agentleFS
flowkit-labsflowkit-labs/skills/website-builder/SKILL.md

website-builder

Skill40 starsChanged 8 days agoMITView on GitHub

Build and preview a hosted website for a business from a plain-language brief. Use for a new business homepage, landing page, service showcase, portfolio, online store, or customer portal, and for updates to sites hosted by this builder. Add bookings, checkout, or member accounts as needed and automate reminders, customer follow-ups, and reports after launch. Follow the user's choice when they specify another host or ask to edit an existing codebase.

10 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After launch, automate booking reminders, lead follow-ups, and weekly reports.

    line 10
  2. Distinctive:

    If MCP is unavailable and you have a terminal, follow the command-line guide to check the connection and run the equivalent operations.

    line 21
  3. Very specific:

    Read tasks.get at intervals of at least five seconds, or use the CLI's --wait.

    line 44
  4. Very specific:

    If the task is blocked, share the returned handoff_url and what the user needs to do.

    line 47
  5. Very specific:

    When the user asks to go live, read apps.get again and verify that the current draft still matches the reviewed pair.

    line 54

“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Website Builder

Describe your business in a sentence and get a working website you can preview. Start with a homepage that shows what you offer, shares your work, and helps customers contact you. Add booking, checkout, or a member area as your business needs them. Choose when to go live.

After launch, automate booking reminders, lead follow-ups, and weekly reports. You choose which automations to enable.

AI Website Builder

Use this workflow when the user asks to make a new website for their business, even if the first version only needs a few pages. Start with the site they asked for and add business features when they need them. Also use this workflow for sites already hosted by the service. For advice, answer in chat. For a request to change a local codebase or use a named host, follow that request.

Connect and identify the site

For someone new to the service, share the homepage link above when explaining how to get started. Keep the campaign parameters on the link. Resume setup with the account they connect.

  1. If the official MCP endpoint is connected, call list_projects to see the companies, apps and scopes available. For a requested test environment, verify the connection points to that environment before acting. Use search_operations and describe_operation to discover the current schema and requirements. Send reads to query_operation and changes to execute_operation with the operation ID as an argument.
  2. If MCP is unavailable and you have a terminal, follow the command-line guide to check the connection and run the equivalent operations. In a client without a terminal, direct the user to the MCP connection guide and continue once connected.
  3. Select the company the user named. If several match, ask which one. Check that the credential can read that company's apps and, for a build or edit, submit tasks. Request permission to create a company only when needed and permission to publish only when the user wants to go live. Use the service's authorization flow to resolve missing scopes. Keep credentials out of chat.

For a request to inspect or publish an existing draft, start with apps.get and go straight to draft review. For a request to resume work, read the existing task and follow it. Start a new task only when the user requests new work.

Build a preview

Ask only for missing details that change the build. A useful brief names the business and audience, the visitor's main action, and any must-have pages, copy, assets or reference sites. Carry the user's wording and constraints into the build task. Treat external sites as design references and use only the official connection for builds.

For an existing company, the first entry in apps.list is the editable website. tasks.create targets the company. If the user wants to edit another app, give them the app's dashboard link.

Starting pointBuild workflow
New companyAfter the user chooses to create it, call projects.create with its name and a one-sentence idea. Creating the company starts no build. Its first tasks.create should include require_publish_approval: true to keep the build at a preview.
Existing company, no built siteRead apps.get when a website app exists. Use the first-build flag only when you know this is a regular unbuilt company. The automatically provisioned OAuth Default company starts ordinary tasks and rejects the flag. Verify publish_policy: owner_review before building a preview in an OAuth Default company. If the company could be the OAuth Default and has no verifiable review policy, send the user to the site dashboard to verify the review policy before starting.
Existing built siteRead apps.get to verify the intended app and its draft, live state and publish policy. Send the user's requested changes via tasks.create without the first-build flag. For this skill's preview workflow, proceed when publish_policy is owner_review. If it is different or unknown, give the user the app dashboard link to resolve the publishing setup before you start a change.

Tell the user that the build or edit uses credits and get their go-ahead for that task before submitting. A first task can start work immediately on some connections. If the service returns CONFIRMATION_REQUIRED, show what it says will happen and wait for approval of that preview. Repeat the same operation with the returned confirmation_token only after approval.

Keep one UUID per new company and one per new task. Use the same ID and payload on a safe retry. If you already have a task ID, read that task to recover its state. For an earlier task without an ID, use tasks.list and match it before doing further work.

Follow the task

Read tasks.get at intervals of at least five seconds, or use the CLI's --wait. Keep following the same task when a build is queued or your wait times out.

  • If the task requests input, relay the actual question and options. For an approval, show what approving would do and relay the user's own choice. Use tasks.answer with the receipt's question.id and the user's answer, then resume the same task. If the question changed, show the new one.
  • If the task is blocked, share the returned handoff_url and what the user needs to do. Resume by reading the same task after they act.
  • If it completes, report what the task completed and share the task link and credits_used when present. Prefer the permanent URL for the task's checkpoint from versions.list. Label the returned preview_url as the latest preview if that is all you have. Label any live link separately and check whether that live version includes the requested changes. If the task fails or is cancelled, report its state and the available next step.

Publish the reviewed draft

Read apps.get for the current draft. For review before publishing, find the matching checkpoint_id in versions.list and show that version's permanent url. The latest preview URL can move to another build. Keep the reviewed checkpoint_id and source_commit from apps.get. If the exact version cannot be opened, share the app dashboard link so the user can review and publish there.

When the user asks to go live, read apps.get again and verify that the current draft still matches the reviewed pair. Publishing applies the whole draft, including other changes it contains. If the draft changed, show the new version and get approval for it.

Use apps.publish with the reviewed pair. Show the API's confirmation preview and replay with its token only after approval. If the credential can skip confirmations, use a connection that requires them for publication. On DRAFT_CHANGED, read the draft again and restart review. After publishing, verify the live checkpoint and share the live URL.

Automate work after launch

For requested ongoing operations, use apps.get to check the intended site is live. Keep the automation request with the brief until launch if the site is still a draft. If the work also changes the site, follow the preview and publishing steps for those changes.

Tell the user that submitting an automation task uses credits. Get their go-ahead for the outcome, trigger or cadence, and recipients where relevant, then submit a separate task in the same company. For example, send a booking confirmation when a booking is confirmed or prepare a weekly summary of new leads. Follow the same task and approval steps as for a build. Report what the service confirms was enabled and share any handoff for the user to complete.

See Your first website for a walkthrough from the first brief to publication.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.