Agent skill infrastructure

Universal skills agents can discover, trust, and run.

SkillHub is the registry, marketplace, and gateway layer for reusable AI-agent capabilities: one manifest format, searchable packages, permission profiles, billing-ready usage, and agent-ready APIs.

Publish a skill View contract Open consoles

Operating evidence chain

One visible path from publisher supply to agent use.

SkillHub is not a directory: one skill can move from contract, to review, to install, to governed runtime, to ledger, payout, and audit.

01Publish contractPublisherDraft -> submittedManifest, version, permissions, runtime, examples, and pricing intent are saved as an organization-scoped draft.02Review gateTrust operatorQueued -> decidedAutomated manifest, runtime, example, and security checks feed a human review decision with notes and SLA.03Marketplace installDeveloperInspect -> installApproved supply becomes searchable listing data with install readiness, publisher trust, pricing, feedback, and alternatives.04Runtime governanceAgent builderInvoke -> logProject keys, policy approval, budget, rate limit, subscription state, REST, and MCP calls share the same gateway path.05Ledger splitFinance adminUsage -> balanceBillable usage and subscription periods post immutable transactions, commission splits, balance rows, refunds, and disputes.06Payout and auditPublisher + adminAvailable -> paidEligible balances move through reserve, finance review, manual PayPal/Alipay transfer, notification delivery, and audit logs.

Published skills4

Verified1

API calls0

Avg latency--

Live rows stay live: empty production data is shown as empty, not replaced by demo supply.

Console access

Where the backend lives.

SkillHub's backend is split by job: account access, developer runtime operations, publisher supply operations, and platform governance.

Current sessionWorkspace connected

No shared passwordProduction access uses email code, configured Google/GitHub OAuth, or a reveal-once team token. Raw tokens and passwords are never published as page copy.

OAuth readiness is visibleIf Google or GitHub is not configured, the login page shows callback URLs and missing environment names instead of fake buttons.

Token fallback is limitedUser tokens remain for bootstrap, invitations, and recovery. Product users should enter through login and account center.

Workspace connected

Login and account centerAll users/login -> /account

Open account

Available for this session

Developer workspaceDeveloper / owner/developer

Create projects, manage installed skills, runtime keys, policies, tests, buyer requests, billing, team, and webhooks.

Open console

Available for this session

Publisher workspacePublisher / owner/publisher

Upload skills, manage review repair, pricing blockers, buyer demand, feedback, revenue, payout readiness, and notifications.

Open console

Available for this session

Admin operationsReviewer / finance / admin/admin

Operate review queues, launch readiness, identity directory, trust reports, incidents, ledger, payouts, and delivery outbox.

Open console

P0 journey map

The three roles share one operating chain: publish, review, install, run, ledger, payout, and audit.

Developer install journey

The developer path proves SkillHub is more than a directory: discovery becomes project install state, runtime governance, logs, cost, and updates.

CurrentMarketplaceNextSkill detail

MarketplaceSearch, filter, and compare trust, runtime, pricing, and publisher signals.
Skill detailReview manifest, permissions, examples, feedback, incidents, and alternatives.
Developer workspaceCreate projects, manage team, billing, webhooks, and buyer requests.
Project commandPin versions, approve policy, create keys, and manage updates.
Runtime testUse REST or MCP through the same policy, budget, subscription, and log path.

Inspect a skill

Publisher monetization journey

The publisher path proves authors have repeat work: draft packaging, exact-version review, repair, pricing blockers, feedback, revenue, and payout readiness.

CurrentPublish entryNextPublisher workspace

Publish entryPaste a manifest, run preflight, save an organization-scoped draft.
Publisher workspaceOperate owned skills, versions, buyer demand, notifications, and revenue.
Review repairUse reviewer notes, SLA, automated evidence, target fields, and next actions.
Pricing and payoutComplete profile, terms, payout readiness, verified review, and pricing state.
Improve and retainRespond to feedback, buyer requests, revenue signals, refunds, and placement appeals.

Open workspace

Admin governance journey

The operator path proves SkillHub is governable: readiness, review, trust, money movement, delivery, and audit stay in one command center.

CurrentAdmin consoleNextLaunch readiness

Admin consoleStart from the operating console.
Launch readinessInspect launch blockers, migrations, templates, and credibility thresholds.
Reviews and trustUse SLA, risk, and automated evidence to approve, reject, or block.
Finance and payoutsOperate ledger, refunds, disputes, commissions, and payout decisions.
Delivery and auditRecover email/webhook delivery and verify every privileged action.

Check readiness

Live marketplace loops

One platform for developers, publishers, and operators.

The product only feels real when a listing becomes install state, runtime state, review state, ledger state, and payout state. SkillHub keeps those loops visible from day one.

01Publish

02Review

03Install

04Invoke

05Ledger

06Payout

Developer / Agent BuilderProject runtime

Find a verified skill, install it into a project, create a reveal-once runtime key, run a governed test call, and monitor cost or incidents later.

Install -> test -> monitorOpen developer console

Publisher / Skill AuthorSupply operations

Upload a manifest, submit an exact version, repair review blockers, activate pricing gates, respond to feedback, and track payout readiness.

Draft -> review -> improveOpen publisher workspace

Platform OperatorTrust and finance

Prioritize reviews, govern abuse and incidents, process ledger states, manage notification delivery, and inspect launch readiness without exposing secrets.

Review -> govern -> launchOpen admin operations

Registry

Agent Skill Registry

New skill

Published skills4

Verified1

API calls0

Avg latency--

Verified share25%

SkillTagsTrustRisk

Browser Research

v0.1.0

Research a web topic and return concise findings with source URLs.

browser-research

VerifiedManifest checked

Medium risk

Dataset Summarizer

v0.1.0

Convert tabular data into structured notes, anomalies, and next actions.

dataset-summarizer

SubmittedManifest checked

Medium risk

Operating proof

A customer can inspect the platform, not just the pitch.

These are not brochure claims. They are the operating surfaces a buyer, publisher, or platform operator can revisit after the first install.

Open launch readiness

Account entry

Email-code login, OAuth readiness, session fingerprints, and connected identity state are modeled before launch.

Review gate

Submitted versions run automated manifest, runtime, example, and security checks before approval.

Runtime governance

REST and MCP calls pass through project install, policy, budget, subscription, logging, and metering controls.

Commercial ledger

Usage and subscriptions post into transactions, splits, balances, invoice records, refunds, disputes, and payouts.

Notification loop

In-app events fan out to email and organization webhooks with templates, retry state, and admin recovery.

Smoke checks

Production smoke commands verify stats, auth providers, launch readiness protection, and key app pages.

Protocol

A skill contract agents can read before they act.

A SkillHub package is not just a prompt. It is a typed, versioned capability with declared runtime, schemas, permissions, pricing, and review state.

Discover

Agents search by task, tags, permission profile, runtime contract, and pricing model.

Validate

Skill manifests declare inputs, outputs, runtime entrypoints, access needs, and version history.

Execute

The gateway exposes HTTP and MCP endpoints so agents can call skills safely and meter usage.

skillhub.json

One manifest for humans, agents, and runtime gateways.

The registry accepts a compact JSON manifest. The same contract powers search results, trust review, SDK generation, billing, and MCP tool discovery.

Typed input and output schemas

HTTP, MCP, or local runtime declarations

Explicit network, browser, filesystem, and secret access

Versioned identity for repeatable agent behavior

skillhub.jsonschema v0.1

{
  "schemaVersion": "0.1",
  "name": "browser-research",
  "runtime": {
    "type": "http",
    "entrypoint": "https://api.useskillhub.com/demo/browser-research"
  },
  "permissions": {
    "network": true,
    "browser": true,
    "filesystem": "none",
    "secrets": []
  }
}

Trust layer

Designed for agents that need guardrails.

Permission aware

Every package carries its network, browser, filesystem, and secret requirements.

Versioned packages

Skills are registered with immutable versions so agents can pin behavior.

Operator control

Paid publishing, review, refunds, and payouts stay behind audited platform controls.