Adopt the sfc-gh-gpope Cortex Agent routing pattern as a canonical sub-skill of snowflake for Runwal Enterprises deployment, with 10 production hardenings vs upstream and AJ-specific account adaptat
Decision
Adopt the sfc-gh-gpope Cortex Agent routing pattern as a canonical sub-skill of snowflake for Runwal Enterprises deployment, with 10 production hardenings vs upstream and AJ-specific account adaptation (SNOWFLAKE_INTELLIGENCE.ROUTING_AGENT schema, WH_INTELLIGENCE warehouse, MCP SERVER object skipped).
Rationale
Verified live in account VM65651: Cortex Agents enabled, DATA_AGENT_RUN GA 2026-03-16, 3 prod agents currently force dropdown picking. Upstream notebook has 6 compile/runtime bugs + 4 production gaps — all fixed in deploy.sql. Premortem identified 5 failure modes, all bounded: deprecation risk low + single-point-of-update; non-deploy still leaves skill as auto-fire architectural reference; routing accuracy measurable via routing_audit; test-agent pattern manually overridable; cross-skill ambiguity mitigated via Composition section + parent skill anti-trigger cascade. Skill at /root/.claude/skills/snowflake-cortex-routing-agent/SKILL.md (229 lines, in budget), SQL at /root/aj-workspace/infra/snowflake-routing-agent/deploy.sql (614 lines, 13 sections, idempotent). Cost <₹500/mo at Runwal scale, <0.5% of Tier 1 budget. Reversible: DROP SCHEMA CASCADE has zero impact on downstream agents.
Alternatives Rejected
Outcome
Pending
Related
- patch-the-4-mcp-gaps-surfaced-in-the-ea-salescrm-scoping-han
- v140-god-grade-audit-complete-agents-architect-core-30-secti
- snowflake-mcp-v2203-upgrade-quality-audit-full-pass-bible-v1
- adopt-phased-tableau-cortex-analyst-alignment-phase-0-1-defi
- supersedes-decision-409649ae-3411-47a0-a5af-fcd5ec5ef4f9-do
- encode-budget-pools-are-separate-never-conflate-or-sum-acros
- r-dash-v120-verdict-pilot-block-execute-3-wave-remediation-w