Skip to content
Sign in
Benchy Docs
Browse the docs

Agent skills

Open, agent-consumable playbooks for Benchy. Give one to Claude Code, Codex, Cursor, or Grok and your agent can run your Benchy account.

What skills are

A skill is a single Markdown document that teaches an MCP-capable agent — Claude Code, Codex, Cursor, Grok, or your own — how to operate a slice of Benchy. Give your agent a skill and an authorized Benchy connection, and it can build your bio, manage your shop, and upload and check models through the hosted MCP server.

Skills encode the things agents otherwise learn the hard way — which tool to call first, which caps are enforced server-side, when to dry-run, what each error code means, and the boundaries that must never be routed around.

Install a skill

Each skill is one SKILL.md instruction file. Installing it does not download Benchy, run a Benchy server, or give the agent access by itself. Put the file wherever your agent reads instructions:

  • Claude Code — save it as .claude/skills/<name>/SKILL.md in the agent's workspace (or ~/.claude/skills/ globally); the frontmatter description tells Claude when to load it.
  • Cursor / Codex / anything with rules or instruction files — drop the file in your rules directory or paste it into the system context.
  • Hosted agents — paste the document into the conversation before asking for Benchy work.
shell
mkdir -p .claude/skills/getting-started-with-benchy
curl -o .claude/skills/getting-started-with-benchy/SKILL.md \
  https://benchy.world/skills/getting-started-with-benchy.md

Pair the instruction file with the hosted MCP connection — Connect your agent covers OAuth, API keys, and each client's exact setup.

The skills

Getting Started with BenchyThe foundation skill for operating a user's Benchy account through the hosted MCP server. It covers safe connection, permissions, approval, and the core Bio, Shop, World, Canvas, Writer, Box, and account-knowledge workflows.Manage Your Benchy ShopThe seller-operations skill: everything commerce on benchy.shop, from a fresh Stripe Connect account to a running catalog with orders, entitlements, gated downloads, subscriptions, and refunds — including the invariants an agent must never route around.Upload and Check 3D ModelsThe pipeline skill: upload, checks, and publishing are separate states. Covers ingest grants, multipart transfer, server-enforced benchy-mesh policies, print-readiness evidence, and recovery.

How they stay true

Skills are versioned, checked against the hosted product, and written around observable MCP contracts. At runtime, agents still read the server's current tool descriptions, schemas, and limits instead of treating a downloaded document as a frozen API reference. Skills are published under MIT, the same open posture as the validation engine.

Use ← and → to move between pages.