docs_dataset_rows
ChatGPTRead rows (form submissions) from a Fluix document dataset. Each row carries the 6 system fields plus the dataset's form fields, keyed by their real names - read them off the row, no schema call needed first. Cursor-paginated with a fixed upstream page size (up to 1000 rows per call): follow nextCursor until hasNext is false. Prefer format: "csv" on wide datasets to keep the output compact. Pass additionalColumns: ["document_id"] to include the source document id (lets you join dataset rows with documents).
docs_dataset_schema
ChatGPTList the columns of a dataset. Returns two groups: systemFields (the 6 predefined fields with descriptions) and formFields (dataset-specific custom form fields). Names match the keys in row exports, so docs_dataset_rows needs no schema call first - use this tool only when the user asks what the columns are.
docs_list_datasets
ChatGPTList all document datasets the user has access to (masters + subviews). master is null for master datasets, {id, name} for subviews. Use the returned IDs with docs_dataset_schema / docs_dataset_rows.
docs_status_rows
ChatGPTFind documents by status, condition, workflow, owner, group, date range, or name. Page-based pagination - limit becomes the upstream page size so pages line up with the rows you receive. Use nextPage for the next call. Filters compose. Every distinct filter combination needs its own call - the rows returned are valid only for the filters passed in that same call, so re-call this tool instead of narrowing an earlier result yourself. Pass format: "csv" for compact text output. Requires an admin role with access to Document status; if the request is refused, say so instead of retrying.
docs_status_schema
ChatGPTList the field names available on every document-status row. 10 predefined fields with fixed meanings - no dynamic fields. Names match the keys returned by docs_status_rows.
tasks_list_reports
ChatGPTList all task reports the user has access to. Each report has a type: - process - process-view report (one row per task) - form - form-view report (one row per submitted form) Pass type from this list into tasks_report_rows to skip an extra lookup on the row call.
tasks_media_thumbnails
ChatGPTResolve media ids from submitted form data to downloadable image URLs, up to 100 per call. An image field's value in tasks_report_rows is one media id. An image gallery's value is several media ids joined by ", " - split it on that separator and pass one id per `mediaIds` entry. A joined value is not a media id and fails the whole batch, so never pass one through as it came. This turns those ids into URLs you can fetch. The URLs are short-lived - a few minutes. Download them immediately; do not collect them across several calls and fetch later, and do not store or return them as a lasting reference. size is the longest edge in pixels, 80-1280, defaulting to 1280. 80 is a UI list thumbnail and is unreadable once anyone zooms in, so ask for a smaller size only when a small image is genuinely what you want. Ids for placeholder media resolve to a placeholder URL rather than failing, so a mixed batch always returns something for every id.
tasks_report_form_fields
ChatGPTDescribe every field of every form behind a task report, grouped by form. This is the whole form, not the report's current column selection - a report whose columns are all process/task fields still returns its forms' fields in full, because the question is what the report's data can contain at all. Generated report forms are left out: they are built from another form's submissions rather than filled in, so their fields restate the source form's. Each form carries its formType - dynamic, pdf or web - and it changes what the fields look like. A dynamic form is the only one whose scheme holds question text, so only its fields carry a label; a choice field there is an enum with its full options, and a checklist question is a composite. On pdf and web forms there are no labels at all, so use the field id; a dropdown or radio group is a plain string with no options because the allowed set is not stored, and a checkbox is a bool whose id names its option, as checkbox-8:Option 1. Fields come back as an ordered list, and every field carries the same eight keys at every depth - a key that does not apply is empty rather than missing, so none of them needs a presence check. id is the field id, matching the keys of a submitted form's data (null on a collection's element, which is anonymous); label the question text a person answered, null where there is none; type one of string, number, bool, date, datetime, time, enum, image, signature, email, collection or composite, or null when the stored definition cannot be read - such a field is still listed. options is an enum's full allowed set and [] otherwise; default the value the field is prefilled with and null where it has none; format how date/datetime/time values are written and null otherwise; children the nested shape - a composite's named children, or the single anonymous shape a collection's items take, [] otherwise; and flagged the subset of a checklist answer's options that the form author declared as raising a flag, [] everywhere else. Trust flagged over the wording of an answer - it is the author's declaration, not a guess from words like "no" or "fail". Use it before reading tasks_report_rows, whose values are all display strings - the label says what a column means, the type how a value may be aggregated, format how to parse it, and options which values were possible at all.
tasks_report_rows
ChatGPTRead rows of a Fluix task report. Works for both process and form reports - pass type if you already know it (from tasks_list_reports), otherwise it's auto-detected. Cursor-paginated: limit is forwarded to upstream as the page size, so nextCursor lines up with the rows you received. Pass format: "csv" for compact text output. An image field's value is a media id; an image gallery's is several media ids joined by ", ". Pass them to tasks_media_thumbnails to get URLs you can fetch.
tasks_report_schema
ChatGPTList the columns of a task report. Works for both process and form reports. Returns systemFields (the report's predefined columns, such as Process Key, Workflow Name, Task Name, Assignee, each with a description) and formFields (custom form fields). The exact set depends on how the report is configured, so read it from the response instead of assuming. Names match the keys returned by tasks_report_rows, so that tool needs no schema call first - use this one only when the user asks what the columns are.
docs_dataset_rows
Claudedocs_dataset_schema
Claudedocs_list_datasets
Claudedocs_status_rows
Claudedocs_status_schema
Claudetasks_list_reports
Claudetasks_report_rows
Claudetasks_report_schema
Claude