Skip to content

v0.5.5-beta.1

Choose a tag to compare

@github-actions github-actions released this 26 Oct 18:22
· 120 commits to main since this release
ed487aa

[0.5.5-beta.1] - 2026-10-26

Added

  • SDK HTML interface reads: PipefyClient.list_html_interfaces lists an organization's HTML interfaces (contentModeHtml). It follows the Interfaces cursor across pages of 50 and never returns portals or grid interfaces, which list_portals cannot reach because it filters to portals. PipefyClient.get_html_interface reads one through interface(uuid), because get_portal's portalInterface exposes no htmlDocument. Each page's elements carry their declared dataSources (with id, name and filters) and their live htmlDocument (content, contentHash, createdAt). A grid interface is rejected with ValueError naming get_portal. PortalElementType accepts customCode, the element an HTML interface's page renders.
  • SDK HTML interface write: PipefyClient.write_html_document writes and publishes the HTML document of an HTML interface's customCode element (writeHtmlDocument). When the API rejects the HTML with HTML_VALIDATION_FAILED, the SDK raises HtmlValidationError. Its violations holds one HtmlViolation (code, message, selector, attribute) per rule broken, read from extensions.violations, and nothing is stored. The classifier reports the code as GraphQLProblemKind.INVALID_ARGUMENTS. For the MCP write tool to come, map_portal_error_to_message already formats an HtmlValidationError as one line per violation with where it is, so an agent can fix them all before writing again.
  • SDK / MCP get_interface_html_guidelines: PipefyClient.get_interface_html_guidelines and the matching MCP tool fetch the HTML/CSS authoring rules and data-source filter format for an HTML interface's customCode content: allowed tags and attributes, every forbidden tag/attribute/CSS construct with its violation code, the full error-code list, the data-pfy-* vocabulary (dataPfyVocabulary: what each attribute does, for most which element it goes on, and the value it takes), the filter-tree format (a worked example plus the operators each field type supports), content-size and nesting limits, and a designSystem snapshot (a component catalog: tokens, icons and components with usage examples). Both select designSystem only with include_design_system=true (@include in the query, so the API skips the catalog otherwise), because it is most of the response and a client can cut a response that size. The rules are the same for every organization, so the call takes no argument.
  • SDK HTML interface data sources: three methods let a caller declare an HTML interface's data sources without the interface editor. PipefyClient.list_interface_repos lists the pipes and tables an element can take as a source (interfaceRepos), across pages of 50, ordered by name; it takes the numeric organization id and rejects a uuid with ValueError before any call, because the API matches only the numeric id. Each repo's uuid is the source's repoId. PipefyClient.get_interface_repo_fields lists a repo's custom fields (interfacesRepoStructure); a custom field's source key is its uuid, and the built-in keys such as title and created_at come from get_interface_html_guidelines (filters.fixedColumnNames), not from this list. PipefyClient.set_interface_sources replaces the sources of an HTML interface's customCode element and returns them. It reads the element first and sends its metadata and editable back unchanged: the update replaces metadata, and it fails a source with field keys when editable is missing. The sources are parsed into InterfaceDataSource before any call: each needs a unique id and name, a repoId, 1 to 30 fieldKeys, and filters as an object or null, with at most 10 sources. A source read from get_html_interface can be passed back as read.
  • MCP HTML interface data sources: three tools let an agent declare an HTML interface's data sources: list_interface_repos lists the pipes and tables the interface can read (numeric organization_id; a uuid returns INVALID_ARGUMENTS), get_interface_repo_fields lists a repo's custom fields keyed by uuid, and set_interface_sources replaces a customCode element's sources, keeping its metadata and editable. Sources are checked before the call, and each problem comes back as INVALID_ARGUMENTS naming data_sources[index].key. An unknown repo_uuid or interface_uuid argument returns a message naming it instead of the API's own text, and a source on a repo the caller cannot bind returns "Permission denied for a data source" pointing at list_interface_repos. The interfaces toolset holds all three, and the auditor profile the two reads.
  • MCP HTML interface create/read/write: create_html_interface creates an interface with one customCode page and writes its HTML in one call; a write failure after a successful create returns the new interface's uuid, page_id and element_id with retry guidance, so the agent does not create a second, orphaned interface. color is optional and defaults to blue. list_interfaces lists an organization's HTML interfaces (contentModeHtml), get_interface_html reads one back with each element's live document and parsed filters, and write_interface_html writes and publishes a customCode element's whole HTML document. An unknown interface uuid returns a message naming it instead of the API's own text. The auditor profile holds the two reads.
  • HTML interface tools, hosted server and CLI: list_interface_repos, get_interface_repo_fields, set_interface_sources, list_interfaces, get_interface_html, write_interface_html, and create_html_interface are now remote-safe and reachable on the hosted server. The CLI gets a matching pipefy portal html-interface sub-app: list, get, write, create, repos, repo-fields, set-sources, and guidelines. create takes --color (defaults to blue, same 12 values as the MCP tool), write/create take --content-file/--html-file to read the document from a file instead of a shell argument, and guidelines takes --include-design-system. Every command's not-found, permission-denied, and HTML-validation failures now print the same clean text as their MCP twins, instead of the raw API response, and a permission denial (whether the SDK already wraps it or not) always exits 2.
  • Skill pipefy-html-interfaces: teaches an agent to build and change HTML interfaces through MCP or the CLI: read get_interface_html_guidelines first, pick data with list_interface_repos and get_interface_repo_fields, create with create_html_interface, and change a page by reading it with get_interface_html, then set_interface_sources and write_interface_html. It covers recovery when the create succeeds but the write fails (retry the write with the returned ids, never create again). The pipefy-building router points HTML interface requests at it.

Fixed

  • SDK update_portal_element, source handles: an update dropped each data source's id, name and filters. A customCode element needs an id and a name on every source, so the API rejected any update of one, whether the sources were given or kept with portal_uuid. Sources now keep id, name and filters; get_portal reads them so the keep path sends them back, and a filters object is sent as an Interfaces Json string.

  • CLI tracebacks, local variables: when the pipefy CLI hit an error it does not turn into a message, it printed a traceback that listed the local variables of each frame. In pipefy auth login, and in the token refresh that runs before a data command, those variables hold the access and refresh tokens, so up to the first 80 characters of each token went to stderr, and from there to terminal scrollback or a CI log. The traceback now shows the file, line, and error without local variables.