# Agent Skill

> Teach coding agents how to install, set up, configure, and use routeup safely.

Routeup publishes an official [Agent Skill](https://agentskills.io) for coding
agents that support the open `SKILL.md` format. The skill teaches an agent how
to install the CLI, perform one-time machine setup, configure a project, choose
between runner and serve modes, expose a route deliberately, and inspect traffic
without leaking credentials.

## Install the skill

Install it globally when you want Routeup available to your agent in every
project:

```bash
npx skills add mukul-mehta/routeup --skill routeup -g
```

The Skills CLI detects installed agents and offers compatible destinations,
including OpenCode, Claude Code, Codex, Cursor, and GitHub Copilot.

To install it only for the current project, omit `-g`:

```bash
npx skills add mukul-mehta/routeup --skill routeup
```

A project installation can be committed when everyone working in the repository
should receive the same guidance. Review the generated agent-specific files
before committing them.

The canonical source is
[`skills/routeup/SKILL.md`](https://github.com/mukul-mehta/routeup/blob/main/skills/routeup/SKILL.md).

## Let the agent set up Routeup

After installing the skill, give the agent a direct request:

```txt
Use the Routeup skill to install and set up Routeup on this machine. Verify it
with routeup doctor, then give this project a stable local HTTPS route.
```

That request authorizes the skill's installation and setup workflow. The agent
will:

1. Check whether `routeup` is already installed.
2. Install it with Homebrew or the official installer when necessary.
3. Run `routeup setup` for the local CA and port 443.
4. Leave Touch ID or administrator-password approval to you.
5. Verify the machine with `routeup doctor --json`.
6. Inspect existing project configuration before changing anything.
7. Start a local `.localhost` route without exposing it publicly.

Machine setup changes the OS trust store and privileged port handling. If you
did not explicitly request setup, the skill tells the agent to explain those
changes and ask first. It never instructs an agent to bypass an interactive
privilege prompt or disable TLS verification.

## Common prompts

Route an application that is already listening:

```txt
Use Routeup to serve the app currently listening on port 3000 as
https://dashboard.localhost. Keep it local only.
```

Configure runner mode for the current project:

```txt
Use the Routeup skill to inspect this project and add the smallest valid runner
configuration. The app must honor the HOST and PORT supplied by Routeup.
```

Create a public webhook endpoint:

```txt
Use Routeup to expose this project's webhook endpoint publicly. Limit exposure
to /api/webhooks/* and redact authorization, cookie, and signature headers from
capture. Ask before changing any existing config.
```

Prepare a local service for an off-machine browser agent:

```txt
Use the Routeup skill to make this dev server reachable by a remote browser
agent. Explain the public exposure before starting it and return the granted URL.
```

## Safety defaults

The skill instructs agents to:

- Prefer trusted local `.localhost` routes over public tunnels.
- Ask before setup or exposure unless the request already authorizes it.
- Keep tokens out of prompts, command output, and project configuration.
- Enable request or response capture only when requested.
- Redact sensitive headers from retained traffic.
- Use structured CLI output instead of parsing terminal styling.
- Stop foreground or detached owners through Routeup's normal lifecycle.
- Treat the installed CLI's `--help` output as the version-specific authority.

The skill is instruction-only. It contains no executable scripts and grants no
tools or shell commands permission in advance; the selected agent still applies
its normal permission and sandbox rules.