Fabric Databricks delivery skill

The fabric-databricks-delivery skill teaches a coding agent to scaffold and package a self-contained Harness Databricks App, then register and deliver the immutable artifact through Runway gates. It keeps Databricks runtime authorization distinct from Fabric release governance and never asks the model to handle a bearer token or secret value.

Install

GitHub CLI 2.90 or newer provides the verified repository installer used by this project:

gh skill preview Fabric-Pro/fabric-runway fabric-databricks-delivery
gh skill install Fabric-Pro/fabric-runway fabric-databricks-delivery \
  --agent codex --scope project

Replace codex with claude-code, cursor, github-copilot, or another supported host. Pin a release tag or commit for controlled environments:

gh skill install Fabric-Pro/fabric-runway fabric-databricks-delivery \
  --agent codex --scope project --pin <release-tag-or-commit>

Databricks aitools install currently installs the skills in Databricks' own manifest; it does not accept an arbitrary third-party repository. This skill therefore uses the open Agent Skills repository distribution path documented by Databricks rather than claiming unsupported aitools publication.

Use

Use $fabric-databricks-delivery to add a Genie-backed OBO agent, build its self-contained
Databricks App artifact, and request a Runway preview deployment for PR 42.

The skill first runs its read-only project inspector, then routes to Harness authoring, portable App packaging, Runway delivery, or both. A complete result names the artifact directory and digest, organization, environment, deployment id, gate state, and live-evidence limitations without printing credentials.

The included inspector can also be run directly:

node .agents/skills/fabric-databricks-delivery/scripts/inspect-project.mjs .

Expected output is JSON with ok, warning, or error for the repository instructions, Harness workspace, runway.json, App artifact, entrypoints, and self-contained dependency check.

Authentication

• Harness workspace operations use a Databricks OBO, OAuth M2M, CLI-profile, or development PAT credential resolved outside prompts and source.
• Runway uses RUNWAY_API_URL, RUNWAY_ORG, and an opaque RUNWAY_API_TOKEN, or the Databricks App identity boundary for hosted MCP.
• Unity Catalog, Warehouse, Genie, AI Search, App, and MCP Service permissions remain authoritative.

Failure behavior

• Inspector errors stop before build or delivery; a missing build artifact is a warning until fh build.
• fh doctor --target databricks-app failures block packaging.
• Runway exit code 2, MCP isError, failed/stale gates, and digest mismatch are governance failures and are never bypassed or converted to warnings.
• Mock success is reported as mock-proven; only retained exact-package workspace evidence is live-proven.

Update and remove

gh skill update fabric-databricks-delivery
gh skill list --agent codex --scope project

Review changes before updating pinned enterprise installations. GitHub CLI does not currently expose an uninstall subcommand; to remove the skill, delete only the exact fabric-databricks-delivery directory reported by gh skill list (normally .agents/skills/fabric-databricks-delivery at project scope).