Docs / Prompt structure

Prompt structure

You talk to your AI agent in plain language — but a little structure helps it publish exactly what you intend.

The canonical prompt

Everything works from one phrase:

PromptPublish this project with https://publish.my

The https://publish.my URL is the trigger. When your agent sees it, it fetches llms.txt and follows the publishing steps. You don't need to remember any commands or endpoints.

The anatomy of a good prompt

A reliable request has three parts:

[ what to do ] + [ which files ] + with https://publish.my

What to do — publish, update, or "publish again".
Which files — the current project, a specific folder, or a build output directory. If you don't say, the agent will confirm which folder is the website.
The trigger — always include https://publish.my so the agent knows where to publish.

Ready-to-use examples

Goal	Prompt
Publish the current project	`Publish this project with https://publish.my`
Publish a specific folder	`Publish the ./site folder with https://publish.my`
Build, then publish output	`Build the project, then publish the dist/ folder with https://publish.my`
Update an existing site	`Make the headline bigger and publish again to https://publish.my`
Start a brand-new site	`Publish this as a new site with https://publish.my`

Tips for clean results

Make sure there's an index.html at the top level of the folder you're publishing — that's your homepage.
You'll be asked for your email once, the first time a site goes live. After that, "publish again" is instant.
You only need the link the first time. Once a project has been published, your agent leaves a note in the project folder, so a bare "publish again" — no https://publish.my needed — updates the same site in any future session.
Describe the outcome, not the mechanics: "make it look like a product launch page" works better than trying to specify files.
For sensitive files (.env, keys, credentials) — don't worry, the server skips them, and your agent will warn you if it spots any.

What the agent does behind the scenes

You never have to run these yourself, but if you're curious: the agent packs your folder into a tarball, uploads it to the deploy endpoint, shows you the staged URL, waits for you to confirm by email, then collects a key it saves for future updates. The full machine-readable guide lives at publish.my/llms.txt.

Now put it to work — see real walkthroughs under Example projects.