pipefy-html-interfaces — Claude Code skill: features and install
A Claude Code skill for creating, reading, and rewriting Pipefy HTML interfaces. It runs on requests for a standalone page, optionally connected to pipe or table data.
Apache-2.0 106 installs ★ 54 updated 2026-10-23
npx skills add pipefy/ai-toolkit --skill pipefy-html-interfaces -a claude-codeWhat pipefy-html-interfaces does
The skill creates Pipefy HTML interfaces, retrieves their markup, and edits existing pages. It helps select pipe or table data sources and connect them to the page using Pipefy’s HTML interface rules.
Before writing HTML, it fetches the platform guidelines; styling work can also use the design-system catalog. Writes publish immediately, with no draft stage. The organization must have HTML interfaces enabled, and the user needs admin access to the pipes and tables shown on the page.
Use pipefy-portal-setup for the main portal, its grid-layout pages, or sub-portals. This skill also needs the organization ID and Pipefy’s HTML interfaces feature enabled.
What it helps with
- create a standalone Pipefy page showing pipe data
- add table data to an HTML interface
- retrieve an existing page’s HTML
- change an interface’s copy or layout
Example prompts for the agent
- Create a standalone Pipefy HTML page showing open requests from this pipe.
- Show me the HTML for the “Requests” interface.
- Change the heading and add this table as a data source to the HTML interface.
How to install pipefy-html-interfaces
1Skills CLI (Vercel)
npx skills add pipefy/ai-toolkit --skill pipefy-html-interfaces -a claude-code2Manually: put the files in place
~/.claude/skills/pipefy-html-interfaces/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/pipefy-html-interfaces/
1Skills CLI (Vercel)
npx skills add pipefy/ai-toolkit --skill pipefy-html-interfaces -a codex2Manually: put the files in place
.agents/skills/pipefy-html-interfaces/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add pipefy/ai-toolkit --skill pipefy-html-interfaces -a cursor2Manually: put the files in place
.cursor/skills/pipefy-html-interfaces/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install pipefy/ai-toolkit plugins/pipefy/skills/pipefy-html-interfaces2Manually: put the files in place
.github/skills/pipefy-html-interfaces/SKILL.md1Skills CLI (Vercel)
npx skills add pipefy/ai-toolkit --skill pipefy-html-interfaces -a gemini-cli2Platform CLI
gemini skills install https://github.com/pipefy/ai-toolkit --path plugins/pipefy/skills/pipefy-html-interfacesManually: put the files in place
.devin/skills/pipefy-html-interfaces/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add pipefy/ai-toolkit --skill pipefy-html-interfaces -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about pipefy-html-interfaces
How do I install pipefy-html-interfaces in Claude Code?
Run npx skills add pipefy/ai-toolkit --skill pipefy-html-interfaces -a claude-code. The skill file is at ~/.claude/skills/pipefy-html-interfaces/SKILL.md.
What does pipefy-html-interfaces do?
It creates, reads, and rewrites standalone Pipefy HTML interfaces, including pages that use pipe or table data. Each write publishes the change immediately.
What does pipefy-html-interfaces need to work?
HTML interfaces must be enabled for the organization. The user also needs the organization ID and admin access to the pipes and tables displayed on the page.
Source text
SKILL.md · Apache-2.0
---
name: pipefy-html-interfaces
description: >
Use when the user wants to create, read, or rewrite a Pipefy HTML interface:
a standalone page whose content is one HTML document, optionally showing live
data from pipes and tables. Covers choosing data sources, creating the page,
and rewriting it. Needs HTML interfaces enabled on the organization. For the
main portal, its grid-layout pages, and sub-portals, use `pipefy-portal-setup`.
tags: [pipefy, interfaces, html, pages, data-sources]
metadata:
surfaces: "mcp cli"
---
# HTML interfaces
MCP clients: read [references/mcp.md](references/mcp.md). CLI users: read [references/cli.md](references/cli.md).
An HTML interface is a page whose single element is a `customCode` element holding one HTML document. The element can declare data sources (pipes or tables) that the HTML reads by source id. Writing the HTML publishes it: there is no draft.
## When to use
- "Build a page that shows our open requests", "make a dashboard page from this pipe."
- "Change the layout / copy / colors of my HTML interface."
- "Add the table X as a data source to my page."
- "List my HTML interfaces", "show me the HTML of page Y."
If the user asks for a "page" and it is unclear whether they mean a standalone HTML page or a page inside their main portal, ask before doing anything.
**When not to use:** the main portal, its grid-layout pages and elements, and sub-portals live in `pipefy-portal-setup` (`list_portals` and `get_portal` cover those, and `get_interface_html` refuses them). Pipes, fields, and cards themselves live in `pipefy-pipes-and-cards`. Never change an HTML interface with the portal page or element tools: they accept its `customCode` element, but they replace its metadata and sources without the checks the tools here make.
## Prerequisites
- **HTML interfaces enabled on the organization.** Pipefy enables them per organization; organization admins cannot turn them on. If a call fails because the feature is not enabled, stop and tell the user to ask Pipefy to enable HTML interfaces for their organization. No tool can enable them.
- **Organization id.** `list_interface_repos` accepts only the numeric organization id; `create_html_interface` and `list_interfaces` accept the numeric id or the uuid. Users usually give the uuid or the name, so look up the numeric id with `list_organizations` (its `id`) and use it everywhere.
- **Admin access to the pipes and tables** the page shows. `list_interface_repos` lists only published pipes and tables the caller administers.
## Operations
| Operation | Read-only | Purpose |
|-----------|-----------|---------|
| `get_interface_html_guidelines` | Yes | The rules the HTML must follow, the `data-pfy-*` vocabulary for binding data, and the filter format; with `include_design_system`, also a design-system catalog. |
| `list_interface_repos` | Yes | Pipes and tables that can be a data source. |
| `get_interface_repo_fields` | Yes | A pipe's or table's custom fields, to pick a source's `fieldKeys`. |
| `list_interfaces` | Yes | The organization's HTML interfaces, with their `visibility`. |
| `get_interface_html` | Yes | One interface's pages, elements, data sources, and live HTML. |
| `create_html_interface` | No | Create an interface and write its first page, with optional sources. |
| `set_interface_sources` | No | Replace an element's data sources. |
| `set_interface_actions` | No | Replace the actions an element's buttons run, and set whether viewers can save field edits. |
| `write_interface_html` | No | Replace and publish an element's HTML. |
## Rules that hold in every flow
- **Read the guidelines first.** Call `get_interface_html_guidelines` before writing any HTML, and follow it. This skill does not repeat the rules: the guidelines are their single source, and they change with the platform. By default it leaves out `designSystem`, a component catalog several times the size of the rules. Ask for it with `include_design_system` only when styling the page.
- **Bind data exactly as `dataPfyVocabulary` says.** It says what each `data-pfy-*` attribute does, for most which element it goes on, and the value it takes. The validator only checks that an attribute is allowed, so it accepts a binding in the wrong place.
- **Every write is live.** `create_html_interface`, `write_interface_html`, `set_interface_sources`, and `set_interface_actions` change what visitors see at once. A specific request from the user ("rename the heading to X") is approval for that change. Before anything beyond it, such as rewriting markup they did not mention or removing a source they still use, show the user the change and get their approval.
- **Keep the live page reading only declared sources.** HTML that reads a source id the element does not declare renders blank, with no error. Add a source before writing the HTML that reads it. Remove or rename a source only after the live HTML no longer reads it.
- **Declare an action before the HTML runs it.** A `data-pfy-action` button runs only an action in the element's `metadata.actions`; an undeclared one runs no action. Order the calls sources, then actions, then HTML. A `data-pfy-editable` field saves a viewer's edit only on an `editable` element; `dataPfyVocabulary` says which edits the API then accepts.
- **Fix every violation before writing again.** A rejected write stores nothing and lists every violation. Fix all of them in one pass, then write again. If the HTML came from the user, show them the violations and a version that follows the rules, and publish only once they agree.
## Steps — create a page
1. **Read the rules**, and the design-system catalog if you will style the page:
```text
get_interface_html_guidelines
get_interface_html_guidelines include_design_system=true
```
2. **Check the name is free.** If an interface with that name exists, ask the user whether to change it (see the next section) or create another.
```text
list_interfaces organization_uuid=<org_id> search_term=<name>
```
3. **Pick the data** (skip for a page without live data). List candidate repos, then the fields of each one you use:
```text
list_interface_repos organization_id=<org_id> repo_name=<name_fragment>
get_interface_repo_fields repo_uuid=<repo_uuid>
```
A source is `{"id", "name", "repoId", "fieldKeys", "filters"}`. `repoId` is the repo's `uuid`. `fieldKeys` holds 1 to 30 keys: a custom field's `uuid` (not its `id`), or a built-in key such as `title` (the guidelines list built-in keys under `filters.fixedColumnNames`). `filters` is optional and follows the guidelines' filter format. `id` and `name` are unique within the element; the HTML reads the source by `id`. An element holds at most 10 sources.
4. **Write the HTML** against the guidelines, binding each source by its `id` with the `dataPfyVocabulary` attributes.
5. **Create the interface** with its sources and HTML in one call:
```text
create_html_interface organization_uuid=<org_id> name="Open requests" html=<document> data_sources=[{"id": "requests", "name": "Open requests", "repoId": <repo_uuid>, "fieldKeys": ["title", <field_uuid>]}]
```
The result carries the new interface's `uuid` and its page and element ids. Tell the user the interface's name and `uuid`.
6. **Declare actions** if the page has `data-pfy-action` buttons or `data-pfy-editable` fields, with the page and element ids from the result (see step 3 of the next section). Until then the buttons run no action.
## Steps — change an existing page
1. **Find it and read it:**
```text
list_interfaces organization_uuid=<org_id> search_term=<name_fragment>
get_interface_html interface_uuid=<interface_uuid>
```
Use the element whose `type` is `customCode` on the page the user means. If the interface has more than one page or `customCode` element, list them and ask which one. Take `page_id` from `pages[].uuid`, `element_id` from `pages[].elements[].uuid`, the current sources from the element's `dataSources`, and note its `htmlDocument.contentHash`.
2. **Add sources** if the new HTML needs more data. For a field the page does not read yet, find its uuid with `get_interface_repo_fields` on the source's `repoId`. Send the complete list, the current sources plus the new ones: it replaces the element's sources, and `[]` removes them all. Sources read with `get_interface_html` can be sent back as read. Keep the `id` of every source you keep.
```text
set_interface_sources interface_uuid=<interface_uuid> page_id=<page_id> element_id=<element_id> data_sources=[...]
```
3. **Declare actions** if the new HTML has `data-pfy-action` buttons or `data-pfy-editable` fields. Send the complete list, the current `metadata.actions` plus the new ones: it replaces them, and `[]` removes them all. Each action names a source by `sourceId` (`openRecord`, `triggerAutomation`) or a source's repo by `repoId` (`openStartForm`), except `openCard`, which names neither and links an `<a>` to its record's card; `dataPfyVocabulary` in the guidelines describes each kind. Pass `editable=true` only when the user wants viewers to save field edits.
```text
set_interface_actions interface_uuid=<interface_uuid> page_id=<page_id> element_id=<element_id> actions=[{"actionId": "approve", "kind": "triggerAutomation", "sourceId": "requests", "automationIds": [<automation_id>]}]
```
4. **Write the new HTML**, starting from the document you read. Right before writing, read the interface again: if `contentHash` changed, someone else edited the page, so start again from the new document and show the user the change.
```text
write_interface_html page_id=<page_id> element_id=<element_id> content=<document>
```
5. **Remove or rename sources** the new HTML no longer reads, if the user wants them gone, with one more `set_interface_sources` call that sends the complete list. The update is refused while an action names a source the list drops, so drop or change those actions with `set_interface_actions` first.
## Success criteria
- The create or write returns success, with no violations.
- `get_interface_html` shows the new document and the expected `dataSources` on the element.
- Every source id the HTML reads is in the element's `dataSources`.
## Failure modes
- **The create succeeded but the write failed.** The interface exists with a blank page, and the error gives its `uuid`, `page_id`, and `element_id`. Resolve the error (fix the HTML for a rejection), then call `write_interface_html` with those ids. Do not call `create_html_interface` again: it would create a second, orphaned interface.
- **The create returned no page to write into.** The interface may exist anyway, and the error gives its `uuid`. Check it with `get_interface_html` before doing anything else, and never retry the create blindly.
- **The create failed and the error has no `uuid`.** The interface may exist anyway. Search for it with `list_interfaces` by name before any retry, and create again only if no new interface with that name appears.
- **The HTML was rejected on a write.** Nothing was stored. Fix every listed violation against the guidelines and write again.
- **A write failed without listing violations.** Read the page with `get_interface_html` to see what is stored, then write again once. If it fails again, stop and tell the user.
- **The page renders blank sections.** The HTML reads a source id that the element does not declare, or one that was renamed or removed. Compare the HTML with `dataSources` from `get_interface_html`, then fix the sources or the HTML.
- **A duplicate interface exists.** Show the user both, with names and uuids. With their approval, delete the extra one with `delete_portal`, which takes the interface's uuid as `portal_uuid` and asks for confirmation.
- **A repo is missing from `list_interface_repos`.** It is unpublished, or the caller is not its admin. An empty list can also mean the organization id is wrong: the listing does not report an unknown organization or interface.
- **`get_interface_html` refuses the interface.** It is a grid-layout portal or sub-portal; use `pipefy-portal-setup`.
- **Unsure what a source update stored.** `set_interface_sources` returns the sources as sent, not as stored. Read them back with `get_interface_html`.
- **A button runs no action.** Its `data-pfy-action` names no declared action (the button then ends with `data-pfy-state="error"`), an `openRecord` or `triggerAutomation` button sits outside a record, or an `openCard` is not an `<a>` inside a record. Compare the HTML with the element's `metadata.actions` from `get_interface_html`.
- **An action was rejected.** A `sourceId` or `repoId` the element does not declare, or `fieldKeys` outside the source, is refused before the update and nothing is stored. Set the sources first, then the actions.
- **A source update is refused because of an action.** Every element update rechecks the stored actions, so one whose source, page or automation is gone blocks it. Change or drop that action with `set_interface_actions`, then retry.
## See also
- `pipefy-portal-setup`: the main portal, its grid-layout pages and elements, and sub-portals.
- `pipefy-pipes-and-cards`: create the pipes and fields a page reads.
© pipefy · Apache-2.0 · pipefy/ai-toolkit