00-uc-resources-foundation

name: 00-uc-resources-foundation description: > Use when any agent or downstream skill needs Unity Catalog schemas and managed volumes (knowledge_sources, agent_outputs, memory tables, benchmark tables, etc.) provisioned idempotently. Foundation Step 0 — runs before MLflow tracing setup and before any track-specific skill. Owned by no track. Reads the resolved spec to discover which volumes a use case needs and creates them alongside the canonical agent + ops UC schemas. license: Apache-2.0 clients: [ide_cli, genie_code] bundle_resource: schemas deploy_verb: bundle_deploy deploy_note: "Provisions UC schemas + managed volumes idempotently under the per-user prefixed schema ({user_schema_prefix}); created via SDK/DDL identically on both clients. On Genie Code run CLI steps through runDatabricksCli (pre-authenticated) and execute on serverless; see skills/genie-code-environment." coverage: full metadata: last_verified: "2026-04-15" volatility: medium upstream_sources: [] author: "prashanth-subrahmanyam" version: "1.1.0" domain: "genai-agents" pipeline_position: "F0" consumes: "uc_catalog, USER_SCHEMA_PREFIX, AgentSpec.required_volumes(optional), AgentSpec.knowledge_base_backend(optional)" produces: "agent_schema, ops_schema, uc_volumes(map of name → /Volumes path), signoff_volume_path" grounded_in: "docs.databricks.com/aws/en/sql/language-manual/sql-ref-syntax-ddl-create-schema, docs.databricks.com/aws/en/sql/language-manual/sql-ref-syntax-ddl-create-volume, docs.databricks.com/aws/en/volumes/managed-vs-external" fields_read: - resources.uc.catalog - resources.uc.user_schema_prefix - agent.required_volumes - agent.knowledge_base_backend

Foundation Step 0: UC Resources (Schemas + Volumes)

Cross-cutting. This skill is consumed by every Foundation step and agent track. It exists so that no downstream skill (KA, agent memory, tool exports, benchmark persistence, prompt registry) has to repeat schema or volume provisioning. If you find yourself writing CREATE SCHEMA or CREATE VOLUME inside another skill, delegate to F0 instead.

What This Skill Owns

signoffs is generic. F0 pre-creates the volume under a workshop-neutral name (signoffs) so SDLC 04b never has to provision its own UC location and SDLC 06 can resolve /Volumes/${catalog}/${ops_schema}/signoffs/v<N>/decision.md deterministically. The volume name does not carry any use-case-specific prefix (no skyloyalty_signoffs, no ${use_case_slug}_signoffs); the per-version directory v<N>/ carries the disambiguation.

Every operation is idempotent (CREATE SCHEMA IF NOT EXISTS, CREATE VOLUME IF NOT EXISTS — AlreadyExists is treated as success). It is safe to call this skill from any prompt that needs to be sure these resources exist; the cost on a warm workspace is one round-trip per resource.

What This Skill Does NOT Own

• Catalog creation. A UC catalog must already exist; F0 reads state://Resources.uc.catalog and validates the caller has USE CATALOG + CREATE SCHEMA on it.
• UC table DDL. Tables for OTeL traces, benchmarks, memory, or monitoring are owned by their respective consuming skills (F2, SDLC 02/04/07).
• Source content. F0 creates an empty knowledge_sources volume; F5 (or any other consumer) is responsible for populating it with files.
• Bronze data. Bronze schema + Delta tables are owned by the bronze layer setup skill (data_product_accelerator/skills/bronze/00-bronze-layer-setup).

When to Use

• Always, near the top of the prompt sequence — before F1 (MLflow env) so that anything downstream can assume schemas + volumes exist.
• Whenever you add a new consuming skill that needs a new volume — extend state://AgentSpec.required_volumes[] rather than putting CREATE VOLUME inside the consuming skill.
• After bootstrap when the resolved spec is available, but before any other Foundation step.

Prerequisites

Inputs

# Resolved by the caller from state — F0 does not read state directly.
uc_catalog:           "string (required)"        # e.g. "main"
agent_schema:         "string (required)"        # default: "${user_schema_prefix}_agent"
ops_schema:           "string (required)"        # default: "${user_schema_prefix}_ops"
warehouse_id:         "string (required)"        # SQL warehouse for DDL execution
required_volumes:                                 # optional — defaults below if not provided
  - { name: "knowledge_sources", schema: "agent", comment: "KA + retrieval source files" }
  - { name: "agent_outputs",     schema: "agent", comment: "Tool-generated artifacts (CSV, charts)" }
  - { name: "signoffs",          schema: "ops",   comment: "SDLC 04b stakeholder + engineering signoff decision.md artifacts" }
  # AgentSpec.required_volumes[] is appended verbatim

Operations

flowchart TD
  start["F0 invoked"] --> grant["Verify USE CATALOG + CREATE SCHEMA"]
  grant --> agentSch["CREATE SCHEMA IF NOT EXISTS<br/>${catalog}.${agent_schema}"]
  grant --> opsSch["CREATE SCHEMA IF NOT EXISTS<br/>${catalog}.${ops_schema}"]
  agentSch --> volLoop["For each required_volume:<br/>create MANAGED volume idempotently"]
  opsSch --> volLoop
  volLoop --> capture["Capture { agent_schema, ops_schema, uc_volumes{name → path} }"]

Reference implementation

from databricks.sdk import WorkspaceClient
from databricks.sdk.service.catalog import VolumeType
from databricks.sdk.errors import AlreadyExists

def provision_uc_resources(
    *,
    uc_catalog: str,
    agent_schema: str,
    ops_schema: str,
    warehouse_id: str,
    required_volumes: list[dict] | None = None,
) -> dict:
    w = WorkspaceClient()

    # 1. Schemas (DDL via statement execution against the workshop warehouse)
    for schema in (agent_schema, ops_schema):
        w.statement_execution.execute_statement(
            warehouse_id=warehouse_id,
            statement=(
                f"CREATE SCHEMA IF NOT EXISTS {uc_catalog}.{schema} "
                f"COMMENT 'F0-managed: agent or ops assets'"
            ),
            wait_timeout="30s",
        )

    # 2. Volumes (SDK; AlreadyExists is success)
    defaults = [
        {"name": "knowledge_sources", "schema": "agent",
         "comment": "KA + retrieval source files"},
        {"name": "agent_outputs", "schema": "agent",
         "comment": "Tool-generated artifacts (CSV, charts)"},
        {"name": "signoffs", "schema": "ops",
         "comment": "SDLC 04b stakeholder + engineering signoff decision.md artifacts"},
    ]
    volumes_spec = (required_volumes or []) + defaults
    seen, paths = set(), {}
    for v in volumes_spec:
        target_schema = agent_schema if v["schema"] == "agent" else ops_schema
        key = (target_schema, v["name"])
        if key in seen:
            continue
        seen.add(key)
        try:
            w.volumes.create(
                catalog_name=uc_catalog,
                schema_name=target_schema,
                name=v["name"],
                volume_type=VolumeType.MANAGED,
                comment=v.get("comment", "F0-managed volume"),
            )
        except AlreadyExists:
            pass
        paths[v["name"]] = f"/Volumes/{uc_catalog}/{target_schema}/{v['name']}"

    return {
        "agent_schema": agent_schema,
        "ops_schema": ops_schema,
        "uc_volumes": paths,
    }

The skill caller passes the result back to vibecoding-state exit so downstream skills can read state://Resources.uc_volumes.knowledge_sources etc.

Outputs (handoff)

Validation Gate

• ${uc_catalog}.${agent_schema} exists (SHOW SCHEMAS IN ${uc_catalog} LIKE '${agent_schema}')
• ${uc_catalog}.${ops_schema} exists
• Every volume in required_volumes is listable via databricks volumes list ${uc_catalog} ${schema} --output json | jq '.volumes[] | select(.name=="<name>")'
• The generic signoffs volume in ${ops_schema} is present (consumed by SDLC 04b / SDLC 06 — name is workshop-neutral)
• agent_schema, ops_schema, the full uc_volumes map, and signoff_volume_path are captured in state

Delegation Map

Limitations to Plan For

• Catalog must exist. F0 does not create catalogs (org-level admin action).
• Workspace warehouse required. Schema DDL goes through the SQL warehouse, not the Python serverless API.
• No drop semantics. F0 is create-only — to remove a schema or volume, delete out of band; do not put DROP in any agent skill.
• Volume size budgeting. Managed volumes inherit the catalog's storage location; if your agent_outputs may grow large, document a retention policy in the consuming skill.

Notes to Carry Forward

Next Step

After F0's gate passes, continue with F1: MLflow GenAI Foundation. Every subsequent skill can now assume schemas + volumes exist.

References

• CREATE SCHEMA — Databricks SQL
• CREATE VOLUME — Databricks SQL
• Managed vs External Volumes
• Databricks SDK for Python — volumes

Version History