00-appkit-navigator

name: 00-appkit-navigator description: > Entry-point navigator for Databricks AppKit + Lakebase development. Routes tasks to the correct skill based on keyword detection: scaffolding, plugin integration, feature building, or deployment. This is a routing skill -- it does not generate code. It directs the agent to the correct specialized skill. Use this skill as the starting point for any AppKit-related task. Triggers on "AppKit", "Lakebase app", "Databricks app", "build app", "deploy app", "scaffold app", "add plugin", "AppKit project", "agent endpoint", "wire agent", "serving plugin", "agent app proxy", "chat history", "feedback". license: Apache-2.0 clients: [ide_cli, genie_code] coverage: full

Client-awareness body layer added in Milestone 06 (batch 6e, criterion 3 / step 4):

Tier-1-thin client callout (client_context via skills/vibecoding-state, skills/genie-code-environment

pointer), npx docs IDE-gated with a Genie Code alternative, $PROFILE annotated as an IDE/CLI concept,

and the Development-Lifecycle deploy(agnostic)/local-test(IDE-only) split — all by reference, no inline behavior.

metadata: author: prashanth subrahmanyam version: "1.1.0" domain: meta role: navigator standalone: true last_verified: "2026-06-02" volatility: low upstream_sources: [] # Routing skill; canonical Apps guidance lives in databricks-agent-skills (see See Also).

AppKit Lakebase Navigator

Route AppKit + Lakebase tasks to the correct specialized skill.

Client awareness (Tier-1-thin). Routing is identical for every client; only the run/test/docs mechanics differ. skills/vibecoding-state detects and gates client_context; on the in-workspace agent load skills/genie-code-environment for behavior (pre-auth, serverless, page-context CLI, project-rooted artifacts — skills load from the .assistant/skills copy). The deploy verb is client-agnostic, but the local dev server / E2E test steps below are IDE/CLI-only — Genie Code verifies against the deployed app (see 03-appkit-deploy and the routed skill's deploy_note). Routed prompts open with a client-specific RULE_0 preamble; follow it.

Genie Code first-run (once). Two locations: git clone the workshop repo into your user project (/Workspace/Users/<your-username>/vibe-coding-workshop = artifact_root, git-backed so generated bundles are recognized), then copy the tree into /Workspace/Users/<your-username>/.assistant/skills/vibe-coding-workshop for skill discovery. After the copy lands, start a NEW Agent-mode chat thread (hard-refresh if skills don't appear) so Genie Code recurses the .assistant/skills copy and loads these skills. Full block: repo-root AGENTS.md ("Genie Code" section) / PRE-REQUISITES.md. Grounded in the Genie Code skills docs.

Development Lifecycle

AppKit apps follow a branch-aware development lifecycle. Each step maps to a skill or prompt:

Step 1          Step 2         Step 3            Step 4
Scaffold ──► Build ──► Deploy mock ──► Setup Lakebase
 (01)        (02)     (03-deploy)    (04-plugin-add + bundle config)
                                      │
                                      ├─► Wire Lakebase ────────────────┐
                                      │    (05-lakebase-wiring)         │
                                      │                                  │
                                      ├─► Wire Serving endpoint ────────┤
                                      │    (06-serving-wiring)           │
                                      │                                  │
                                      └─► Wire separate Agent App ──────┤
                                           (06d-agent-app-proxy)         │
                                                                         ▼
                                      Optional Chat History ──► Feedback ──► Deploy+E2E
                                      (07-appkit-chat-history) (08-appkit-feedback) (03-deploy)

Lakebase, serving endpoint wiring, and separate Agent App proxy wiring are independent branches. An app may use Lakebase only, agent chat only, or both. Choose 06 for Model Serving / Agent Serving endpoints and 06d for an agent deployed as its own Databricks App.

Use 06d-appkit-agent-app-proxy instead of 06-appkit-serving-wiring when the agent is deployed as its own Databricks App rather than a Model Serving / Agent Serving endpoint. 06d is the canonical 2-Apps path used by the root genai-agents/PROMPT-GUIDE.md.

Steps 7 and 8 are optional and apply to agent-chat apps. 07-appkit-chat-history requires Lakebase (05) and either the Serving endpoint path (06) or the separate Agent App proxy path (06d). 08-appkit-feedback requires 07-appkit-chat-history (it reads the Vote table and traceId column that 07 creates).

Start at Scaffold, Build & Test for new projects. Jump to any step if prior steps are complete.

Lakebase flow: After Wire Lakebase Backend, the app has Lakebase code but runs locally with mock fallback data. Deploy and E2E Test deploys so the Service Principal creates database objects, then runs E2E verification.

Skill-driven vs prompt-driven steps: Scaffold/Build and Wire Lakebase Backend are skill-driven — the prompt delegates to skills for reusable patterns. Setup Lakebase and Deploy and E2E Test are prompt-driven — the prompt contains application-specific procedures with skills as supplementary references. Skills encode reusable, PRD-independent knowledge (how to scaffold, deploy, register a plugin, wire a database). Prompts encode orchestration and application-specific context (variable values, step prerequisites, handoff instructions).

Task Routing Table

Match the user's request keywords to the correct skill. Read the skill's SKILL.md before proceeding.

Routing Algorithm

1. User request received
2. Detect keywords from the routing table above
3. IF "create" / "new" / "scaffold" / "init"         → Read apps_lakebase/skills/01-appkit-scaffold/SKILL.md
4. IF "add plugin" / "integrate" / "add lakebase"     → Read apps_lakebase/skills/04-appkit-plugin-add/SKILL.md
5. IF "build" / "implement" / "PRD" / "UI"            → Read apps_lakebase/skills/02-appkit-build/SKILL.md
6. IF "setup lakebase" / "add lakebase plugin"         → Read apps_lakebase/skills/04-appkit-plugin-add/SKILL.md (plugin install)
                                                        + apps_lakebase/prompts/03-setup-lakebase.md (bundle resources)
7. IF "wire lakebase" / "connect lakebase" / "DDL"   → Read apps_lakebase/skills/05-appkit-lakebase-wiring/SKILL.md (patterns)
                                                        + apps_lakebase/prompts/04-lakebase-wiring.md (orchestration, no deploy)
8. IF "wire agent" / "serving plugin" / "agent endpoint" / "model serving"
                                                      → Read apps_lakebase/skills/06-appkit-serving-wiring/SKILL.md
9. IF "agent app proxy" / "two apps" / "separate Agent App" / "OBO forwarding"
                                                      → Read apps_lakebase/skills/06d-appkit-agent-app-proxy/SKILL.md
10. IF "chat history" / "save conversations" / "persistent chat" / "conversation sidebar"
                                                      → Read apps_lakebase/skills/07-appkit-chat-history/SKILL.md
11. IF "feedback" / "thumbs" / "MLflow assessment" / "rate response" / "user rating"
                                                      → Read apps_lakebase/skills/08-appkit-feedback/SKILL.md
12. IF "e2e test" / "test lakebase" / "deploy lakebase" → Read apps_lakebase/prompts/05-e2e-test.md (deploy + E2E)
13. IF "lakebase CLI" / "lakebase troubleshoot"       → Read and follow databricks-lakebase agent skill
14. IF "deploy" / "ship" / "fix deploy"               → Read apps_lakebase/skills/03-appkit-deploy/SKILL.md
15. IF ambiguous or multi-step                        → Ask user to clarify, or follow
                                                        the lifecycle order (scaffold → build → deploy)

Skill Inventory

To see the full directory tree, run: find apps_lakebase/skills/ -type f -name "*.md" | sort

Skill Dependencies

• 02-appkit-build requires a scaffolded project (01-appkit-scaffold must run first)
• 02-appkit-build cross-references 03-appkit-deploy in its "What's Next" section
• 03-appkit-deploy expects $APP_NAME and $PROFILE to be set by the caller. $PROFILE is an IDE/CLI concept (a ~/.databrickscfg profile); on the in-workspace agent there is no profile (pre-auth/serverless) — see skills/genie-code-environment.
• 04-appkit-plugin-add can run at any point after scaffolding
• 05-appkit-lakebase-wiring requires the Lakebase plugin to be registered first (04-appkit-plugin-add)
• 06-appkit-serving-wiring requires the Serving plugin to be registered first (04-appkit-plugin-add)
• 06-appkit-serving-wiring is independent of 05-appkit-lakebase-wiring — an app may use either or both
• 06d-appkit-agent-app-proxy requires a deployed Agent App and AppKit app resource binding (CAN_USE on the Agent App)
• 07-appkit-chat-history requires 05-appkit-lakebase-wiring plus either 06-appkit-serving-wiring or 06d-appkit-agent-app-proxy (persists agent responses to Lakebase)
• 08-appkit-feedback requires 07-appkit-chat-history (reads the chat."Vote" table and chat."Message".traceId column created there)

AppKit Documentation (Live)

For the latest API details, always consult the live docs first:

# IDE/CLI clients (local Node + npx):
npx @databricks/appkit docs              # documentation index
npx @databricks/appkit docs "<query>"    # search for a specific topic
npx @databricks/appkit docs --full       # full index with all API entries

Genie Code: npx (local Node) is not available. Use the AppKit docs hub under See Also (databricks.github.io/appkit) or the per-skill reference docs, and run any AppKit commands via the workspace page-context CLI (runDatabricksCli) — see skills/genie-code-environment.

See Also

• Authoritative upstream: databricks-agent-skills / databricks-apps — back-up reference for canonical Databricks Apps platform guidance.
• AppKit docs hub: databricks.github.io/appkit