00-course-orchestrator

name: 00-course-orchestrator description: > GenAI agent skill navigator for the genai-agents/ folder. Start here to route agent-building, MLflow GenAI, tracing, Databricks Apps Track A, SDLC, evaluation, deployment, monitoring, prompt iteration, and vibecoding runtime tasks to the correct specialized skill. Uses tiered loading and keyword routing so agents read only the skills needed for the current task. license: Apache-2.0 clients: [ide_cli, genie_code] bundle_resource: none deploy_verb: none deploy_note: "Course navigator/router — selects which genai-agents skills to load; no deployed resource and no deploy verb. Client-agnostic routing. Client-awareness layer present (M06): client_context is detected/gated by skills/vibecoding-state; skills/genie-code-environment referenced by the Runtime Contract and the A7 deploy route. No inline behavior detail." coverage: full metadata: author: "prashanth-subrahmanyam" version: "5.0.0" domain: "genai-agents" role: "navigator" scope: "strictly genai-agents/; alternate tracks only linked as optional accelerator mirrors" last_verified: "2026-06-02" volatility: low upstream_sources: [] references: - "references/canonical-walkthrough-ordering.md" - "references/skyloyalty-variants.md" - "references/alternate-methods-catalog.md" - "../../skills/vibecoding-state/references/state-template.md" - "../../skills/vibecoding-state/references/retrospective-template.md"

GenAI Agent Skill Navigator

Purpose

This skill is the entry point and router for all skills under genai-agents/. It chooses the right skill to load for a task, preserves the canonical course progression, and keeps context small by using progressive loading.

The navigator does not duplicate implementation instructions. Each worker skill owns its own commands, validation gates, references, scripts, and notes to carry forward.

For the full workshop-style walkthrough with per-step Activity / Output / Gate tables, load references/canonical-walkthrough-ordering.md.

When to Use

Pipeline Progression Overview

The canonical path is Track A: custom Python agent built with OpenAI Agents SDK and deployed on Databricks Apps.

Canonical ordering and time estimates are in references/canonical-walkthrough-ordering.md.

Runtime Contract

For vibecoding workshops, always load ../../skills/vibecoding-state/SKILL.md before acting on a prompt. That sibling skill owns bootstrap, enter, exit, applicability checks, state-file updates, and retrospectives. This navigator only routes to the correct domain skill.

Client awareness (by reference). vibecoding-state also detects and gates client_context. The course routes the same for every client; only deploy/run mechanics differ. On the in-workspace agent, load ../../skills/genie-code-environment/SKILL.md (pre-auth, serverless, page-context CLI, project-rooted artifacts — skills load from the .assistant/skills copy) — local dev servers and CLI/profile authentication do not apply there (see PRE-REQUISITES §11 for the IDE/CLI auth path). Routed prompts open with a client-specific RULE_0 preamble; follow it as written.

Genie Code first-run. Clone the whole repo into /Users/<your-username>/.assistant/skills/vibe-coding-workshop, then start a NEW Agent-mode chat thread so the skills load. Full block: repo-root AGENTS.md ("Genie Code" section) / PRE-REQUISITES.md.

Live state files are gitignored. Templates live in:

Routing Algorithm

1. Receive the user request.
2. If this is an Instructions.md / WALKTHROUGH.md workshop prompt, load
   ../../skills/vibecoding-state/SKILL.md first.
3. Detect the task family using the tables below.
4. If the task is end-to-end, route to the first skill in that stage and let
   each skill's Next Step / Notes to Carry Forward section advance the flow.
5. If the task is specific, load only that worker skill.
6. Load references/ files only when the selected skill asks for details.

Skill Structure And Numbering

How To Navigate A Skill

1. Read the selected SKILL.md first.
2. Follow its prerequisites, gates, and validation commands.
3. Load that skill's references/ files only when needed.
4. Carry forward only the skill's summary outputs before loading the next skill.

Context Budget Management

Budget Zones

Task Detection And Skill Routing

Bootstrap / Runtime Routes

Foundation Routes

Track A Routes

SDLC Routes

Progressive Loading Strategy

Step 1: Detect

Read this navigator and choose one route from the tables above.

Step 2: Load The Selected Skill

Load exactly one routed SKILL.md. For an end-to-end path, start with the first skill in the stage, then follow that skill's Next Step.

Step 3: Execute Or Explain

If implementing, follow the selected skill's commands and gates. If answering a question, cite only the selected skill and any relevant references.

Step 4: Carry Forward A Small Summary

At a stage boundary, keep only:

• The live state-file values, if this is a workshop run.
• The selected skill's outputs / Notes to Carry Forward.
• The next skill path.

Step 5: Load References Only On Demand

Prefer task-specific references/ files over broad reading. Treat scripts as black-box utilities unless debugging them.

Complete Skill Directory Map

genai-agents/
├── 00-course-orchestrator/SKILL.md                      # Navigator: routes all genai-agents tasks
│   └── references/
│       ├── canonical-walkthrough-ordering.md             # Canonical course narrative and timing
│       ├── skyloyalty-variants.md                        # SkyLoyalty V1-V5 chooser
│       └── alternate-methods-catalog.md                  # Optional non-canonical paths, outside main scope
│   (vibecoding-state relocated to repo-root skills/vibecoding-state/  # Runtime contract: state, gates, retrospectives)
├── foundation/
│   ├── 00-uc-resources-foundation/SKILL.md               # F0: UC schemas and managed volumes
│   ├── 00b-agent-spec-and-tool-plan/SKILL.md            # F0b: PRD -> Agent Spec -> Tool Plan
│   ├── 01-mlflow-genai-foundation/SKILL.md               # F1: MLflow GenAI setup and autolog
│   ├── 02-experiment-tracing-and-uc-storage/SKILL.md     # F2: experiments, tracing, UC OTEL storage
│   ├── 02b-typescript-tracing/SKILL.md                   # F2b: Node/TypeScript MLflow tracing
│   ├── 02c-trace-context-and-environments/SKILL.md       # F2c: users, sessions, env/version metadata
│   ├── 03-tools-and-data-access/SKILL.md                 # F3: MCP, tools, resource grants
│   ├── 04-ai-gateway/SKILL.md                            # F4: AI Gateway, guardrails, usage tracking
│   └── 05-knowledge-assistant/SKILL.md                   # F5: Knowledge Assistant lifecycle
├── tracks/
│   └── A-custom-agent-apps/
│       ├── 01-clone-and-run/SKILL.md                     # A1: clone template and run locally
│       ├── 02-agent-framework/SKILL.md                   # A2: OpenAI Agents SDK + ResponsesAgent
│       ├── 03-tools-and-mcp/SKILL.md                     # A3: function tools and MCP tools
│       ├── 04-authentication/SKILL.md                    # A4: SP, OBO, resource permissions
│       ├── 05-lakebase-memory/SKILL.md                   # A5: conversation memory and thread IDs
│       ├── 06-evaluation/SKILL.md                        # A6: pre-deploy agent evaluation
│       ├── 07-deploy-and-query/SKILL.md                  # A7: Databricks Apps deploy and query
│       └── 08-debugging/SKILL.md                         # A8: local/deployed agent debugging
└── sdlc/
    ├── 01-prompt-registry/SKILL.md                       # S1: UC prompt registry and aliases
    ├── 02-evaluation-datasets/SKILL.md                   # S2: eval datasets and trace/label import
    ├── 03-scorers-and-judges/SKILL.md                    # S3: scorers, judges, thresholds
    ├── 04-evaluation-runs/SKILL.md                       # S4: mlflow.genai.evaluate and labels
    ├── 04b-stakeholder-signoff/SKILL.md                  # S4b: business/compliance sign-off gate
    ├── 04c-end-user-feedback/SKILL.md                    # S4c: production feedback assessments
    ├── 05-logged-model-and-uc-registration/SKILL.md      # S5: log/register/promote agent model
    ├── 06-deployment-and-automation/SKILL.md             # S6: deploy, CI/CD, AI Gateway handoff
    ├── 07-production-monitoring/SKILL.md                 # S7: registered scorers and dashboards
    ├── 08-prompt-optimization/SKILL.md                   # S8: GEPA / optimize_prompts
    └── 08b-prompt-handauthoring/SKILL.md                 # S8b: manual prompt iteration

Post-Completion: Skill Usage Summary

After a routing or orchestration session, output a concise Skill Usage Summary reflecting what was actually loaded.

End with:

• Totals: X navigator/runtime skills, Y foundation skills, Z Track A skills, W SDLC skills, R reference files.
• Routing path: The skill sequence actually used.
• Skipped: Skills considered but not needed, with reason.
• Unplanned: Skills read outside the normal route, with reason.

Maintenance

When adding or moving a genai-agents/ skill:

1. Keep the skill's SKILL.md focused; move detailed patterns to references/.
2. Add or update frontmatter fields: name, description, metadata.domain, metadata.pipeline_position or metadata.role.
3. Add a routing row in the correct Task Detection table above.
4. Add one line in the Complete Skill Directory Map.
5. If the skill changes canonical order or timing, update references/canonical-walkthrough-ordering.md.
6. If the skill affects workshop state/gates, update ../../skills/vibecoding-state/SKILL.md or its references.
7. Verify every linked SKILL.md exists.

References

Local references

• references/canonical-walkthrough-ordering.md
• references/skyloyalty-variants.md
• references/alternate-methods-catalog.md
• ../../skills/vibecoding-state/SKILL.md
• ../../skills/vibecoding-state/references/state-template.md
• ../../skills/vibecoding-state/references/retrospective-template.md

Official documentation

• Author an AI agent and deploy it on Databricks Apps
• Authentication for AI agents
• Debug AI agents
• Agents development workflow
• Store MLflow traces in Unity Catalog
• Collect end-user feedback
• Automatically optimize prompts

Version History