Agent Skills Marketplace

What is it

An Agent Skills Marketplace is a storefront where AI agents download capabilities they don't natively have. Think of it as an app store, but the "apps" are discrete, composable skills — a shell command executor, a PDF parser, a Stripe refund handler, a web scraper — packaged so that any compatible agent framework (Claude's tool use, OpenAI's function calling, LangChain, or a shell-based runner) can install and invoke them at runtime.

The technical essence: a skill is a versioned bundle containing a manifest (name, description, input/output schema), the execution code, permission scopes, and metadata for discovery. The marketplace is the registry, the search layer, and increasingly the billing layer.

The business significance is bigger than it looks. If agents become the dominant software consumer, whoever owns the registry where agents discover and install capabilities sits in the same position the App Store occupied in 2009. Right now the space is fragmented — Skillstore, WorkBuddy's skill market, and open-source repos like mattpocock/skills — which means the registry layer is unowned. That's the opening. You're not selling a skill; you're selling the distribution channel for everyone else's skills.

Why now

Three things converged in the last 12 months to make this viable, and none of them existed a year earlier.

First, agent frameworks standardized enough to make portable skills possible. Claude's tool-use schema, OpenAI's function calling, and the Model Context Protocol (MCP) gave developers a common shape for "here is a capability, here is its schema." Before that, every skill was bespoke glue code.

Second, agents got capable enough to actually need a marketplace. When an agent could only summarize text, it needed one tool. Now that agents execute multi-step workflows — browse, extract, write to a database, send an email — the number of capabilities per task exploded, and hand-rolling each one stopped scaling.

Third, distribution channels opened up. The Show HN and OSChina signals in the last week show developers are actively publishing and comparing skill collections. mattpocock/skills going public is a tell: respected engineers are treating skills as shareable artifacts, not internal config.

The window is now because the registry layer is still unclaimed. In 12 months, either a framework vendor (Anthropic, OpenAI) or a platform (Vercel, Cloudflare) will ship a first-party registry. Today, nobody has.

Market Evidence

The signal is thin but directional: 3 independent sources, 3 total mentions, 100% growth rate, stage classified as nascent. That's not a market yet — it's a leading indicator.

Break it down honestly. Three sources across Show HN, OSChina, and GitHub means the interest is real but geographically and culturally split (Western hacker news + Chinese dev community + open-source). That cross-pollination is actually a stronger signal than three mentions in one place — it means the concept is independently converging in different ecosystems, which usually precedes a real category.

The 100% growth rate is a vanity metric at this volume. Three mentions becoming six is 100% growth and means nothing. Don't over-read it.

What matters more: the mentions are product mentions, not think-pieces. Skillstore and WorkBuddy's skill market are live products people are trying. mattpocock/skills is a public repo people are cloning. That's builders building, which is the only signal that matters at the nascent stage.

Verdict: real demand, unproven market. This is the phase where you validate cheaply, not where you raise money. The absence of a dominant player is the opportunity; the absence of volume is the risk.

Who's Behind It

The current cast is small and fragmented, which is exactly why it's enterable.

Skillstore is the most product-shaped entrant — a marketplace framing for agent skills. Weakness: unclear distribution and no visible network effect yet. WorkBuddy's skill market bundles skills into a broader agent product, meaning the marketplace is a feature, not the business — they'll never optimize for third-party skill vendors. mattpocock/skills is an open-source collection from a well-known TypeScript educator; it signals that credible developers will publish skills publicly, but it's a repo, not a registry.

The whales are absent but circling. Anthropic owns the tool-use schema and MCP; OpenAI owns function calling; LangChain owns the framework mindshare. Any of them could ship a first-party registry overnight. The question isn't whether they will, but whether they'll do it well — framework vendors historically build mediocre marketplaces because marketplace ops (review, payments, discovery, trust) aren't their core competency.

That's the gap: the whales will build registries, but they won't build good ones. A focused independent operator can win the long tail before the whales care.

TAM & Market Size

Be realistic: the current addressable market is small, and the provided scores (Opportunity 0/100, Demand 0/100) reflect that honestly. This is a bet on a future market, not a present one.

Who buys? Three segments. (1) Agent developers — indie hackers and small teams building agents who need capabilities fast and would rather pay $10–50/month than write and maintain 20 integrations. (2) Skill authors — developers who build skills and want distribution and revenue share; they're supply, not demand, but they're the flywheel. (3) Enterprises piloting agents — teams that need vetted, permissioned, auditable skills and will pay for compliance metadata.

Price tolerance is low today. Indie developers pay $19–49/month for dev tools (see the entire Vercel/Railway/Supabase tier). Skill authors will accept 20–30% revenue share — the App Store's 30% is the anchor. Enterprises are the only segment with real budget, but they won't buy a nascent marketplace; they'll buy from a vendor with SOC 2 and a support contract.

The honest TAM framing: today maybe 50k–200k developers globally are building agents seriously. At $30/month average, that's a $18M–72M annual market ceiling if you capture all of it — which you won't. The upside case is that agents become the primary software consumer and the registry becomes infrastructure. That's a $1B+ outcome, but it's a 3–5 year bet.

Competitive Landscape

The field is wide open and the competition score (0/100) reflects that no one has claimed the position.

Framework-native registries (Anthropic, OpenAI, LangChain) are the biggest threat but the weakest current products. They have distribution and schema control but no marketplace DNA. They'll ship something functional and unloved.

Skillstore is the closest direct competitor — marketplace framing, early. Its weakness is the classic marketplace cold-start: no supply means no demand means no supply. It hasn't cracked the flywheel.

WorkBuddy's skill market is a feature inside an agent product. It will never serve third-party vendors well because its incentive is to lock skills to its own agent.

Open-source collections (mattpocock/skills and similar) are the real supply today — free, GitHub-hosted, no billing. They prove demand for skills but undercut any paid marketplace. Your job is to be better than GitHub for this use case: search, versioning, permissions, one-command install, and billing.

The differentiation gap: trust and distribution. Nobody owns "the npm of agent skills." The first operator to nail one-command install, semantic search, and a clear security model (what can this skill touch?) wins the long tail before the whales arrive. You have roughly 12–18 months before a first-party registry becomes credible.

Business Model

Take a clear position: marketplace with revenue share, plus a paid pro tier for authors and consumers. Freemium alone won't work — you need the supply side incentivized.

Structure:

• Free tier: browse, install public skills, 3 private skills. Drives adoption and SEO.
• Pro ($29/month): unlimited private skills, team sharing, usage analytics, priority search placement. Targets indie devs and small teams.
• Marketplace take rate (20%): on paid skills. Lower than Apple's 30% to attract authors; high enough to fund ops.
• Enterprise (custom, $500+/month): SSO, audit logs, permission policies, SOC 2 roadmap. Targets agent-piloting companies.

Why this fits: the free tier builds the registry's network effect (more skills = more value = more users), the pro tier monetizes the developers who need privacy and teams, and the take rate captures value from the paid-skill long tail without gating discovery.

12-month forecast:

• Conservative: 800 free users, 60 pro ($1,740 MRR), $500 marketplace GMV → ~$1.8k MRR.
• Base: 3,000 free, 250 pro ($7,250 MRR), $4k GMV at 20% ($800) → ~$8k MRR.
• Optimistic: 10,000 free, 900 pro ($26,100 MRR), $20k GMV ($4k) → ~$30k MRR.

CAC: near-zero via content and GitHub distribution (skills are inherently shareable). Paid CAC ~$40–80 if you run dev-tool ads. Payback: under 2 months on pro at $29/month. The model works because distribution is organic — that's the whole reason to build in this category.

MVP Blueprint

Ship in 5–7 days. Cut everything that isn't registry + install + search.

Core features (only these):

1. Skill manifest spec — a simple skill.json (name, version, description, input/output schema, permissions, entrypoint). Mirror MCP's shape so it's familiar.
2. Registry + search — Postgres full-text search over name/description/tags. No vector search yet.
3. One-command install — npx skills install <name> that drops the bundle into a local .skills/ directory and prints the manifest. This is the whole product.
4. Skill detail page — manifest, README, install count, author, permission warnings.
5. Auth + publishing — GitHub OAuth, npx skills publish from a repo.
6. Basic billing — Stripe for pro tier only. Skip paid skills in v1.

Stack: Next.js + Postgres (Neon) + Vercel. CLI in Node/TypeScript published to npm. Auth via GitHub OAuth. Total infra cost under $50/month.

Fastest path to launch: build the CLI and registry first, seed it with 10–20 skills you write yourself (shell executor, web fetch, PDF parse, JSON transform, etc.), then post to Show HN with the angle "npm for agent skills." The CLI is the moat — if npx skills install becomes muscle memory, you win.

Skip in v1: vector search, ratings, paid skills, enterprise features, team workspaces, analytics dashboards. All of it. Ship the install command.

Commercial Opportunities

1. The registry itself (primary bet). A focused, well-designed skill registry with a killer CLI and a security model. Target: indie agent developers and small teams. Expected revenue: $3k–15k MRR within 12 months on pro subscriptions plus take rate. Why it beats alternatives: it's the only position with a network-effect moat — every skill added makes the registry more valuable, and no framework vendor will do the ops work well.

2. Skill auditing and security service. Agents executing third-party code is a real risk. Offer automated permission scanning, sandbox verification, and a "verified" badge for skills. Target: teams that need to justify installing third-party skills (enterprise-adjacent). Expected: $2k–8k MRR as an add-on to the registry or standalone consulting. Why it beats alternatives: trust is the #1 blocker to marketplace adoption, and nobody's solving it.

3. White-label skill registry for agent platforms. Sell the registry infrastructure to agent-product companies who need an in-app skill store but don't want to build it. Target: WorkBuddy-style agent products, vertical AI SaaS. Expected: $5k–20k per contract, recurring. Why it beats alternatives: it monetizes the long tail of platforms that will never build their own marketplace, and it's a natural B2B upsell from the consumer registry.

Product Ideas

🥇 Skills Registry + CLI ("npm for agent skills"). Value prop: One command to install any agent capability, with a security manifest so you know what it touches. Target user: Indie agent developers, small AI teams. Why now: The registry layer is unclaimed, MCP standardized the shape, and the CLI creates a habit moat before framework vendors ship first-party stores.

🥈 Skill Audit & Verified Badge. Value prop: Know exactly what permissions a skill requests before you install it — automated scanning plus a verified badge for trusted authors. Target user: Teams installing third-party skills; agent-platform operators. Why now: Security is the top objection to agent marketplaces, and no one has built the trust layer. It's a natural paid add-on with high willingness to pay.

🥉 White-Label Skill Store for Agent Platforms. Value prop: Drop a full skill marketplace — discovery, install, billing — into your agent product in a week. Target user: Agent-product companies (WorkBuddy competitors), vertical AI SaaS. Why now: Every agent product wants a skill store; none want to build the ops. You sell the same infrastructure repeatedly at B2B prices.

SEO Opportunity

Search volume is emerging, not established — terms like "agent skills marketplace" and "install agent skills" are low-volume today but the trend line is up. SEO difficulty is effectively zero (0/100), which means ranking is a matter of publishing first and consistently.

Long-tail targets: "how to install agent skills," "claude agent skills marketplace," "mcp skill registry," "agent skill manifest format," "best agent skills for developers."

Content strategy: every skill detail page is an SEO asset (name + description + install instructions). Publish a canonical "Agent Skills Manifest Spec" doc and become the reference — reference docs rank and earn backlinks. Pair with comparison posts ("Skillstore vs. [your registry]") to capture competitor traffic.

Risk Assessment

The thesis breaks if framework vendors ship a good first-party registry. If Anthropic or OpenAI launches a polished, well-ops'd skill store with native install, your distribution advantage evaporates. Mitigation: move fast, own the CLI habit, and stay framework-agnostic (support Claude, OpenAI, and LangChain from day one).

Market risk: agents don't become the primary software consumer. If agents stay a niche workflow tool, the registry never reaches escape velocity. Watch for whether agent frameworks retain developers past the demo phase.

Execution risk: the cold-start problem. A marketplace with no skills has no users and vice versa. Mitigation: seed 20+ skills yourself, make publishing frictionless, and don't gate supply behind review in v1.

Cheap validation: build the CLI and a bare registry in a weekend, seed 10 skills, post to Show HN. If you get 100+ installs and 3+ third-party publishes in two weeks, the flywheel is turning. If not, walk away — you've spent a weekend, not a quarter.

Action Plan

Today: Write the skill.json manifest spec (steal MCP's shape) and build a 10-line npx skills install prototype that pulls a skill from a GitHub repo. That's the entire thesis in one command.

This week: Ship a bare registry (Next.js + Postgres) with search, detail pages, and the install CLI. Seed it with 10–15 skills you write yourself: shell executor, web fetch, PDF parse, JSON transform, Stripe refund, email send. Publish to npm. Post to Show HN with "npm for agent skills — install any capability in one command."

Month 1: Measure install counts and third-party publishes. Add GitHub OAuth and npx skills publish. Launch the pro tier ($29/month) with private skills. Target 500 installs and 5 external authors.

Month 3: Add the verified-badge audit layer (manual at first), start content SEO with the manifest spec doc, and reach out to 3 agent-platform companies about white-label. Target 200 pro subscribers and $6k MRR. If installs are flat after month 1, kill it and move on — the signal will be clear fast.

Related Terms

Model Context Protocol (MCP) — Anthropic's standard for connecting agents to tools and data. It's the schema layer beneath skill marketplaces; the marketplace is the distribution layer on top. Watch MCP adoption as a leading indicator.

AI Agent Frameworks (LangChain, CrewAI, AutoGen) — the runtime that consumes skills. Framework-native registries are your biggest competitive threat.

Agent Observability — tooling to trace what agents actually do. Adjacent opportunity: skills that ship with built-in telemetry become more trustworthy, feeding the audit/verified-badge play.