Claude Code skill

An installable agent skill that teaches Claude Code, Cursor, Codex, and other coding agents how to build against the MPCKit SDKs end-to-end.

The MPCKit repo ships an open agent skill at skills/mpckit. Install it into your coding agent and it will read the SDK surface, the onboard-sign flow, and the error taxonomy directly into the agent's context the moment you ask it to integrate MPCKit.

Skills are how you give an agent durable, procedural knowledge about a library without re-pasting docs into every prompt. The skill loader (skills.sh) is open source at vercel-labs/skills and supports Claude Code, Cursor, Codex, OpenCode, Cline, and 50 other agents.

Install

In any project where you use a coding agent:

npx skills add Iamknownasfesal/mpckit --skill mpckit

The loader detects which agents you have configured (Claude Code, Cursor, etc.) and installs the skill for each. To target one explicitly:

npx skills add Iamknownasfesal/mpckit --skill mpckit -a claude-code

To install once for your user account, available in every project:

npx skills add Iamknownasfesal/mpckit --skill mpckit -g

You can also point the loader directly at the skill's tree URL, which is useful in CI where you do not want to depend on the loader's project resolution:

npx skills add https://github.com/Iamknownasfesal/mpckit/tree/main/skills/mpckit

What the agent learns

The skill is a hub (SKILL.md) plus five reference files. The hub is always in context once the skill is matched; reference files are read on demand only when the task touches them.

TypeScript reference

@mpckit/sdk full surface: MPCKit class, onboard, sign, billing, introspection, crypto engines, the raw HTTP escape hatch.

React reference

MPCKitProvider plus all 10 hooks, worker setup by bundler, SSR notes, and the typed query key factory.

Rust reference

mpckit crate builder, async surface, default vs the `crypto` feature, two-phase sign.

Flows

Cold-start-to-signed-message recipes, the passkey-PRF seed pattern, idempotent retries, broadcasting per chain.

Errors

MPCKitError taxonomy, error codes, retry policy, production logging.

When the agent invokes it

The skill's frontmatter description lists the triggers. In short, any task that mentions:

@mpckit/sdk, @mpckit/react, or the mpckit Rust crate.
The MPCKit class, MPCKit::builder, MPCKitProvider, or any of the hooks (useMPCKit, useOnboard, useSign, useDWallets, ...).
dWallet onboarding or signing.
The api.mpckit.xyz or api.testnet.mpckit.xyz endpoints.

When matched, the agent has the SDK surface, the network hosts, the API-key prefix conventions, the valid (curve, signature algorithm, hash scheme) combinations, and the common-mistake table at hand without any extra prompting from you.

Compatibility

Agent	Status
Claude Code	tested in CI
Cursor	tested in CI
Codex	tested in CI
OpenCode, Cline, Gemini, Copilot, Windsurf, +50 more	supported via `skills.sh` loader

The skill is plain Markdown with YAML frontmatter; nothing about it is agent-specific.

Authoring your own

If you fork MPCKit or run a self-hosted backend with a divergent surface, you can fork the skill in place:

Edit skills/mpckit/SKILL.md and the matching references/*.md.
Bump metadata.json:version to match your SDK minor.
Install from your fork: npx skills add yourorg/yourfork --skill mpckit.

The skill source is the authoritative document for the agent; keep it in lockstep with the SDK packages under packages/sdk-ts, packages/sdk-react, and packages/sdk-rust.

Why not just point the agent at this docs site?

Two reasons.

Triggers and budget. The skill loader puts the hub into context exactly when the agent's task matches the description, and only loads reference files on demand. Pasting the entire docs site into every prompt is wasteful and noisy.
Procedural knowledge. This page tells you what MPCKit is. The skill tells the agent how to use MPCKit, with conventions, common mistakes, the install commands, and a build-verification checklist that are aimed at code generation, not at humans reading prose.

The two are complementary. Humans read the docs; agents load the skill.