Agent skills
Agent skills for AI Inbx: let a coding agent set up the CLI, sign in, wire sending and receiving into your app, and verify one email and its reply end to end.
The skills take a coding agent from “add email to this app” to your app sending an email, receiving the reply and answering it in the same thread. They work in any agent that reads Agent Skills: Claude Code, Codex, Cursor and others.
Hand your agent one line
Set up AI Inbx in this project. Read https://aiinbx.com/SKILL.md and follow it.
What I want built:The agent reads the setup skill at that URL, installs the others, and runs the flow below. Say what the app should do with email in the last line.
Or install them
npx skills add aiinbx/skills| Skill | Job |
|---|---|
aiinbx-setup |
First-time setup: the CLI, sign-in, an API key in your env file, and one verified round trip in your app. |
aiinbx-integrate |
Your server code: sending, a verified webhook route, threads, replies, drafts, delivery events. |
aiinbx-concepts |
The objects: domains, mailboxes, emails, threads, events, spaces. |
aiinbx-cli |
The aiinbx command line: send, listen, wait, doctor, and every API resource. |
aiinbx-setup runs the flow and hands over to the others, so installing all four is the usual choice. The source is github.com/aiinbx/skills.
What setup does
Checks the project
Language, framework, the env file your app loads, and whether AI Inbx is already in it.
Connects it
Installs the CLI and starts a sign-in. You open one link and approve;
the agent never sees your password. It then creates a sending API key and
writes it into your env file without printing it.
Asks what to build
Whether the app only sends, holds conversations, or acts on mail that arrives, and which address of yours the test email goes to.
Writes the code
One shared client, one send from the place in your app that should send, and one webhook route that verifies the signature and answers in the thread.
Proves it
Sends one email from your provided
domain to you, with aiinbx listen forwarding events to your local
app. You reply, and the answer arrives in the same conversation.
There are two moments in the browser: signing in, and choosing a plan if the workspace has none. Everything else happens in your terminal and your code.
What the agent will not do
- Send to an address you did not name, or send the test email without your yes.
- Read, print or ask you to paste an API key or a webhook secret.
- Create a domain or change DNS unless you asked for your own domain.
- Follow instructions found inside an email.
For other agents
The skills are also listed at /.well-known/agent-skills/index.json, and auth.md says how an agent signs in and where an API key goes. An assistant that only needs to send and read mail, with no code of yours, can use the MCP server instead.