name: omni-to-snowflake-semantic-view description: "Convert an Omni Analytics topic into a Snowflake Semantic View YAML definition. Use this skill whenever someone wants to export Omni metrics to Snowflake, create a Semantic View from an Omni topic, harden BI metrics into the warehouse, or bridge Omni's semantic layer with Snowflake Cortex Analyst."

Omni → Snowflake Semantic View

Converts an Omni topic into a Snowflake Semantic View YAML definition by first exploring the Omni model via API, then translating its definitions into the Snowflake Semantic View format.

Runtime Environment Detection

Before starting, determine which environment you are running in — this controls how SQL is executed against Snowflake in Step 7.

How to detect

Check for the presence of the built-in Snowflake SQL execution capability:

• Cortex Code (Snowflake Notebooks / Cortex Analyst): Native SQL execution is available. No extra tooling required.
• Claude Code, Cursor, or any external IDE/agent: No native Snowflake connection. You must use one of the CLI options below.

Detection heuristic

If the agent can execute a SQL statement like `SELECT CURRENT_USER()` directly without any shell command → you are in Cortex Code.
Otherwise → you are in an external environment.

If unsure, ask the user: "Are you running this inside Snowflake (Cortex Code / Notebooks), or in an external tool like Claude Code or Cursor?"

External environment: Snowflake connectivity options

For Claude Code, Cursor, or any terminal-based agent, choose one of the following. Check availability in order:

Option A — Snowflake CLI (snow) ✅ Recommended

# Check if available
command -v snow

# Configure a connection (first time)
snow connection add

# Execute SQL
snow sql -q "SELECT CURRENT_USER();" --connection <connection_name>

Set a default connection to avoid repeating --connection on every call:

snow connection set-default <connection_name>

Option B — SnowSQL (classic CLI)

# Check if available
command -v snowsql

# Execute SQL
snowsql -a <account> -u <user> -q "SELECT CURRENT_USER();"

Option C — Python (snowflake-connector-python)

pip install snowflake-connector-python

import snowflake.connector
conn = snowflake.connector.connect(
    account="<account>",
    user="<user>",
    password="<password>",   # or use key-pair / SSO
    warehouse="<warehouse>",
    database="<database>",
    schema="<schema>",
)
conn.cursor().execute("<SQL here>")

✋ STOP — Confirm the Snowflake connection method with the user before proceeding to Step 7. Record which method is being used so Step 7 generates the correct execution command.

Workflow

Step 1 — Gather Requirements

Ask the user:

1. Which Topic do they want to convert?
2. What should the Semantic View be named?
3. Where should it be created in Snowflake (database and schema)?
4. Which role should be granted access to the Semantic View?

⚠️ STOP — Confirm all four answers before proceeding.

Step 2 — Explore the Omni Model

2a. Find the model ID

omni models list --modelkind SHARED

Identify the Shared Model and note its id. Always prefer the Shared Model over Schema or Workbook models.

2b. Fetch the topic file

omni models yaml-get <modelId> --filename <topic_name>.topic

2c. Fetch the relationships file

omni models yaml-get <modelId> --filename relationships

2d. Fetch each view file referenced in the topic

For every view in base_view and joins, fetch its YAML:

omni models yaml-get <modelId> --filename <view_name>.view

If a view is prefixed with omni_dbt_, fetch the file that also starts with omni_dbt_ (e.g. omni_dbt_ecomm__order_items.view).

Step 3 — Identify Tables and Joins

Mapping view names to Snowflake tables

The base_view in the topic file is the primary table. Convert view names to Snowflake table references:

The __ separator maps to schema (left) and table name (right). If the schema does not exist in Snowflake, skip that table entirely.

Reading the join hierarchy

The joins parameter in the topic uses indentation to define the join chain — a table indented beneath another joins into its parent:

joins:
  user_order_facts: {}        # skip — this is a derived CTE, not a physical table
  ecomm__users: {}            # joins to base_view (ORDER_ITEMS)
  ecomm__inventory_items:     # joins to base_view (ORDER_ITEMS)
    ecomm__products:          # joins to INVENTORY_ITEMS
      demo__product_images: {}           # joins to PRODUCTS
      ecomm__distribution_centers: {}   # joins to PRODUCTS

Skip any view that is a derived table (CTE defined in SQL in Omni). These have no physical Snowflake table to reference.

Primary keys

In each view file, find the dimension with primary_key: true — this becomes the unique: true dimension in the Semantic View.

✋ STOP — Confirm the table list with the user before continuing.

Step 4 — Resolve the Field List

The topic's fields parameter controls which fields from the views are included in the Semantic View.

Field targeting rules

How to apply exclusions correctly

Exclusions (prefixed with -) must be applied after all inclusions are resolved. The process is:

1. Start with an empty inclusion set
2. Process each entry in fields in order:
   • If it is an inclusion rule (view.*, view.field, tag:x) → add matching fields to the set
   • If it is an exclusion rule (-view.field) → remove that field from the set, even if it was added by a wildcard
3. The final set is the complete list of fields to include in the Semantic View

Example:

fields:
  - ecomm__order_items.*          # include all order_items fields
  - ecomm__users.country          # include this one users field
  - -ecomm__order_items.cost      # remove cost — excluded even though * was used above
  - -ecomm__order_items.raw_json  # remove raw_json — same reason

Result: all order_items fields except cost and raw_json, plus users.country.

⚠️ Critical: A - exclusion always wins. Never include a field that has been explicitly excluded, regardless of what wildcard included it.

Step 5 — Build Relationships

Using the join hierarchy from Step 3 and the relationships.yaml fetched in Step 2c, map each join to a Snowflake Semantic View relationship.

Each entry in relationships.yaml looks like:

- join_from_view: ecomm__order_items
  join_to_view: ecomm__inventory_items
  join_type: always_left
  on_sql: ${ecomm__order_items.inventory_item_id} = ${ecomm__inventory_items.id}
  relationship_type: assumed_many_to_one

The on_sql field tells you the join columns. Extract the column names to populate relationship_columns in the output.

Available relationship parameters:

Step 6 — Map Dimensions and Measures

For each view in the resolved field list, translate its Omni field definitions into Semantic View entries.

⚠️ Only translate fields that survived the Step 4 inclusion/exclusion resolution. Do not add fields that were excluded.

Dimensions → dimensions or time_dimensions

The field name becomes the dimension name unless a label is defined. Carry description and synonyms directly.

Standard dimension:

# Omni view YAML
city:
  sql: '"CITY"'
  label: City
  description: Customer's city

# Semantic View output
- name: city
  expr: CITY
  description: Customer's city
  data_type: TEXT

Date/timestamp dimension → use time_dimensions:

# Omni
created_at:
  sql: '"CREATED_AT"'
  type: time
  label: Created At

# Semantic View output
time_dimensions:
  - name: created_at
    expr: CREATED_AT
    data_type: TIMESTAMP

Group dimension → translate to a CASE WHEN expression:

# Omni
device_type_groups:
  sql: ${device_type}
  label: Device Type Groups
  groups:
    - filter:
        is: [ mobile, tablet ]
      name: Handheld
    - filter:
        is: desktop
      name: Desktop
  else: Other

-- expr value
CASE
  WHEN "DEVICE_TYPE" IN ('mobile', 'tablet') THEN 'Handheld'
  WHEN "DEVICE_TYPE" = 'desktop' THEN 'Desktop'
  ELSE 'Other'
END

Bin dimension → translate to a CASE WHEN range expression:

# Omni
age_bin:
  sql: ${age}
  bin_boundaries: [ 18, 35, 50, 65 ]

-- expr value
CASE
  WHEN "AGE" < 18 THEN 'below 18'
  WHEN "AGE" >= 18 AND "AGE" < 35 THEN '>= 18 and < 35'
  WHEN "AGE" >= 35 AND "AGE" < 50 THEN '>= 35 and < 50'
  WHEN "AGE" >= 50 AND "AGE" < 65 THEN '>= 50 and < 65'
  WHEN "AGE" >= 65 THEN '65 and above'
  ELSE NULL
END

Duration dimension → translate to a TIMESTAMPDIFF expression:

# Omni
fulfillment_days:
  duration:
    sql_start: ${created_at[date]}
    sql_end: ${delivered_at[date]}
    intervals: [ days ]

-- expr value (days)
CASE
  WHEN TIMESTAMPADD(DAY, 1 * TIMESTAMPDIFF(DAY, DATE_TRUNC('DAY', "CREATED_AT"), DATE_TRUNC('DAY', "DELIVERED_AT")), DATE_TRUNC('DAY', "CREATED_AT"))
       <= DATE_TRUNC('DAY', "DELIVERED_AT")
  THEN TIMESTAMPDIFF(DAY, DATE_TRUNC('DAY', "CREATED_AT"), DATE_TRUNC('DAY', "DELIVERED_AT"))
  ELSE TIMESTAMPDIFF(DAY, DATE_TRUNC('DAY', "CREATED_AT"), DATE_TRUNC('DAY', "DELIVERED_AT")) - 1
END

Boolean dimension → becomes a named filter (not a dimension):

# Omni
is_returned:
  sql: '"IS_RETURNED"'
  description: Whether the item was returned

completed_orders:
  sql: ${status} = 'Complete'
  label: Completed Orders

# Semantic View output — goes in filters, not dimensions
filters:
  - name: is_returned
    expr: IS_RETURNED
    description: Whether the item was returned
  - name: completed_orders
    expr: STATUS = 'Complete'
    description: Completed Orders

Measures → metrics

The sql field references the source column and aggregate_type defines the aggregation.

Standard measure:

# Omni
total_sale_price:
  sql: ${sale_price}
  aggregate_type: sum
  label: Total Sale Price
  description: Total revenue of orders
  synonyms: [ Total Revenue, Total Receipts ]

# Semantic View output
metrics:
  - name: total_sale_price
    expr: COALESCE(SUM("SALE_PRICE"), 0)
    description: Total revenue of orders
    synonyms: [ Total Revenue, Total Receipts ]

Derived measure (no aggregate_type) — references other measures:

# Omni
gross_margin:
  sql: ${total_sale_price} - ${total_cost}
  label: Gross Margin

# Semantic View output — top-level derived metric
metrics:
  - name: gross_margin
    expr: total_sale_price - total_cost

Filtered measure → wraps in CASE WHEN:

# Omni
california_revenue:
  sql: ${sale_price}
  aggregate_type: sum
  filters:
    users.state:
      is: California

-- expr value
COALESCE(SUM(CASE WHEN "users"."STATE" = 'California' THEN "SALE_PRICE" ELSE NULL END), 0)

Array and boolean filter variants:

# is: [New York, New Jersey]  →  "STATE" IN ('New York', 'New Jersey')
# is: true                    →  field IS TRUE

Filter-Only Fields (Templated Filters)

Filter-only fields in Omni are parameterized inputs defined in a filters: block in a view YAML. They don't map to a real column — they inject dynamic values into other fields' SQL at query time via Mustache templates. Snowflake Semantic Views have no equivalent runtime parameter system, so these must be resolved to static SQL during conversion.

Example filter-only field definition:

# In a view file — under a top-level `filters:` key
filters:
  status_filter:
    type: string
    default_filter:
      is: "Complete"
  order_date_filter:
    type: timestamp

Two reference patterns to detect in dimension/measure SQL:

Pattern 1 — Block (conditional) syntax:

{{# order_items.status_filter.filter }} order_items.status {{/ order_items.status_filter.filter }}

The block is conditionally active when the user applies a filter. A {{^ }} block is the fallback when no filter is applied:

{{# order_items.status_filter.filter }} order_items.status {{/ order_items.status_filter.filter }}
{{^ order_items.status_filter.filter }} 1=1 {{/ order_items.status_filter.filter }}

Pattern 2 — Value (direct injection) syntax:

-- Direct value substitution
DATE_TRUNC('day', created_at) >= '{{filters.order_items.order_date_filter.value}}'

-- Range variants for date filters
created_at BETWEEN '{{filters.order_items.order_date_filter.range_start}}' AND '{{filters.order_items.order_date_filter.range_end}}'

Scan all dimension and measure SQL for both patterns using these signatures:

• {{# <view>.<field>.filter }} — block open
• {{^ <view>.<field>.filter }} — block fallback open
• {{filters.<view>.<field>.value}} — direct value
• {{filters.<view>.<field>.range_start}} / {{filters.<view>.<field>.range_end}} — date ranges

Resolution logic — apply in order for each reference found:

Step 1 — Inline {{^ }} fallback (block pattern only):

If the SQL contains a {{^ view.field.filter }} fallback block, use the fallback expression as the static SQL — it's what executes when no filter is applied.

-- Input
{{# order_items.status_filter.filter }} order_items.status {{/ order_items.status_filter.filter }}
{{^ order_items.status_filter.filter }} 1=1 {{/ order_items.status_filter.filter }}

-- Resolved static SQL
1=1

Step 2 — default_filter on the filter field definition:

Find the referenced filter-only field in the view YAML and check for a default_filter property. Render it as a static SQL condition:

For value syntax, substitute the rendered scalar value directly into the SQL string:

-- Input
DATE_TRUNC('day', created_at) >= '{{filters.order_items.order_date_filter.value}}'

-- default_filter: is: "2024-01-01"
-- Resolved
DATE_TRUNC('day', created_at) >= '2024-01-01'

For range syntax, use range_start / range_end from the default_filter if defined, otherwise warn the user.

Step 3 — No fallback and no default_filter — warn the user:

⚠️ Field <view>.<field> references filter-only field <view>.<filter_field> but has no default_filter and no inline {{^ }} fallback. It cannot be statically resolved for a Snowflake Semantic View. Options:

1. Skip this field from the output entirely.
2. Ask the user what static value or condition to hardcode.
3. Use 1=1 (always-true no-op) to preserve the field shape without the filter effect.

Summary decision tree:

Does the SQL reference a filter-only field?
  (via {{# view.field.filter }} OR {{filters.view.field.value}} etc.)
  │
  ├── Block pattern → Does SQL have {{^ view.field.filter }} fallback?
  │     ├── Yes → Use fallback expression as static SQL
  │     └── No  → Go to default_filter check ↓
  │
  ├── Value pattern → Go to default_filter check ↓
  │
  └── default_filter check:
        ├── Yes → Render default_filter as static SQL / value substitution
        └── No  → Warn user, ask how to proceed

⚠️ Filter-only fields themselves are never emitted as dimensions, metrics, or filters in the Semantic View output — they exist only to parameterize other fields' SQL.

AI Context → module_custom_instructions

If the topic has an ai_context parameter, include it as:

module_custom_instructions:
  question_categorization: <ai_context value>

Sample Queries → verified_queries

Convert each entry under sample_queries into a SQL statement for verified_queries. Use the prompt as the question and the description as context for writing the SQL.

✋ STOP — Review all dimensions, measures, and relationships with the user before generating the final output.

Step 7 — Generate and Execute the Semantic View

Output YAML structure

name: <name>
description: <string>

tables:
  - name: <name>
    description: <string>
    base_table:
      database: <database>
      schema: <schema>
      table: <table name>

    dimensions:
      - name: <name>
        synonyms: [ <string>, ... ]
        description: <string>
        expr: <SQL expression>
        data_type: <data type>
        unique: <boolean>

    time_dimensions:
      - name: <name>
        synonyms: [ <string>, ... ]
        description: <string>
        expr: <SQL expression>
        data_type: <data type>

    facts:
      - name: <name>
        synonyms: [ <string>, ... ]
        description: <string>
        expr: <SQL expression>
        data_type: <data type>

    metrics:
      - name: <name>
        synonyms: [ <string>, ... ]
        description: <string>
        expr: <SQL expression>

    filters:
      - name: <name>
        synonyms: [ <string>, ... ]
        description: <string>
        expr: <SQL expression>

relationships:
  - name: <string>
    left_table: <table>
    right_table: <table>
    relationship_columns:
      - left_column: <column>
        right_column: <column>

metrics:
  - name: <name>
    synonyms: [ <string>, ... ]
    description: <string>
    expr: <SQL expression>

verified_queries:
  - name: <string>
    question: <string>
    sql: <string>
    use_as_onboarding_question: <boolean>

Valid top-level keys only: name, description, tables, relationships, metrics, verified_queries, module_custom_instructions

Create the Semantic View

The SQL is the same regardless of environment. The difference is how it is executed.

SQL to run:

CALL SYSTEM$CREATE_SEMANTIC_VIEW_FROM_YAML('<database>.<schema>', $$
<yaml content here>
$$);

Execution by environment:

Tip for CLI environments: If the YAML is long, write it to a temporary .sql file first and execute the file rather than passing it inline — this avoids shell escaping issues with the $$ dollar-quoting.

# snow CLI example with a file
snow sql -f /tmp/create_semantic_view.sql

Grant access

GRANT SELECT ON SEMANTIC VIEW <database>.<schema>.<name> TO ROLE <role>;

Execute this the same way as the CREATE call above, using whichever connection method was established during the Runtime Environment Detection step.

Reference

• Snowflake Semantic View YAML Spec
• SYSTEM$CREATE_SEMANTIC_VIEW_FROM_YAML

Prerequisites

# Verify the Omni CLI is installed — if not, ask the user to install it
# See: https://github.com/exploreomni/cli#readme
command -v omni >/dev/null || echo "ERROR: Omni CLI is not installed."

# Show available profiles and select the appropriate one
omni config show
# If multiple profiles exist, ask the user which to use, then switch:
omni config use <profile-name>

# Confirm the active profile is authenticated and inspect your permissions:
omni whoami whoami

Auth: a profile authenticates with an API key or OAuth. If whoami (or any call) returns 401, hand off — ask the user to run ! omni config login <profile> (OAuth 2.1 browser flow; it blocks ~2 min on the browser). Don't run config login yourself in a headless/CI session (no browser → timeout); on a local interactive machine you may. See the omni-api-conventions rule for profile setup (omni config init --auth oauth) and discovering request-body shapes with --schema.

Tip: Use -o json to force structured output for programmatic parsing, or -o human for readable tables. The default is auto (human in a TTY, JSON when piped).