What SkillVault is

SkillVault is a hosted registry for AI agent skills. You register a skill once — a name, a description, a body of instructions — and SkillVault exposes it through three surfaces simultaneously: an authenticated URL your agents can fetch, a JSON REST API your apps can call, and a CLI wrapper that drops a single-skill fetch into local AI harnesses like Claude Code, Cursor, Windsurf, Aider, and Codex. One source of truth. Three delivery surfaces. Same library everywhere.

The "vault" metaphor is deliberate. A skill vault is something you own and control — not a marketplace, not a model router, not an analytics dashboard. Your skills live in your vault, served from a domain you authenticate against, behind a token you can rotate without touching your password. The URL you give to an AI stays the same forever; the token is the lock that determines who can read what.

The problem we are solving

AI assistants do not share state. The custom instructions you wrote for Claude on the web are not visible to Cursor on your laptop. Neither of them has access to the prompt library you carefully built up in ChatGPT. Skills you write are local to one surface — and when you move to a new device, a new chat, a new agent harness, you start over.

This fragmentation is a tax. Every prompt you re-author is a tax. Every inconsistency between how you instruct Claude and how you instruct GPT is a tax. Every onboarding doc you write for a new tool is a tax. SkillVault exists to retire that tax.

How the vault works

When you create a vault, SkillVault gives you three things:

A vault URL. https://app.skillvault.dev/v/<your-username>/<skill-name> returns the body of one skill — a Markdown document your AI reads as instructions. The URL is stable. The token is bound to the vault, not to any single skill, and you can rotate it from the dashboard at any time. AI assistants that fetch the web (Claude.ai, ChatGPT with browsing, Perplexity, Gemini, mobile chat apps) just need the URL; the token is appended or sent as a header depending on the surface.

A JSON REST API. GET /api/v1/skills lists your skills. GET /api/v1/skills/<name> returns one. The endpoints are documented in the OpenAPI spec at /openapi and can be called from any HTTP client. AI apps with a connector or action config — ChatGPT Custom GPTs, Claude.ai custom integrations, Gemini extensions — point at the JSON endpoints and read skills as structured data.

A CLI wrapper. npx @skillvault/cli bootstrap installs the wrapper, asks for your vault URL, and writes a single-skill fetch into whichever AI harness folder it detects. Claude Code gets a skill.md. Cursor gets a .cursor/rules/skillvault.mdc. Windsurf, Aider, OpenCode, Codex, and Cline each get the equivalent. From that moment, the harness treats your vault as one of its own skills.

The three surfaces share the same backend. Update a skill once, push to the vault, and the next fetch from any surface returns the new version.

What lives in a skill

A skill has four fields:

Name. Lowercase, dash-separated, unique to your vault. Used as the URL slug, the CLI identifier, and the OpenAPI path.
Description. One sentence. The first thing an AI reads when it discovers the skill; also shown in your dashboard list.
Body. The actual instructions. Markdown, no length limit, no required structure. Write the prompt you wish you'd had three months ago.
Tags. Optional. Useful for grouping skills inside your own vault.

There is no schema for what a skill body looks like. Some people write them as system prompts. Some write them as few-shot examples. Some write them as playbooks — "if the user asks X, do Y; if they ask Z, do W." Whatever shape matches your workflow.

How skills reach the agents

The authenticated vault URL is the canonical surface. AI assistants that can fetch the web — Claude.ai with the web-browsing tool, ChatGPT with Browse enabled, Gemini, Perplexity, and the mobile apps for all of them — accept the URL and read the skill body as context. You give the agent the URL once; it remembers across sessions (depending on the surface) and reads the latest version on each fetch.

For agents that can't fetch the web — Claude Code on the desktop, Cursor, Windsurf, Aider, Codex — the CLI wrapper drops a single file into the harness's skills folder. The file is generated to point at your vault URL with your token embedded; the next time the harness reads its skills, your library shows up alongside the built-ins. One file. Same library as the web surface.

For programmatic access — Custom GPTs, Claude.ai custom integrations, internal tools, third-party apps you don't control — the JSON REST API is the right surface. Authenticate with a vault token in the Authorization: Bearer header, list skills, fetch bodies, parse the JSON, and inject the body into whatever prompt your app constructs.

What we don't do

We don't run models. SkillVault doesn't answer your questions. It doesn't choose between GPT and Claude. It doesn't decide when to invoke your skill — the model in front of you does that. We store your skills, surface them to whichever model you choose, and stay out of the way.

We don't track which agents read which skills. There's no analytics dashboard on our side. There's no model-side attribution. The vault log (when enabled in a paid tier) shows what you read, not what your agents read. If you want telemetry, build it on your side by reading the JSON API in your own application.

We don't host a marketplace. You can't publish skills to other people's vaults. You can't subscribe to someone else's library. The vault is yours.

Pricing

SkillVault has three live tiers and one Enterprise option:

Free — 3 skills, 100 MB. For trying the product end-to-end with a small, real vault. No card required.
Pro — €6.95/month. 50 skills, 1 GB. For people who actually use this every day across more than one agent.
Power — €18.95/month. Unlimited skills, 10 GB, priority routing. For people whose vault has become part of how they work.
Enterprise — Shared team vaults, SSO, signed DPAs, dedicated support. Email enterprise@skillvault.dev.

All three tiers include all three surfaces (URL, JSON, CLI). The tier is the cap on what you can put in the vault; the surfaces are not paywalled.

Why the token is separate from your password

Your password proves who you are to SkillVault. Your vault token proves which vault a fetch is for. They are deliberately separate so you can:

Rotate the vault token without changing your password. Suspected leakage of the URL? Rotate the token; the URL stays the same; every agent that had the old URL stops working and gets the new one when you re-issue.
Give different tokens to different surfaces. The CLI installs the token in a local config file; the JSON API bearer is sent in HTTP headers; the URL surface encodes the token in the path. Rotating one surface's token doesn't disturb the others.
Share the vault without sharing the account. Power tier supports delegated tokens per device. Revoke one, the others keep working.

The vault token is stored hashed at rest on SkillVault's side. The plaintext is shown to you exactly once at issuance, and never again. Rotation produces a new plaintext, also shown once.

Security and privacy

SkillVault stores your skill bodies in S3-compatible object storage with encryption at rest. Vault metadata (names, descriptions, tags, ownership) lives in a managed Postgres database in the EU. In transit, every byte between your browser, your CLI, your AI agents, and SkillVault is TLS. The vault URL is https:// only — there is no plaintext HTTP surface.

We don't sell, share, or transfer your skill content. The vault content is yours; we hold it for you. The full privacy policy is at /privacy-policy.

How to start

The fastest path is curl -fsSL https://app.skillvault.dev/install.sh | bash. The CLI installs, asks for your vault URL, and writes the wrapper file. Total time: under a minute.

If you'd rather see the dashboard first, create an account at /auth/signup. Free tier, no card. From there you can register your first skill and copy the vault URL.

Get in touch

Email hello@skillvault.dev for general questions. Enterprise customers: enterprise@skillvault.dev. Security disclosures: security@skillvault.dev.