Align all skills to best practices: ⚠️ stopping points, Output sections, updated templates

- Update CONTRIBUTING.md with skill best practices section covering
  required sections, stopping point format, Phase 0 consent pattern,
  line budget, and automation pipeline steps
- Rewrite both TEMPLATE/SKILL.md files with best-practices structure
- Add ⚠️ MANDATORY STOPPING POINT markers to all 17 action skills
  (replacing informal **STOP** and > **STOP.** variants)
- Add ## Output section to 8 skills that were missing it
- Add ## Stopping Points section to docker-dev-setup, drizzle-orm-setup
- Sync all agent rules (cursor, claude, windsurf, gemini)

Author: iamontheinet

.claude/rules/cortex-agents.md

@@ -75,7 +75,7 @@ Present the following briefing to the user:
 >
 > Ready to proceed?
 
-**STOP** — Wait for user approval before continuing.
+**⚠️ MANDATORY STOPPING POINT**: Wait for user approval before continuing.
 
 ---
 
@@ -344,6 +344,12 @@ GRANT USAGE ON AGENT {{DATABASE}}.{{SCHEMA}}.{{AGENT_NAME}} TO ROLE {{CONSUMER_R
 GRANT DATABASE ROLE SNOWFLAKE.CORTEX_USER TO ROLE {{ROLE}};
 ```
 
+## Output
+
+- A Cortex Agent object that routes natural language questions to the appropriate tool (Analyst, Search, or custom UDFs)
+- Supporting prerequisites: a semantic view for structured data and a Cortex Search Service for unstructured data
+- Tested agent responses demonstrating multi-tool orchestration across the user's data sources
+
 ## Agent Lifecycle
 
 ```sql

.claude/rules/cortex-ai-pipeline.md

@@ -70,7 +70,7 @@ Present the following briefing to the user:
 >
 > Ready to proceed?
 
-**STOP** — Wait for user approval before continuing.
+**⚠️ MANDATORY STOPPING POINT**: Wait for user approval before continuing.
 
 ---
 
@@ -87,7 +87,7 @@ Where should I set up the AI pipeline?
 - Use demo data or your own table?
 ```
 
-**STOP** — Wait for response.
+**⚠️ MANDATORY STOPPING POINT**: Wait for response.
 
 If user has their own table, ask which text column(s) to enrich and skip the demo data setup in Step 2.
 
@@ -244,6 +244,12 @@ LIMIT 10;
 
 The newer `AI_*` functions (AI_CLASSIFY, AI_SENTIMENT, etc.) are the recommended namespace. The legacy `SNOWFLAKE.CORTEX.*` functions (COMPLETE, SENTIMENT, SUMMARIZE, TRANSLATE, EXTRACT_ANSWER) still work but the `AI_*` versions have more features including image and multi-label support.
 
+## Output
+
+- An `ENRICHED_TICKETS` table (or enriched version of the user's table) with AI-generated columns: category, sentiment score, summary, and extracted fields
+- Cross-row aggregated insights via AI_AGG summarizing themes and recurring patterns
+- Optional custom AI_COMPLETE analysis and translations for multilingual data
+
 ## Access Control
 
 All Cortex AI Functions require the `SNOWFLAKE.CORTEX_USER` database role. Grant it with:

.claude/rules/cortex-mcp-server.md

@@ -77,7 +77,7 @@ Present the following briefing to the user:
 >
 > Ready to proceed?
 
-**STOP** — Wait for user approval before continuing.
+**⚠️ MANDATORY STOPPING POINT**: Wait for user approval before continuing.
 
 ---
 
@@ -102,7 +102,7 @@ Also:
 - MCP Server name: (e.g., MY_MCP_SERVER)
 ```
 
-**STOP** — Wait for response.
+**⚠️ MANDATORY STOPPING POINT**: Wait for response.
 
 ---
 
@@ -171,7 +171,7 @@ CREATE OR REPLACE MCP SERVER {{DATABASE}}.{{SCHEMA}}.{{MCP_SERVER_NAME}}
   $$;
 ```
 
-**STOP** — Show spec to user. Ask: "MCP server spec looks good?"
+**⚠️ MANDATORY STOPPING POINT**: Show spec to user. Ask: "MCP server spec looks good?"
 
 ---
 
@@ -249,7 +249,7 @@ Add to `claude_desktop_config.json`:
 }
 ```
 
-**STOP** — Help user connect their MCP client and test tool discovery.
+**⚠️ MANDATORY STOPPING POINT**: Help user connect their MCP client and test tool discovery.
 
 ---
 
@@ -323,3 +323,9 @@ DESCRIBE MCP SERVER {{DATABASE}}.{{SCHEMA}}.{{MCP_SERVER_NAME}};
 -- Drop server
 DROP MCP SERVER IF EXISTS {{DATABASE}}.{{SCHEMA}}.{{MCP_SERVER_NAME}};
 ```
+
+## Output
+
+- A Snowflake-managed MCP SERVER object exposing the user's chosen tools (Analyst, Search, Agent, SQL, custom UDFs)
+- RBAC grants configured for both server-level and tool-level access control
+- Client connection instructions with authentication setup (OAuth or PAT) ready for use in any MCP-compatible AI client

.claude/rules/cortex-search-rag.md

@@ -74,7 +74,7 @@ Present the following briefing to the user:
 >
 > Ready to proceed?
 
-**STOP** — Wait for user approval before continuing.
+**⚠️ MANDATORY STOPPING POINT**: Wait for user approval before continuing.
 
 ---
 
@@ -92,7 +92,7 @@ Where should I set up the RAG pipeline?
 - If your own table: which column contains the searchable text?
 ```
 
-**STOP** — Wait for response.
+**⚠️ MANDATORY STOPPING POINT**: Wait for response.
 
 If user has their own table, adapt the search service to use their text column. They must have change tracking enabled on the table (`ALTER TABLE ... SET CHANGE_TRACKING = TRUE`).
 
@@ -159,7 +159,7 @@ SELECT PARSE_JSON(
 )['results'] AS search_results;
 ```
 
-**STOP** — Show search results to user. Ask: "Search looks good? Ready to build the RAG pipeline?"
+**⚠️ MANDATORY STOPPING POINT**: Show search results to user. Ask: "Search looks good? Ready to build the RAG pipeline?"
 
 ---
 
@@ -228,7 +228,7 @@ Test with several questions to check quality:
 -- 3. "What shipping options are available?"
 ```
 
-**STOP** — Show results. Ask if the user wants to:
+**⚠️ MANDATORY STOPPING POINT**: Show results. Ask if the user wants to:
 1. Adjust the number of retrieved documents (limit)
 2. Add attribute filtering (e.g., only search specific categories)
 3. Try a different LLM model
@@ -314,3 +314,9 @@ SHOW CORTEX SEARCH SERVICES IN SCHEMA {{DATABASE}}.{{SCHEMA}};
 -- Manually refresh (usually not needed — auto-refresh handles it)
 ALTER CORTEX SEARCH SERVICE {{DATABASE}}.{{SCHEMA}}.KB_SEARCH_SERVICE RESUME;
 ```
+
+## Output
+
+- A Cortex Search Service with hybrid keyword + vector search over the user's data
+- A tested RAG query template that retrieves relevant documents and generates grounded answers via AI_COMPLETE
+- Working SQL patterns for filtered search, attribute-based narrowing, and Python API integration

.claude/rules/data-product-sharing.md

@@ -99,7 +99,7 @@ FROM raw_table
 WHERE is_public = TRUE;  -- row-level filtering
 ```
 
-**STOP**: Confirm which objects to share before proceeding.
+**⚠️ MANDATORY STOPPING POINT**: Confirm which objects to share before proceeding.
 
 ### Step 3: Create Share or Listing
 
@@ -199,6 +199,12 @@ When sharing with accounts in different regions, Snowflake automatically replica
 
 ---
 
+## Output
+
+- A configured Snowflake share or listing granting consumers live, read-only access to the specified tables/views
+- Consumer-side SQL to mount the shared database and verify data access
+- End-to-end verification confirming provider and consumer can query shared objects
+
 ## Stopping Points
 
 - **Step 2**: Confirm which objects to share

.claude/rules/docker-dev-setup.md

@@ -43,6 +43,13 @@ docker-dev-setup/
     └── devcontainer.json         # VS Code Dev Container config
 ```
 
+## Stopping Points
+
+- Phase 1: User confirms stack, entry point, and services before Dockerfile creation
+- Phase 2: User reviews Dockerfile and verifies it builds
+- Phase 3: User confirms Compose setup and verifies it starts
+- Phase 4: User confirms Dev Container config and tests it
+
 ## Workflow
 
 Follow these phases in order. **Stop after each phase** to confirm with the user before continuing.
@@ -78,7 +85,7 @@ Follow these phases in order. **Stop after each phase** to confirm with the user
    - Go → `Dockerfile.go`
    - Other → Use `references/dockerfile-patterns.md` to build from scratch
 
-> **STOP.** Confirm the stack, entry point, and which services the user needs.
+**⚠️ MANDATORY STOPPING POINT**: Confirm the stack, entry point, and which services the user needs.
 
 ---
 
@@ -106,7 +113,7 @@ Follow these phases in order. **Stop after each phase** to confirm with the user
    - Use `npm ci` not `npm install` in Docker builds
    - For Go: `CGO_ENABLED=0` for static binaries that run on `scratch`
 
-> **STOP.** Review the Dockerfile with the user. Verify it builds: `docker build -t app .`
+**⚠️ MANDATORY STOPPING POINT**: Review the Dockerfile with the user. Verify it builds: `docker build -t app .`
 
 ---
 
@@ -153,7 +160,7 @@ Follow these phases in order. **Stop after each phase** to confirm with the user
    - Set environment variables via `environment:` block, not `.env` files in the image
    - Use `restart: unless-stopped` for infrastructure services
 
-> **STOP.** Confirm the Compose setup. Verify it starts: `docker compose up`
+**⚠️ MANDATORY STOPPING POINT**: Confirm the Compose setup. Verify it starts: `docker compose up`
 
 ---
 
@@ -182,7 +189,7 @@ Follow these phases in order. **Stop after each phase** to confirm with the user
    }
    ```
 
-> **STOP.** Confirm the Dev Container config. Test by reopening in container.
+**⚠️ MANDATORY STOPPING POINT**: Confirm the Dev Container config. Test by reopening in container.
 
 ---
 

.claude/rules/drizzle-orm-setup.md

@@ -40,6 +40,13 @@ drizzle-orm-setup/
     └── db.ts                 # Database client setup
 ```
 
+## Stopping Points
+
+- Phase 1: User confirms dialect, driver, and schema organization
+- Phase 2: User reviews schema structure, column types, and relations
+- Phase 3: User confirms client setup and DATABASE_URL configuration
+- Phase 4: User confirms migrations ran successfully
+
 ## Workflow
 
 Follow these four phases in order. **Stop after each phase** to confirm with the user before continuing.
@@ -72,7 +79,7 @@ Follow these four phases in order. **Stop after each phase** to confirm with the
    - Single `schema.ts` file (small projects)
    - `schema/` directory with one file per table (recommended for >3 tables)
 
-> **STOP.** Confirm dialect, driver, and schema organization with the user.
+**⚠️ MANDATORY STOPPING POINT**: Confirm dialect, driver, and schema organization with the user.
 
 ---
 
@@ -110,7 +117,7 @@ Follow these four phases in order. **Stop after each phase** to confirm with the
 - Always define BOTH sides of a relation (e.g., `users → many(posts)` AND `posts → one(users)`).
 - Use `$onUpdateFn(() => new Date())` for `updatedAt` columns, NOT database-level triggers.
 
-> **STOP.** Review the schema with the user. Confirm table structure, column types, and relations.
+**⚠️ MANDATORY STOPPING POINT**: Review the schema with the user. Confirm table structure, column types, and relations.
 
 ---
 
@@ -172,7 +179,7 @@ Follow these four phases in order. **Stop after each phase** to confirm with the
    }
    ```
 
-> **STOP.** Confirm the client setup and config. Verify the DATABASE_URL is available (env var, .env file, etc.).
+**⚠️ MANDATORY STOPPING POINT**: Confirm the client setup and config. Verify the DATABASE_URL is available (env var, .env file, etc.).
 
 ---
 
@@ -214,7 +221,7 @@ Follow these four phases in order. **Stop after each phase** to confirm with the
 - Migration files are append-only. Don't edit generated SQL files.
 - The `./drizzle/meta/` directory tracks migration state. Commit it to version control.
 
-> **STOP.** Confirm migrations ran successfully. Check for any errors.
+**⚠️ MANDATORY STOPPING POINT**: Confirm migrations ran successfully. Check for any errors.
 
 ---
 

.claude/rules/dynamic-tables-pipeline.md

@@ -69,7 +69,7 @@ Present the following briefing to the user:
 
 Ask the user: **"Shall I proceed with Step 1 (Gather Target Details)?"**
 
-Do NOT proceed until the user confirms.
+**⚠️ MANDATORY STOPPING POINT**: Do NOT proceed until the user confirms.
 
 ---
 
@@ -86,7 +86,7 @@ Propose defaults and let the user override:
 | **Warehouse** | `COMPUTE_WH` | Any X-Small is fine |
 | **Role** | `ACCOUNTADMIN` | Must have CREATE DYNAMIC TABLE privilege |
 
-Confirm with the user before proceeding.
+**⚠️ MANDATORY STOPPING POINT**: Confirm with the user before proceeding.
 
 ---
 

.claude/rules/iceberg-tables.md

@@ -74,7 +74,7 @@ Present the following briefing to the user:
 
 Ask the user: **"Shall I proceed with Step 1 (Choose Catalog Type)?"**
 
-Do NOT proceed until the user confirms.
+**⚠️ MANDATORY STOPPING POINT**: Do NOT proceed until the user confirms.
 
 ---
 
@@ -92,7 +92,7 @@ Ask the user which approach to use:
 
 Default recommendation: **Snowflake-managed** unless the user has a specific external catalog requirement.
 
-Confirm the user's choice before proceeding.
+**⚠️ MANDATORY STOPPING POINT**: Confirm the user's choice before proceeding.
 
 ---
 
@@ -114,7 +114,7 @@ The user must update their IAM role trust policy with the values from:
 - `STORAGE_AWS_IAM_USER_ARN`
 - `STORAGE_AWS_EXTERNAL_ID`
 
-Do NOT proceed until the user confirms the trust policy is updated.
+**⚠️ MANDATORY STOPPING POINT**: Do NOT proceed until the user confirms the trust policy is updated.
 
 **Alternative for demo/testing:** If the user just wants to try Iceberg tables and already has an external volume set at the account or database level, skip this step.
 
@@ -163,7 +163,7 @@ Verify:
 SHOW ICEBERG TABLES IN SCHEMA {{DATABASE}}.{{SCHEMA}};
 ```
 
-Confirm with the user that the table was created.
+**⚠️ MANDATORY STOPPING POINT**: Confirm with the user that the table was created.
 
 ---
 

.claude/rules/ml-model-registry.md

@@ -73,7 +73,7 @@ Present the following briefing to the user:
 >
 > Ready to proceed?
 
-**STOP** — Wait for user approval before continuing.
+**⚠️ MANDATORY STOPPING POINT**: Wait for user approval before continuing.
 
 ---
 
@@ -92,7 +92,7 @@ Let me set up the ML pipeline. A few questions:
 - Use demo data or your own dataset?
 ```
 
-**STOP** — Wait for response.
+**⚠️ MANDATORY STOPPING POINT**: Wait for response.
 
 If the user has their own data, ask about:
 - Target column (what to predict)
@@ -156,7 +156,7 @@ mv = reg.log_model(
 predictions = mv.run(X_test)
 ```
 
-**STOP** — Show model metrics to user. Ask if they're satisfied or want to retrain with different parameters.
+**⚠️ MANDATORY STOPPING POINT**: Show model metrics to user. Ask if they're satisfied or want to retrain with different parameters.
 
 ---
 
@@ -205,7 +205,7 @@ Your model is registered. How would you like to run inference?
 3. Both — Register for warehouse now, deploy service later
 ```
 
-**STOP** — Wait for response.
+**⚠️ MANDATORY STOPPING POINT**: Wait for response.
 
 ### Warehouse Inference
 
@@ -240,6 +240,12 @@ mv.create_service(
 
 ---
 
+## Output
+
+- A trained ML model registered in Snowflake Model Registry with version tracking and logged metrics
+- Warehouse-based SQL inference or an SPCS REST endpoint for real-time predictions
+- Access control grants so analysts can run inference without managing model artifacts
+
 ## Supported Frameworks
 
 | Framework | log_model() Support | Notes |