One skill library for your AI coding tools.
Claude Code, Codex, Grok and AGY read the same signed skills from one install on each machine. Every skill version is signed offline, and your machine checks the signature before your agent reads it.
Free tier, no cardChecked on every readTwo-step sign-in for every account
A skill is only as safe as the path it took to you.
Most skill marketplaces are a list of links. SkilVault signs every version offline and makes your machine prove it — so even a hacked server cannot change what your agent reads.
Read and run
Every skill is read, run and reviewed before it can be published.
Signed offline
SkilVault signs each version with a key that never touches the server.
Published
The server serves exactly what was signed, and takes no writes from the internet.
Checked on your machine
The client verifies the signature and the pack's checksum before anything is read.
No rollbacks
Once you have used a newer version, an older one is refused.
If a skill is altered, your client refuses it.
Verification is silent when everything checks out and loud when it does not. A skill that is not signed by a trusted key is never handed to your agent.
$ skilvault read changelog-entry-on-completion
skilvault: refusing 'changelog-entry-on-completion' v1.0.1: it is not signed by a trusted keyConnect once. Your agent does the rest.
No cloning, no vendoring, no hunting through lists of unreviewed links.
Connect once
Open Install Client in your dashboard, install on your machine and approve its code in your browser. Terminal setup is available for each supported platform.
skilvault listFind by task
Claude Code and Codex also get a prompt hook that names the skills fitting each request. Grok and AGY have no usable prompt hook, so there the agent runs skilvault use to find one. Either way the whole catalog is matched on your machine, and the request is not sent.
skilvault use "review a pull request"Read on demand
Skills are fetched when used and checked against SkilVault's offline signature. Nothing is loaded into your agent's context until it is needed.
skilvault read commit-lintSee what a skill does before you install it.
brave-news-search
Search and filter news articles through the Brave Search news API: freshness or custom date ranges, country and language, SafeSearch, domain boost/downrank/exclude via Goggles, pagination and extra excerpts. Use when the user wants current or breaking news on a topic, articles from a specific date range, news for a country or language, to promote or suppress specific outlets, or article metadata such as source, age, URL or thumbnail — even when they never mention an API.
css-scroll-containers
Decides whether a layout needs an inner scroll container or should be restructured for page-level scroll, writes one-axis overflow CSS, and defines auto-scroll, scroll-restore and error-scroll behavior for chat/log panels. Use when the user mentions scroll, overflow, scrollable, nested or inner scroll in any CSS or component task, or when building tables, panels, drawers, sidebars, modals, chat logs or any fixed-height or long-list layout that might overflow — even if they only ask whether a region should scroll independently, or want existing overflow handling in CSS/TSX/JSX reviewed.
dense-ops-ui-design
Design guidance for dense, workflow-driven UIs that trained specialists — dispatchers, warehouse staff, planners — use as their all-day work tool. Covers table density, in-place accordion hierarchies, status colours, filter separation, keyboard flows, bulk-feedback tiers, persistent workspace state, and a review checklist. Use when designing a dispatch, warehouse, logistics or scheduling tool, building a B2B operations interface for expert or power users, laying out dense data tables or planning UIs, or structuring filters, status indicators or keyboard flows — even if the user just calls it an internal tool or admin screen.
graduate-learnings
Promotes a stable lesson from .groove/memory/learned/<topic>.md into AGENTS.md so every future session sees it: appends a one-sentence bullet with provenance to a comment-delimited Graduated Learnings block (creating AGENTS.md if absent), then marks the source entry graduated without deleting it. Use when the user says a lesson is stable, proven, or should always apply going forward, or asks to graduate, promote, or persist a learned insight, even if they do not mention AGENTS.md. Confirms exact text before writing and skips exact duplicates.
intent-interview
Turns a vague request into a user-confirmed statement of intent through a one-question-at-a-time interview where every question carries your own best guess, ending in goal, in/out scope, constraints, and completion criteria. Use when an ask like 'build me X' is missing who it's for, why now, success criteria, or binding constraints, when the user says 'interview me', 'grill me', 'are we sure?', or 'stress-test my thinking', or when you catch yourself silently filling in ambiguous requirements before any plan or code exists — even if they never ask for an interview.
pay-for-http-request
Make HTTP requests with automatic x402 payment support using the purl command line interface.
More than a list. A library that works with you.
Skill chains
Link skills into an ordered chain your agent reads step by step. A step your plan cannot read is named, and the chain carries on. Available on plans that include chains.
Loaded only when needed
Each CLI gets its own folder of lightweight entries, so your agent's context stays small until a skill is actually read.
Learns on your machine
Rules you confirm are kept as notes on your machine and shown after the skill. Sharing back is off unless you turn it on.
Suggestions that fit the project
Open a repo and SkilVault names the skills that match its stack — once per session, never on every edit.
Two-step sign-in, everywhere
Every account uses two-step verification, with passkeys supported, and keys travel in a header rather than the address.
Start free. No card required.
Start free with five active skills for automatic loading and manual access to the whole library.