Author Snowflake MCP + skill upgrade plan to v22.0.0 / Bible v19.1.7.…

Decision

Author Snowflake MCP + skill upgrade plan to v22.0.0 / Bible v19.1.7. Hard-ban 5 Gen 1 Cortex tools (JSON-RPC -32006 + 3 migration helpers); add +79 net new capability tools (Cortex Code, Agent Teams, Semantic View Autopilot, Cortex Fine-Tuning lifecycle, Trust Layer Feedback, Postgres-on-Snowflake/OpenFlow/dbt-Projects lifecycles, SPCS observability, Iceberg admin); single coordinated cutover; Tier 1 R&D classification per R16; deliver inline per new feedback_present_inline_not_files.md.

Rationale

11 of 12 Bible v19.1.7 Completion Gate checks currently fail or partial. Server self-declares Bible v18.1; current canonical v19.1.7 has critical R24 docket gate (~4% idle CPU reclaim per server), Pattern 1 verb-prefix outputSchema (system-wide Check 12.1 fix without 856 hand-authored Pydantic models), Pattern 3 install_annotator (Check 12.2/12.3 backend-aware injection). Five Gen 1 tools surface live, breaking the skill mandate at the wire. Hard-ban over soft-remove is the only in-band signal preventing ghost-capability bypass via raw SQL. Single cutover means one Contract sign, one Health Census, one R13 cascade. Tier 1 selected per R16 anti-gold-plating. Premortem identified 5 material risks with explicit mitigations baked into phase ordering.

Alternatives Rejected

Outcome

Pending