Use with AI agents

Give your coding agent the current Form, Please documentation, so that it does not rely only on its training data. You can do this in two ways:

• Paste a prompt that points the agent to the LLM files and their fallback links. Your agent needs web or HTTP access.
• Install the Form, Please skill. This works without a prompt.

The llms.txt file is the documentation index.

Give your agent the documentation

Copy this prompt into your agent session, and then add your task:

Read https://r13v.github.io/form-please/llms.txt and follow its links to the
HTML pages that concern my task. If your web tool rejects the direct URL,
open https://github.com/r13v/form-please, follow "Use with AI agents", and
then follow "llms.txt" in the Documentation files table. Use link clicks,
not another direct open of the file URL. If a direct HTML URL also fails,
follow its navigation link from the AI agents page. With shell/HTTP access,
fetch the same URL with an available HTTP client instead. If text/markdown is
unsupported, read the corresponding HTML page or the relevant section of
llms-full.txt, linked from the same table. Treat those pages as the source
of truth for Form, Please. Use only documented APIs. If you cannot read
the pages for my task, report the access failure before implementing code.

Documentation files

In the page Markdown URL, <page> is the page path without the first slash. For example, the Markdown of /form-please/get-started is /form-please/assets/md/get-started.md, and the Markdown of the overview is /form-please/assets/md/index.md.

If a documentation URL fails

Use this sequence when your agent cannot read a direct URL:

1. Open the Form, Please repository.
2. Follow Use with AI agents in its Documentation list.
3. Follow llms.txt in the table above.
4. Read the HTML pages listed in the index for your task. If direct opens fail, follow their navigation links from the AI agents page.

Codex web.run has read llms.txt through this sequence while rejecting a direct open of the same URL. Use click for each link in that tool. The cause of this difference is unknown.

With shell/HTTP access, fetch the same URL with an available HTTP client. For example, use Node.js:

node --input-type=module -e 'const r = await fetch("https://r13v.github.io/form-please/llms.txt"); if (!r.ok) throw new Error(`HTTP ${r.status}`); console.log(await r.text())'

If curl is available, use:

curl -fsSL https://r13v.github.io/form-please/llms.txt

curl is not required. The Node.js command uses its built-in HTTP client.

Some web tools reject the text/markdown response of a page Markdown URL. Read the corresponding HTML page instead. For example, read Get started for assets/md/get-started.md. For Markdown with the supported text/plain response, follow llms-full.txt in the table above. An HTTP client can fetch that file from its URL too. Find the relevant page heading in the full file. If the tool truncates the response, read further or use the individual HTML page.

Both LLM files return text/plain; charset=utf-8 on the published site. Codex web.run has read both through links on this page. If none of these paths gives your agent the pages for its task, it must report the access failure before it implements code that depends on them.

Vocs can leave [!include …] references in Markdown examples. Read the corresponding HTML page for the complete code of those examples.

Install the skill

The skill follows the open Agent Skills format. It gives instructions to your agent. It does not install the Form, Please npm package or change your application.

Run one of these commands from your project directory:

npx skills add r13v/form-please --skill form-please

npx skills add r13v/form-please --skill form-please --global

Follow the CLI prompts to select your agent and the installation method. Use the project scope when only this project uses Form, Please.

The command downloads the skill from the public Form, Please repository. Review SKILL.md before you install or update it, because your agent follows the instructions in that file.

Install without the CLI

Copy the complete skills/form-please directory from the repository to a skill directory that your agent supports. For agents that support the shared project convention, use this path:

.agents/skills/form-please/SKILL.md

Check your agent documentation if it uses a different skill directory.

Use the skill

A compatible agent can load the skill when your request concerns Form, Please. Name the skill in your request when you want to activate it explicitly. You do not need a product-specific slash command.

For an implementation task, use a request such as:

Use the form-please skill. Build a profile form with Zod and the native preset.
Keep the existing design system.

For a review task, use a request such as:

Use the form-please skill. Review this form definition for invalid schema paths,
validation mistakes, and submission errors.

When the skill activates, it tells the agent to read llms.txt and the pages for its task. It includes the same web and HTTP fallback sequence described above. The agent uses that documentation as its source of truth.

Verify the installation

List the installed skills:

npx skills list

Confirm that form-please appears for the agent that you selected. Start a new agent session if the current session does not discover the new skill. Then send one of the example requests. If your agent shows its activity, confirm that it reads SKILL.md and the LLM documentation before it changes code.

Update the skill

Update the skill with this command:

npx skills update form-please

Add --global if you installed the skill for all projects.