Anthropic, LangChain, and LangSmith cover model interfaces, RAG, graphs, and evals; GEO and scrape skills feed proof and extraction. Not legal, tax, or medical advice—tooling for execution.

Skills split by mission

Prefer focused pages instead of one long view. Same dataset, no rows removed: AI engineer, UGC creator engine, HITL & quality, GTM & product judgment.

Stack for a one-person company

A tiny team wins when specialists stay in their lane: marketing agents never draft filings; coding agents never pick pricing. Route work across the areas below—then open playbooks for job-shaped groupings or scroll to the full table.

Acquire · narrative · revenue

Marketing — SEO, ads, lifecycle, social, experiments. Surface area for inbound and proof.

Decide · roadmap · economics

Product — discovery, PRDs, positioning, prioritization, SaaS metrics. What to build and why.

Orchestrate · ship discipline

Code shipping habits — PRD/issue flows, triage, TDD, domain language, git guardrails. How delivery stays sane.

Sync · backend · realtime

Convex — quickstart, auth, migrations, performance. Shared truth for agents and users.

Ship · UI · platform

Vercel — React/Next practices, deploy, RN, design guidelines. What customers touch.

Model · docs · artifacts

Generative AI (Anthropic) — Claude API, MCP builder, Office files, PDFs, web artifacts, app testing. Official anthropics/skills reference pack.

Ground · graph · evaluate

LangChain & LangGraph — RAG, Deep Agents, persistence, HITL. LangSmith — datasets, evaluators, traces. From langchain-ai/langchain-skills and langchain-ai/langsmith-skills.

Prove · crawl · extract

AI search, audits & scrape — answer-engine SEO (seo-geo), deep site audits, Firecrawl, Playwright. Complements marketing skills like ai-seo / seo-audit.

Add finance, legal, and accounting professionals where stakes require it—agent skills handle repeatable procedure, not incorporation law.

Playbooks · same library, clearer jobs

Choose a lens: the nine-beat operator loop (mirrors the orbit cards), the AI engineer syllabus—APIs, MCP, RAG, LangGraph, LangSmith, Convex, Vercel, scrape, and repo discipline—the HITL and quality cluster for checkpoints and evals, or GTM and judgment for growth and product. Each group expands into chips that open the matching skills.sh listing. The sortable table at the bottom is the full set.

What is an agent skill?

A skill ships as a small repo folder: instructions the model reads first, optional scripts, and sometimes reference docs (examples, checklists, platform-specific notes). Installers copy that bundle into your agent’s skill path so the behavior is repeatable across projects.

Skills vs rules vs prompts

Project rules (e.g. .cursor/rules) stay tied to one codebase.
Prompts are one-off instructions or library snippets.
Skills are for large, maintained playbooks you want versioned or updated from upstream.

How we pick quality listings

We favor repos with clear summaries, install activity on skills.sh, and maintained GitHub sources. Read each SKILL.md before you install.

Gen-AI model stacks you can run with skills

You asked for stronger coverage beyond base skill packs. These stacks combine skills.sh workflows with model/tool execution lanes (image, video, motion, voice) so you can ship end-to-end output as a one-person team with human approvals.

These stacks are execution bundles: model runtime + repeatable skills. Use playbooks above to map ownership, and the full table below to add each skill.

Browse & install

skills.sh lists thousands more skills (Azure, browser automation, design systems). Use the exact npx skills add line from each listing page.

Marketing & discovery — coreyhaines31/marketingskills

$ npx skills add https://github.com/coreyhaines31/marketingskills --skill ai-seo

Product — deanpeters/Product-Manager-Skills

$ npx skills add https://github.com/deanpeters/Product-Manager-Skills --skill prd-development

Engineering orchestration — mattpocock/skills

$ npx skills add https://github.com/mattpocock/skills --skill to-prd

Convex — get-convex/agent-skills

$ npx skills add https://github.com/get-convex/agent-skills --skill convex-quickstart

Vercel Labs — vercel-labs/agent-skills

$ npx skills add https://github.com/vercel-labs/agent-skills --skill vercel-react-best-practices

Generative AI (Anthropic reference) — anthropics/skills

$ npx skills add https://github.com/anthropics/skills --skill claude-api

RAG & agents — langchain-ai/langchain-skills

$ npx skills add https://github.com/langchain-ai/langchain-skills --skill langchain-rag

Evals & tracing — langchain-ai/langsmith-skills

$ npx skills add https://github.com/langchain-ai/langsmith-skills --skill langsmith-trace

GEO / site audit / scrape — resciencelab, squirrelscan, firecrawl, …

$ npx skills add https://github.com/resciencelab/opc-skills --skill seo-geo

$ npx skills add https://github.com/firecrawl/cli --skill firecrawl-scrape

Install flags differ per listing (e.g. React rulebook uses vercel-react-best-practices while the skills.sh path is react-best-practices). Use Gen AI for anthropics/skills, RAG & agents for LangChain / LangGraph, Evals & tracing for LangSmith. For GEO, audits, and scraping from mixed repos, use Search & scrape—especially playwright-cli, which may be thin; verify before relying on it.

176 curated skills

Full directory: every curated row is below. Orbit counts are computed from the same data as the filters. For AI engineer and human-in-the-loop groupings, start with the playbooks section—then use the chips or these filters.

Grow → decide → orchestrate → sync → ship → model → ground → measure → extract

Card counts match the table. Use Playbooks for the AI engineer syllabus and HITL clusters without memorizing repo codes.

How to read the table: the first column is the kind of work (you do not need short codes). The second line in each cell is the skill-id you pass to npx skills add. Filters match the same groups as the cards above.

What it is · skill name	Listing	Notes

Go deeper on skills.sh for leaderboards and search beyond this set.

CuratedAI prompts

The library ships 293 prompts total; 25 sit in Agent skills & workflows (MCP specs, eval harnesses, API briefs, governance snippets). Use prompts for one-off copy; install skills when you want a maintained playbook inside the agent.

Open prompts — Agent skills