Skip to main content
This is two things working together: Claude reads the PDF the way it reads any attached document, and the formbharo skill builds the agent from the plain-English question list Claude produces. The skill has no PDF reading of its own.

​
Prerequisites

  • The formbharo skill installed. See Agent Skills.
  • FORMBHARO_API_KEY and FORMBHARO_API_URL set in your environment. See Agent Skills.
1

Attach the PDF and ask Claude to read it

Attach the form PDF to your conversation and ask Claude to list every question or field on the form.
2

Review and correct the extracted list

Check the list against the form. Fix anything Claude misread, and tell it the language and tone you want, for example, “ask these in Hindi, keep it friendly.”
3

Ask Claude to build the agent

Ask Claude to build a FormBharo agent from the list. From here it is the same skill workflow as building from a typed description. See Agent Skills.
4

The skill creates and publishes the agent

The skill creates the agent as a draft (POST /api/v1/agents), then publishes it (PUT /api/v1/agents/{agent_id}).
5

Review the agent before real calls

Check the agent in the web app, or with GET /api/v1/agents/{agent_id}. PDFs are often subtly wrong on:
  • Field types. A “date of birth” field should become response_type: "date", not "string". See Agent Config Schema.
  • Option lists. Check that every choice on a single-select or multi-select field made it into options[], spelled the way you want it read aloud.
  • Validation the form implied but did not spell out. A phone number field should usually get a phone_number validation rule. See Agent Config Schema.
OCR can misread a scanned or handwritten PDF. Check the extracted question list against the source form before publishing, especially field types and which questions are required.

​
Next steps

Agent config schema

See the full shape of an agent’s JSON config

Agent Skills

Install the skill and see everything it can do