Runwal Snowflake Intelligence Layer — Bot Capability Matrix (2026-05-06)

Runwal Snowflake Intelligence Layer — Bot Capability Matrix

Account: RUNWALENTERPRISES-DLH (locator VM65651, AZURE_CENTRALINDIA) Canonical agent home: SNOWFLAKE_INTELLIGENCE.AGENTS schema (auto-discovered by Snowflake Intelligence UI) Shared utility: SNOWFLAKE_INTELLIGENCE.AGENTS.SEND_EMAIL(VARCHAR, VARCHAR, VARCHAR) — stored procedure (Python 3.12), USAGE granted to RNWL_DATA_ENGINEER

Live Production Bots

BotLocationOwnerSemantic ViewToolsStatus
RUNWAL_BOT (display: RUNWAL_SALES_BOT)SNOWFLAKE_INTELLIGENCE.AGENTSACCOUNTADMINRNWL_PROD.RNWL_SALES.RNWL_SALES_REPORTINGRUNWAL_SALES_ANALYST + Send_Email✅ Live, full instructions, 6 sample questions
RUNWAL_CRM_BOTSNOWFLAKE_INTELLIGENCE.AGENTSACCOUNTADMINRNWL_PROD.RNWL_CRM.RNWL_CRM_REPORTINGRNWL_CRM_ANALYST + Send_Email (added 2026-05-06)✅ Live, instructions present, no sample questions
RUNWAL_CONSTRUCTION_BOTRNWL_PROD.RNWL_CONSTRUCTION (off-canonical)RNWL_DATA_ENGINEERRNWL_PROD.RNWL_CONSTRUCTION.RNWL_CONSTRUCTION_REPORTINGRNWL_CONSTRUCTION_REPORTING + Send_Email⚠️ Functional but invisible to Snowflake Intelligence UI picker; missing instructions + sample questions

Production Semantic Views

  • Sales (RNWL_PROD.RNWL_SALES): RNWL_SALES_CORE, RNWL_SALES_BROKERAGE, RNWL_SALES_TARGET, RNWL_SALES_REPORTING
  • CRM (RNWL_PROD.RNWL_CRM): RNWL_CRM_CASES, RNWL_CRM_REPORTING
  • Construction (RNWL_PROD.RNWL_CONSTRUCTION): RNWL_CONSTRUCTION_REPORTING (Colabtools — labour, progress, quality, schedule, tasks)

Known Gaps (open)

  1. CRM bot analyst tool warehouse is empty string (tool_resources.RNWL_CRM_ANALYST.execution_environment.warehouse: "") — falls back to caller’s default, leaks cost-attribution clarity. Sales bot pins WH_INTELLIGENCE. Should match.
  2. Construction bot off-canonical — lives in RNWL_PROD.RNWL_CONSTRUCTION instead of SNOWFLAKE_INTELLIGENCE.AGENTS, so it doesn’t appear in the chat picker. Also missing instructions + sample questions.
  3. CRM bot has no sample questions (Sales has 6) — first-time users see a blank slate.
  4. Test debris — 4 test agents (FINAL_TEST_AGENT, TEST_AGENT, TEST_AGENT_001), 6 test/demo semantic views (DEMO, DEMO2, TEST × multiple schemas), 2 test Cortex Search services (TEST_DOCS_SEARCH, TEST_SEARCH).

Behavioral Note for Future Agents

When asked to modify any of these bots, use CREATE OR REPLACE AGENT with the full updated spec via execute_statement. The alter_cortex_agent_v2 MCP tool exists but pass-through behavior on tools list is unverified. The invoke_cortex_agent_v2 and test_cortex_agent MCP tools currently have client-side regressions (path kwarg / positional args mismatch in CompleteSnowflakeClient) — agents must be tested via Snowflake Intelligence UI, REST API, or direct SELECT FROM SEMANTIC_VIEW(...) probes against the bound semantic view.

Send_Email Side-Effects Discipline

Calling SNOWFLAKE_INTELLIGENCE.AGENTS.SEND_EMAIL from non-user-initiated contexts (e.g. agent-prompted smoke tests) will be permission-blocked by the harness and should be — actual email dispatch needs explicit user authorization in chat. Functional verification of the bot’s email path should be done by the user in the Intelligence UI, not by background calls.