discover-source

Discover Source Workflow

You are identifying the user's data source and registering it with Elliot. The goal is to land at a state where elliot_discover_source has succeeded and the next skill (build-connector) can draft tools against the discovered schema.

Step 0 — Set context if not already set

If elliot_get_session_state shows no product_context, call:

• elliot_set_context with name (slug-style, e.g. acme-crm) and a one-sentence description.

If the user hasn't told you these yet, ask in one combined question. Don't ask two separate questions.

Step 1 — Identify the source type

Ask the user (one question, listing the options):

"What kind of source is this? REST API, Postgres, MySQL, or a file (CSV / JSON / JSONL)?"

Map their answer to a source_type:

User says	source_type arg
"REST", "API", URL ending in /api	rest (aliases: api, http)
"Postgres", "PG", postgres://	postgres (alias: db, postgresql)
"MySQL", mysql://	mysql
"CSV", "JSON", "JSONL", a .csv / .json file	file (aliases: csv, json)

If the user has an OpenAPI spec or Postman collection, this is the wrong skill — switch to onboard-product and call elliot_import_api_collection instead. elliot_discover_source does not ingest OpenAPI/Postman.

If the user wants to wrap a SQLite database, Elliot does not support it as a source today; ask them to export the relevant tables to CSV or JSON first.

Step 2 — Collect the right config fields for that source type

elliot_discover_source takes exactly three arguments:

elliot_discover_source(source_type=<one of above>, config=<dict>, name=<table name>)

Everything that varies by source type goes inside the config dict. Ask in one batched message — never field-by-field, it wastes turns.

REST

{
  "url": "https://api.acme.com/v1/customers",
  "method": "GET",
  "auth": {
    "type": "bearer",
    "secret_key": "{{ env:ACME_API_TOKEN }}"
  },
  "data_path": "data",
  "pagination": {"strategy": "offset", "page_size": 50, "max_pages": 10}
}