Add new Apps Deployment skill — full lifecycle via MCP tools and CLI

[CheeYuTan]

Summary

New skill documenting the Databricks Apps deployment lifecycle: list → create → upload → deploy → verify → logs → debug → delete.

Why this is needed: Existing skills (databricks-app-python, databricks-app-apx) cover building apps but not the deployment operations. There are 6 MCP tools for app management (list_apps, get_app, get_app_logs, create_app, deploy_app, delete_app) with zero skill documentation.

What's in the skill (310 lines)

• MCP Tools Reference — all 6 tools with correct parameters, types, defaults, and CLI equivalents
• Deployment Lifecycle — 7-step walkthrough with verified response shapes
• CLI Fallback Commands — for when MCP tools have bugs
• Troubleshooting — 9 real error messages with causes and fixes
• Status Reference — 3 separate tables for compute_status, app_status, and deployment.status

Test evidence — 10 iterations of test-and-fix

This skill was tested through 10 iterations against a live Databricks workspace. Each iteration tested a claim, and if it failed, the skill was fixed and retested.

Full E2E lifecycle test

Step	Tool	Input	Result	Status
1	list_apps	{limit: 5}	Returned 5 apps with name, URL, status, deployments	PASS
2	create_app (MCP)	{name: "skill-test-app"}	SDK error — but app created server-side	BUG documented
2b	create_app (CLI)	databricks apps create skill-verify-test	Created successfully, compute_status.state = STARTING	PASS
3	get_app	{name: "skill-test-app"}	Returned full details, ACTIVE, no active_deployment	PASS
4	upload_folder	minimal app.yaml + main.py	Uploaded 2 files to workspace	PASS
5	deploy_app (MCP)	{app_name: ..., source_code_path: ...}	SDK error — KeyError on source_code_path	BUG documented
5b	deploy_app (CLI)	databricks apps deploy ... --source-code-path ...	Deployed successfully	PASS
6	get_app (poll)	{name: "skill-test-app"}	deployment.status.state = SUCCEEDED	PASS
7	get_app_logs (MCP)	{app_name: "skill-test-app"}	Import error: get_api_client	BUG documented
7b	get_app_logs (CLI)	databricks apps logs skill-test-app	Returned logs with [SYSTEM]/[APP] prefixes	PASS
8	delete_app	{name: "skill-test-app"}	Returns full app object with DELETING state	PASS
9	get_app (verify)	{name: "skill-test-app"}	DELETING → does not exist or is deleted	PASS

What the 10 iterations found and fixed

Iteration	What was wrong	Fix
2	Response shapes used flat enum strings (ComputeState.ACTIVE) — real API uses nested objects (compute_status.state: "ACTIVE")	Rewrote all response structures
3	CLI create uses positional NAME, not --name	Fixed all CLI commands
5	Delete response was {"message": "deleted"} — real API returns full app object with DELETING state	Fixed delete section
5	Can't delete while STARTING — undocumented	Added to troubleshooting
6	Deployment mode was AppDeploymentMode.SNAPSHOT — real value is "SNAPSHOT"	Fixed
9	app_status field only in get_app, not list_apps	Clarified

3 MCP tool bugs discovered and documented

Tool	Bug	Workaround
create_app	SDK v0.49.0 expects App object, not kwargs	CLI: databricks apps create <name>
deploy_app	SDK expects AppDeployment object, not flat params	CLI: databricks apps deploy <name> --source-code-path <path>
get_app_logs	cannot import name 'get_api_client'	CLI: databricks apps logs <name>

Test plan

• CI validation passes (validate_skills.py — 27 skills)
• SKILL.md frontmatter valid (name: 27 chars, description: 284 chars)
• Registered in install_skills.sh (DATABRICKS_SKILLS + description)
• Full E2E lifecycle: create → deploy → verify → logs → delete → confirm
• All response shapes verified against real API responses
• All CLI fallback commands verified
• All 9 troubleshooting errors are real (observed or documented)
• Test apps cleaned up after testing