Agent Skills Marketplaces

What is it

Agent Skills Marketplaces are platforms and libraries that package AI agent capabilities into distributable, reusable modules. Instead of every developer hand-crafting prompts and tool integrations from scratch, a skill marketplace lets them install a pre-built "skill" — say, a web-scraping routine, a PDF parser, or a CRM-update workflow — the same way npm packages or Shopify apps get installed today. The technical essence is a standardized skill format (think a manifest file plus executable code plus a prompt template) that any agent runtime can load. The business significance is bigger: it turns prompt engineering from a craft into a commodity, creating a two-sided market where skill authors earn money and agent builders save weeks of work. WorkSkill and addyosmani/agent-skills are early examples — one commercial, one open-source — and both point at the same emerging layer: the "app store" moment for AI agents.

Why now

Three forces converge in late 2026. First, agent runtimes have standardized enough that a skill written for one framework can plausibly run on another — MCP (Model Context Protocol) adoption gave us a common transport layer, and the industry is now arguing over the skill packaging layer above it. Second, the cost of running agents collapsed roughly 10x over the past 18 months, so the bottleneck moved from model cost to integration labor. Developers now spend more time wiring tools than tuning prompts. Third, the freelance prompt-engineering market proved people will pay for AI expertise, but it doesn't scale — a marketplace does. The 100% growth rate and "nascent" stage in our data confirm this is the inflection point, not the peak. Last year the runtimes were too fragmented; next year the incumbents will have shipped their own stores. The window is roughly the next 6-12 months.

Market Evidence

The signal here is thin but directional: 2 independent sources (GitHub and V2EX), 4 total mentions, 100% growth rate, stage classified as nascent. Read that honestly — 4 mentions is not demand, it's an early whisper. But the composition matters. GitHub activity around addyosmani/agent-skills suggests developers are already building and sharing skill packages without a marketplace to host them, which is exactly the pre-market behavior you want to see. V2EX discussion adds a Chinese-developer-community signal, meaning this isn't a single-ecosystem phenomenon. Compare this to where MCP was 18 months before it exploded: similar mention counts, similar "why is nobody building the registry" threads. The risk is that this is a feature, not a company — OpenAI, Anthropic, or LangChain could ship a built-in skill store and flatten the standalone opportunity. The 64/100 trend score reflects that tension: real momentum, unproven durability.

Who's Behind It

The early movers split into three camps. Commercial: WorkSkill is positioning as a paid skill marketplace, essentially the "Envato for agents." Open-source: addyosmani/agent-skills (from Addy Osmani, a well-known Google engineer) gives the movement credibility and a reference implementation. Platform players: Anthropic with MCP, OpenAI with its Assistants/tool ecosystem, and LangChain with its hub are all adjacent whales who could absorb this layer. The community gravity is on GitHub and in developer Discord/Slack groups, with V2EX showing Asian-market interest. No single company owns the category yet — that's the opportunity and the danger. If Anthropic ships a first-party skill registry inside Claude, a standalone marketplace's only moat is cross-platform neutrality and curation quality.

TAM & Market Size

Buyers come in three tiers. Tier 1: indie agent developers and small AI startups — maybe 200,000-500,000 globally — who'd pay $10-50/month for a skill library that saves them integration time. Tier 2: enterprises building internal agents, with real budgets ($500-5,000/month) but long sales cycles and security requirements. Tier 3: skill authors themselves, who pay a marketplace commission (15-30%) to distribute and monetize. Our demand score sits at 0/100, which is honest: nobody has proven willingness to pay yet. But the analogy is instructive — the Shopify App Store pays out over $1B annually to developers, and the npm ecosystem proves developers adopt package registries fast when installation is one command. Realistic near-term TAM for a focused indie play: $5-20M ARR ceiling in year one, expanding if cross-platform standards hold. Price tolerance is highest among time-poor solo founders, lowest among enterprises until you have SOC 2 and audit logs.

Competitive Landscape

Direct competition is nearly nonexistent — WorkSkill and a handful of GitHub repos are the entire field, which is why our competition score reads 0/100 (a gift, not a verdict). Adjacent threats are the real story. Anthropic's MCP registry, OpenAI's GPT Store, and LangChain Hub all sit one product decision away from eating this category. Their strength: distribution and default placement. Their weakness: each is walled to its own ecosystem, and developers increasingly want framework-agnostic skills. That's your gap — be the Switzerland of agent skills, the registry that works with Claude, GPT, and open-source runtimes alike. Secondary competitors are generic code marketplaces (GitHub Marketplace, Gumroad) that host skills incidentally but offer no discovery, versioning, or runtime integration. If Big Tech enters seriously, you have roughly 6-12 months of runway. Move fast, own the cross-platform narrative, and build switching costs through author loyalty programs before the walls go up.

Business Model

Go marketplace-plus-subscription hybrid, not pure either. Pure marketplace is a cold-start death sentence — you need supply and demand simultaneously. Instead: free tier with 5 skill installs/month to seed adoption, then a $19/month Pro tier for unlimited installs plus private skills, and a $99/month Team tier with shared skill libraries and usage analytics. Take a 20% commission on paid skill sales to incentivize authors, undercutting the 30% app-store standard to attract supply early. Why this fits: developers already pay $20-30/month for tools like Cursor and GitHub Copilot, so $19 is inside the mental budget, and commission revenue scales without extra CAC. 12-month forecast — conservative: 800 paying users × $19 = $182K ARR; base: 3,000 users + $40K commission = $724K ARR; optimistic: 10,000 users + $150K commission = $2.4M ARR. Blended CAC estimate: $40-70 via developer content and GitHub-adjacent channels, giving a 2-3 month payback at the $19 tier. Keep it annual-discounted (2 months free) to pull cash forward.

MVP Blueprint

Ship in 5-7 days. Core features only: (1) a skill manifest spec — a simple YAML/JSON file declaring name, version, runtime compatibility, permissions, and entry point; (2) a CLI (npx skill install <name>) that fetches and drops skills into a local directory; (3) a web registry with search, install counts, and a one-click copy of the install command; (4) Stripe-powered paid skills with author payouts; (5) basic auth and a "publish" flow via GitHub OAuth. Cut everything else — no ratings, no comments, no analytics dashboard, no enterprise SSO. Tech stack: Next.js on Vercel for the registry, Postgres (Supabase) for skill metadata, Stripe Connect for payouts, and a thin Node CLI published to npm. The fastest path to launch is to seed the registry by scraping and curating the best existing open-source skills from GitHub, so day-one visitors see 50+ skills instead of an empty store. Launch on Product Hunt, Hacker News, and the MCP Discord. Suggested product types — SaaS, Tool, API — all fit; start with the CLI+registry combo.

Commercial Opportunities

Three concrete directions. First, a curated cross-platform skill registry targeting indie agent developers — expected $5K-25K MRR within 6 months if you hit 1,000-3,000 paying users, because discovery and trust are the scarcest resources in a fragmented ecosystem. Second, a white-label skill store for enterprises — sell the marketplace software itself to companies wanting an internal skill catalog, priced $2K-10K/month per seat-block; this beats consumer plays because enterprises pay for governance and audit trails, not just convenience. Third, a skill-authoring toolkit and certification — sell courses, templates, and a "verified skill author" badge at $199 one-time, monetizing the supply side before the demand side matures. The registry is the wedge; the enterprise white-label is where the real money hides. Each direction has a distinct buyer, so you can test all three cheaply without cannibalizing focus.

Product Ideas

🥇 SkillHub — "npm for AI agent skills, cross-platform." Target indie agent developers who waste days wiring tools. One command installs a vetted skill into any runtime. Why now: MCP standardized transport, but nobody owns the packaging layer, and GitHub repos prove organic supply exists.

🥈 SkillForge — "Author once, sell everywhere." Target prompt engineers and AI consultants who want recurring income. Upload a skill, set a price, get paid across multiple agent runtimes via one dashboard. Why now: the freelance prompt market is hot but unscalable; a distribution layer captures the value that currently leaks to Upwork.

🥉 SkillGuard — "Security scanning and permissions for agent skills." Target enterprise AI teams terrified of installing third-party code into production agents. Static analysis, permission manifests, and a trust score for every skill. Why now: the moment skills become installable, supply-chain attacks become inevitable — this is the npm-audit moment waiting to happen, and it's a natural paid add-on to any registry.

SEO Opportunity

Search volume for "agent skills," "MCP skills," and "AI agent marketplace" is climbing from a low base — early but compounding, mirroring "MCP server" searches in early 2025. Target long-tail keywords: "install AI agent skills," "cross-platform agent skill registry," "how to publish an agent skill," "MCP skill marketplace," and "agent skill manifest format." SEO difficulty registers 0/100, meaning almost no competition — you can rank on page one within weeks by publishing authoritative docs and comparison posts. Content strategy: own the specification. Publish the canonical "Agent Skill Manifest Spec" and let everyone link to you.

Risk Assessment

When is this thesis wrong? If the major platforms each keep skills walled inside their own stores, cross-platform neutrality becomes worthless and the standalone marketplace dies. Top three risks: (1) Platform absorption — Anthropic or OpenAI ships a first-party registry and defaults every user into it; (2) Market timing — the category is genuinely too early, developers keep hand-rolling skills, and paying demand never materializes beyond a few hundred enthusiasts; (3) Execution — you solve the cold-start problem badly, ending up with a graveyard of unmaintained skills and no trust. Validate cheaply before building: post a "would you pay $19/month for this" landing page and run it against Hacker News traffic for two weeks; count email signups, not compliments. Talk to 20 agent developers and ask what they spent last month on integration labor. Walk away if fewer than 50 people sign up or if a major platform announces a first-party store during your validation window.

Action Plan

Today: publish a landing page describing SkillHub with an email waitlist and a fake "install" command, then post it to Hacker News and the MCP Discord. This costs $0 and takes 3 hours. Low-cost validation: over the next 7 days, DM 20 agent developers asking one question — "What did you build by hand last month that you'd have paid to install?" If 5+ name the same category of skill, you have a wedge. If signal confirms: build the CLI + registry MVP in week 2-3, seed 50 curated skills, and launch publicly. Timeline — Week 1: landing page, 100 waitlist emails, 20 interviews. Month 1: MVP live, 500 installs, first 10 paying users. Month 3: 1,000 paying users or $10K MRR, plus one enterprise pilot signed. If you miss the month-3 bar by half, pivot to the enterprise white-label play, which has a longer but more reliable sales cycle.

Related Terms

Three adjacent trends feed this one. MCP (Model Context Protocol) is the transport layer skills ride on — its standardization is the precondition for a cross-platform marketplace. AI Agent Frameworks (LangChain, CrewAI, AutoGen) are the runtimes that consume skills, and their fragmentation is precisely why a neutral registry wins. Prompt Marketplaces (PromptBase and similar) are the failed first attempt at monetizing AI expertise — they proved demand exists but lacked executable packaging, which is exactly the gap Agent Skills Marketplaces close. Watch all three: when MCP adoption and framework consolidation peak, this category goes mainstream.