Let a coding agent add Configure

A skill is a page written for an agent instead of a person. You paste its link into the agent, and the agent does the work. There is nothing to install and nothing to configure. Configure publishes two.

Add Configure to your project

Paste this into Claude Code, Cursor, or Codex, as a message, in the project you want to add Configure to:

Read https://configure.dev/skill.md and add Configure to this project.

The agent reads the skill and builds the same six steps the Quickstart teaches, in your code, for the surfaces your app has. Here is what happens, in order:

1. It reads your code and decides two things without asking: whether your app has a chat, an onboarding flow, or both, and whether it has users of its own. If it does, profiles are keyed by your user id. Only an app with no accounts at all gets Sign in with Configure.
2. It installs the SDK. If your credentials are not in the environment yet, it asks you to run npx configure setup --users, because that step opens a browser and needs you.
3. It adds the entry point: the chip in your chat, wired to send the fixed message, or the button in onboarding, wired to go to Configure's page and come back.
4. It answers the fixed message with profile.connect() and mounts Configure's card, or adds the onboarding return route, and makes the entry point flip to Connected.
5. It adds the daily refresh, the first-turn injection, a commit at the end of every turn, and the configure_* tools on your model call, Anthropic, OpenAI, or Vercel, whichever you use.
6. It writes a test against the sandbox, runs it, and shows you the result before it says it is done.

What you do: run setup when asked, read the diff, and run the app. Expect it to take a few minutes.

What it will not do, because the skill forbids it: build a Configure link by hand, keep a connected flag of its own, put your secret key anywhere a browser can see it, or leave the refresh and the commit to the model.

TIP

No account yet? The skill starts from the sandbox, so the agent can build and test the whole loop before you sign up.

Make your agent use what it knows

Giving a model the Configure tools is not the same as the model using them. Left alone, a model calls a search tool when it wants an answer and never when it should check what it already knows about the user. This skill is the set of habits that closes that gap, and it goes into your agent's instructions, not your code:

• Search the profile before saying you do not know something about the user.
• Save a durable fact the moment the user states it, without asking permission.
• Keep project conventions and code style out of the user's profile.

https://docs.configure.dev/skill/agents-block.md

Append it to your agent's system prompt or rules file. For Claude Code, this installs it along with a hook that loads a short digest of the user at the start of every session:

curl -fsSL https://docs.configure.dev/skill/install.mjs | node -

Per-client steps for Codex, Cursor, Windsurf, and Gemini CLI are on the coding agents page.

For platforms that host agents

If you run a platform where other people's agents live, the integration moves one level up: your platform provisions the keys and every agent on it gets the loop. Maritime has a skill for exactly that today, and a section for platforms is coming.

https://configure.dev/configure-maritime-skill.md

How a skill earns its place

Every skill on this page is tested before it ships, the same two ways:

• A quiz. A small model is given only the skill and asked the questions a developer asks when adding Configure to a chat app: where the entry point goes, what the fixed message triggers, how to tell whether a user is connected, when to refresh, what to commit, how to wire each model provider, and what never to do. It has to answer every one correctly.
• A build. A fresh agent is given only the skill and a real app, and what it produces has to run against the sandbox.

If you write a skill for your own platform, those two tests are yours to run too, and two lines decide whether it passes. Name the host, api.configure.dev, in full, because an agent that is not told the address guesses one. And show the tool wiring, mcp_servers handed to the model, not only the session mint, because a session nobody hands to the model fails silently.