folio export

folio export datapackage   <SHEET> [--out <path>] [--stdout]
folio export claude-skills <SHEET> [--out <dir>]

datapackage

folio export datapackage <SHEET> [--out <path>] [--stdout]

Generates a Frictionless v1 Data Package descriptor (datapackage.json) that describes the sheet’s contract and points at records.jsonl. External readers — frictionless validate, datacontract-cli, dbt-style data-contract tooling — can pick the file up without bespoke glue.

Default

$ folio export datapackage ./customers
wrote datapackage.json (4 fields)

$ ls customers/
contract.yaml  datapackage.json  derivations/  provenance.jsonl  records.jsonl  scripts/

By default the descriptor is written to <SHEET>/datapackage.json.

Pipe to stdout

folio export datapackage ./customers --stdout > /tmp/customers.dp.json

Useful when you want to commit the descriptor outside the sheet, or pipe to frictionless validate directly.

What’s in the descriptor

{
  "name": "example-customers",
  "title": "customers",
  "version": "1.0.0",
  "description": "...",
  "resources": [
    {
      "name": "items",
      "path": "records.jsonl",
      "format": "jsonl",
      "profile": "tabular-data-resource",
      "schema": {
        "primaryKey": "id",
        "fields": [
          {"name": "id", "type": "string", "constraints": {"required": true}},
          {"name": "country_code", "type": "string", "folioDerived": true},
          ...
        ]
      }
    }
  ]
}

Folio field	Frictionless field
`id`	top-level `name`
`name`	top-level `title`
`version`	top-level `version`
`description`	top-level `description` (if set)
`schema[0].name`	`resources[0].name`
`properties[].logicalType`	`fields[].type` (mapped)
`properties[].required: true`	`fields[].constraints.required: true`
`properties[].x-derived: true`	`fields[].folioDerived: true` (custom extension key)

logicalType mapping (§14.2 of the design overview):

Folio	Frictionless
`string`	`string`
`integer`	`integer`
`number`	`number`
`boolean`	`boolean`
`date`	`date`
`timestamp`	`datetime`
`array`	`array`
`object`	`object`

What’s not in the descriptor

The contract’s derivations/ are not surfaced (Frictionless has no matching concept).
x-editable-by patterns are not surfaced (also Folio-specific).
Provenance is not surfaced (it’s not part of the data — it’s about the data).

The descriptor is therefore a lossy projection. It’s enough to make a Frictionless-aware tool read the rows; it isn’t enough to reconstruct the sheet from scratch.

Validate the export

pip install frictionless
folio export datapackage ./customers
frictionless validate ./customers/datapackage.json

frictionless is not a Folio dependency — install it explicitly when you need it.

claude-skills

folio export claude-skills <SHEET> [--out <dir>]

Emits one Claude Code-compatible SKILL.md directory per entry in <SHEET>/skills/. Default output dir is .claude/skills/ (relative to the current working directory). The Claude frontmatter (name, description) is mapped 1:1; Folio-specific metadata (audience, tools, allowed_actors, arguments) drops to a “Constraints” section in the body.

$ folio export claude-skills ./examples/onboarding --out /tmp/claude
wrote 1 SKILL.md file(s) under /tmp/claude/
  /tmp/claude/example-onboarding__advance-onboarding/SKILL.md

The output is derived, read-only: re-run the export whenever you change a skill. The sheet’s skills/ directory remains the source of truth.

This sub-command pairs with the folio-agent-skills npm package: the npm pack ships Folio’s product-level skills, while folio export claude-skills projects a sheet’s domain skills into the same Claude Code shape.

Future formats

folio export is namespaced by format (datapackage, claude-skills, <whatever>). The design doc §14.2 mentions OpenAPI surfaces and dbt models as future candidates; neither is implemented.