schema-discovery（模式发现）

Schema Discovery

Purpose

Fetch and interpret the profile schema to find fields that match natural language descriptions. Used by other skills (especially audience-builder) to map user intent to actual schema fields.

Environment

Requires authenticated API access. See ../references/auth.md for credential resolution.

Inputs

• Natural language description of the data concepts to find (e.g., "country", "purchase history", "email engagement")
• Table name (default: user)

API Endpoints

List All Fields

curl -s "${LYTICS_API_URL:-https://api.lytics.io}/v2/schema/${TABLE:-user}/field" \
  -H "Authorization: ${LYTICS_API_TOKEN}"

Returns all fields with their types, descriptions, and merge operations.

Lightweight Field Names

curl -s "${LYTICS_API_URL:-https://api.lytics.io}/v2/schema/${TABLE:-user}/fieldnames" \
  -H "Authorization: ${LYTICS_API_TOKEN}"

Returns just field names -- use for quick inventory.

Field Info with Value Distributions

curl -s "${LYTICS_API_URL:-https://api.lytics.io}/api/schema/${TABLE:-user}/fieldinfo" \
  -H "Authorization: ${LYTICS_API_TOKEN}"

Returns field info including value distributions and sample data.

Field Value Suggestions

curl -s "${LYTICS_API_URL:-https://api.lytics.io}/api/schema/${TABLE:-user}/fieldsuggest/${FIELD}?q=${QUERY}" \
  -H "Authorization: ${LYTICS_API_TOKEN}"

Returns suggested values for a specific field matching the query substring. The q parameter is required -- it performs case-insensitive substring matching against field values.

Identity Configuration

curl -s "${LYTICS_API_URL:-https://api.lytics.io}/v2/schema/${TABLE:-user}/idconfig" \
  -H "Authorization: ${LYTICS_API_TOKEN}"

Available Streams

curl -s "${LYTICS_API_URL:-https://api.lytics.io}/v2/stream/names" \
  -H "Authorization: ${LYTICS_API_TOKEN}"

Behavior

Step 1: Fetch Field Inventory

Call the field names endpoint to get the complete list of available fields.

Step 2: Score Candidate Fields

For each concept in the user's description, score candidate fields by:

1. Name match (highest priority): exact or substring match on field name
2. Type compatibility: field type supports the intended operation (see ../references/field-types.md)
3. Description match: field's ShortDesc or LongDesc matches

Step 3: Confirm with Value Distributions

For the top 2-3 candidates per concept, fetch field value suggestions to confirm:

• The field contains the expected kind of data
• Sample values match what the user described (e.g., "US" values in a country field)

Step 4: Return Field Mapping

Return a mapping of:

• Concept -> chosen field name, field type, confidence level, sample values
• Any ambiguities that need user input

Field Matching Heuristics

Common natural language to field name patterns:

• "country" -> country, geo_country, country_code
• "email" -> email, _e
• "purchased/bought" -> products_purchased, purchase_categories, orders
• "visited/browsed" -> urls_visited, pages_viewed, visit_count
• "signed up/registered" -> created, signup_date, joined
• "clicked" -> click_count, clicks
• "score" -> scores.*, engagement_score

Error Handling

• If no fields match a concept, report that clearly and suggest the user rephrase or check available fields
• If multiple equally strong candidates exist, present them for user selection
• If the table doesn't exist, report and suggest user as default

Dependencies

• References: ../references/auth.md, ../references/api-client.md, ../references/field-types.md