Solutions Installer

Install Snowflake Industry Solutions

Installs a pre-built industry solution from the sf-solutions repository into the user's Snowflake account. Each solution contains a manifest.json describing its metadata and a scripts/ directory with SQL setup/teardown files.

When to Use

• User wants to install a solution: $solutions-installer <solution-name> where <solution-name> is a directory name in the sf-solutions repository (e.g., manufacturing-predictive-maintenance)
• User wants to tear down / uninstall a solution
• User wants to set up a demo environment for an industry use case
• Do NOT use for editing the solutions catalog website

What This Skill Provides

• Clones or locates the sf-solutions repository
• Reads the solution's manifest.json to understand what will be created
• Executes install_scripts (setup SQL) against the user's Snowflake account
• Supports teardown via teardown_scripts
• Verifies the installation by checking created objects

Solution Directory Convention

Each solution follows this structure:

<solution-slug>/
├── manifest.json          # REQUIRED: solution metadata (see schema below)
├── README.md              # Solution overview, architecture, prerequisites
└── scripts/
    ├── setup.sql          # Main installation script
    └── teardown.sql       # Cleanup script (drops all created objects)

manifest.json Schema

{
  "name": "solution-slug",
  "display_name": "Human-Readable Solution Name",
  "version": "1.0.0",
  "industry": "Manufacturing",
  "source": "https://github.com/...",
  "license": "MIT",
  "database": "DATABASE_NAME",
  "schemas": ["SCHEMA_A", "SCHEMA_B"],
  "role": "ACCOUNTADMIN",
  "requires_warehouse": true,
  "install_scripts": ["scripts/setup.sql"],
  "teardown_scripts": ["scripts/teardown.sql"],
  "features": ["Cortex Analyst", "Semantic View", "Streamlit in Snowflake"]
}

Instructions

Step 0: Resolve the Solution

Actions:

1. Parse the solution slug from the user's prompt (e.g., manufacturing-predictive-maintenance)
2. If no slug is provided, list available solutions by scanning <repo>/**/manifest.json and present a table:

   | Slug | Name | Industry | Database | Role Required |
   

3. Ask the user to pick one if not specified

If the slug does not match any directory:

• Check if the repo has new solutions (git pull or re-scan)
• If still not found, inform the user and stop

Step 1: Fetch the Solution Repository

Actions:

1. Check if the sf-solutions repo exists locally. Search these paths in order:
   • ~/project/sf-solutions/
   • ./sf-solutions/
   • The current working directory (check for manifest.json files)
2. If not found locally, clone it:

   git clone https://github.com/Snowflake-Labs/sf-solutions.git /tmp/sf-solutions
   

3. Verify the solution directory exists: <repo>/<slug>/manifest.json

Output: Path to the solution directory.

Step 2: Read manifest.json and Present Installation Plan

Actions:

1. Read <slug>/manifest.json
2. Validate all required fields are present
3. Read <slug>/README.md for additional context (architecture, prerequisites)
4. List the install scripts that will be executed

⚠️ STOPPING POINT: Present the installation plan to the user:

Solution: <display_name> (v<version>)
Industry: <industry>
Source:   <source>

Will create:
  Database: <database>
  Schemas:  <schemas joined with ", ">
  Role required: <role>
  Features: <features joined with ", ">

Scripts to execute:
  1. <install_scripts[0]>
  2. <install_scripts[1]> (if any)

Proceed with installation?

Wait for user confirmation before proceeding.

Step 3: Execute Install Scripts

Actions:

1. For each script in install_scripts (in order): a. Read the SQL file b. Split into individual SQL statements (split on ;, respecting comments and string literals) c. Execute each statement sequentially using snowflake_sql_execute
2. For long-running statements (ML models, large data generation, SPCS compute pools):
   • Use timeout_seconds: 600
   • Inform the user that the statement may take several minutes
3. For statements that depend on RESULT_SCAN(LAST_QUERY_ID()):
   • Execute them immediately after the preceding statement
   • Do NOT batch these
4. After each major section, log progress:
   • "Created database "
   • "Created schema "
   • "Loaded rows into "

If a statement fails:

• "insufficient privileges": Show the required role from manifest.json, ask the user to switch roles or grant privileges
• "object already exists" (without CREATE OR REPLACE): Ask user whether to skip or drop-and-recreate
• Other errors: Show the error and the failing SQL, ask user for guidance

Output: Confirmation of each script completed.

Step 4: Verify Installation

Actions:

1. Run verification queries using the database and schemas from manifest.json:

   SELECT TABLE_SCHEMA, TABLE_NAME, ROW_COUNT
   FROM <database>.INFORMATION_SCHEMA.TABLES
   WHERE TABLE_SCHEMA IN (<schemas>)
   ORDER BY TABLE_SCHEMA, TABLE_NAME;
   

2. Check that all schemas from manifest.json exist
3. Report any schemas that are empty or missing

Output: Summary table of all created objects and row counts.

Step 5: Post-Install Summary

Present to the user:

Solution installed: <display_name> v<version>

Objects created:
  Database: <database>
  Schemas:  <list schemas with object counts>
  Tables:   <count> tables (<total rows> total rows)
  Views:    <count> views

Features enabled: <features>

Next steps:
  - See README: <slug>/README.md
  - Teardown: $solutions-installer teardown <slug>

Best Practices

• Always read manifest.json first — it is the source of truth for the solution
• Execute SQL statements one at a time to isolate errors
• Never modify the source SQL files — read and execute them as-is
• Log each step clearly so the user knows what was created
• For ML model training or large data loads, set timeout to 600s
• Always present the installation plan and wait for confirmation before executing

Stopping Points

• ✋ Step 2 — After presenting the installation plan, wait for user confirmation
• ✋ Step 3 — If privilege errors occur, stop and ask user for role
• ✋ Teardown — Before dropping any objects, confirm with user

Resume rule: Upon user approval, proceed directly to the next step without re-asking.

Output

• Fully provisioned Snowflake environment with the chosen solution
• Database, schemas, tables, views, and ML models ready to use
• Summary of all created objects with row counts

Examples

Example 1: Install manufacturing predictive maintenance

User: $solutions-installer manufacturing-predictive-maintenance Assistant:

1. Finds repo at ~/project/sf-solutions/
2. Reads manufacturing-predictive-maintenance/manifest.json
3. Presents plan: Database SNOWCORE_INDUSTRIES, schemas BRONZE/SILVER/GOLD, role ACCOUNTADMIN
4. User confirms → executes scripts/setup.sql statement by statement
5. Verifies all tables created, shows row counts
6. Shows summary with next steps

Example 2: List available solutions

User: $solutions-installer Assistant: Scans repo for manifest.json files, shows table of available solutions, asks user to pick one