Reference Solution

A complete, working snapshot of the workshop project after all core parts are finished. Use it when something will not import, a step took longer than expected, or you joined late and need to catch up.

Read this before copying

Copying files skips the learning, not the setup. You still need an active watsonx Orchestrate environment (Part 1) and you still have to import every file you copy — the platform does not pick up files from disk on its own.

How to use it

1. Find your part in the table below.
2. Copy only the files listed for the parts you have already passed, plus the part you are stuck on, into your own workspace.
3. Run the import commands for that part.
4. Continue the workshop from the next step.

Each file links directly to its source, so you can download a single file instead of the whole folder.

What goes where

Copy into the root of your workshop project, keeping this layout:

your-project/
├── workspace_config.yaml
├── requirements.txt
├── .bob/                 ← from bob-config/ (renamed, see note below)
│   ├── mcp.json
│   ├── custom_modes.yaml
│   └── rules/wxo-dev-rule-enhanced.md
├── agents/
├── tools/
├── knowledge-bases/
└── evaluation/

bob-config/ is really .bob/

Git and MkDocs both skip dot-directories, so Bob's configuration is stored here as bob-config/. Copy it to .bob/ in your own project.

Part-by-part checkpoints

Part 1: Setup

File	Copy to
workspace_config.yaml	workspace_config.yaml
bob-config/mcp.json	.bob/mcp.json
bob-config/custom_modes.yaml	.bob/custom_modes.yaml
bob-config/rules/wxo-dev-rule-enhanced.md	.bob/rules/wxo-dev-rule-enhanced.md
scripts/add-wxo-env.sh	add-wxo-env.sh (optional helper)

Edit .bob/mcp.json and replace <ABSOLUTE_PATH_TO_YOUR_WORKSPACE> with the full path to your project folder, then reload the Bob IDE window.

Use bob-config/mcp.json for the regular setup. Windows users should use the Windows fallback mcp.json only if the regular MCP setup does not work.

add-wxo-env.sh registers and activates an environment interactively:

./add-wxo-env.sh

Part 2: First Agent

File	Copy to
agents/hello-agent.yaml	agents/hello-agent.yaml

orchestrate agents import -f agents/hello-agent.yaml

Part 3: Custom Tools

File	Copy to
tools/check_order_status.py	tools/check_order_status.py
tools/process_refund.py	tools/process_refund.py
requirements.txt	requirements.txt
agents/customer_support_agent.part4.yaml	agents/customer_support_agent.yaml

orchestrate tools import -k python -f tools/check_order_status.py -r requirements.txt
orchestrate tools import -k python -f tools/process_refund.py -r requirements.txt
orchestrate agents import -f agents/customer_support_agent.yaml

Part 4: Knowledge & Collaborators

File	Copy to
knowledge-bases/FAQ.pdf	knowledge-bases/FAQ.pdf
knowledge-bases/customer-support-faq.yaml	knowledge-bases/customer-support-faq.yaml
agents/escalation-agent.yaml	agents/escalation-agent.yaml
agents/customer_support_agent.part5.yaml	agents/customer_support_agent.yaml

orchestrate knowledge-bases import -f knowledge-bases/customer-support-faq.yaml
orchestrate knowledge-bases status -n customer_support_faq
orchestrate agents import -f agents/escalation-agent.yaml
orchestrate agents import -f agents/customer_support_agent.yaml

Wait for the knowledge base to finish indexing before testing FAQ answers.

Part 5: Guidelines & Guardrails

File	Copy to
tools/data_access_guardrail.py	tools/data_access_guardrail.py
agents/customer_support_agent.yaml	agents/customer_support_agent.yaml

orchestrate tools import -k python -f tools/data_access_guardrail.py -r requirements.txt
orchestrate agents import -f agents/customer_support_agent.yaml

This is the final state of the main agent: tools, knowledge base, collaborator, guidelines, and the pre-invoke guardrail plugin.

Part 6: Evaluation & Red-Teaming

File	Copy to
evaluation/config.yaml	evaluation/config.yaml
evaluation/datasets/*.json (12 files)	evaluation/datasets/
evaluation/red-teaming-attacks/*.json (24 files)	evaluation/red-teaming-attacks/

Fill in <region>, <instance-id>, and <environment-name> in evaluation/config.yaml before running. Never add an API key to that file.

orchestrate evaluations quick-eval -c evaluation/config.yaml -t tools
orchestrate evaluations red-teaming run -a evaluation/red-teaming-attacks -o evaluation/red-teaming-results

The datasets cover six happy paths, three edge cases, and three error scenarios. The attack set covers instruction override, crescendo, emotional appeal, role playing, prompt leakage, jailbreaking, topic derailment, and unsafe topics.

Part 7: Deployment

Deployment produces no new files. Use the Part 5 agent state above, then follow Part 7.

Known differences from the walkthrough text

This snapshot is one completed run of the workshop, not a regenerated copy of the examples embedded in each part. A few things differ:

• Agent files use customer_support_agent.yaml (underscores); the part instructions write customer-support-agent.yaml. Only the name: field inside the YAML matters to the platform.
• The greeting agent here is named hello_agent, while Part 2 uses hello_world_agent.
• The guardrail plugin here is data_access_guardrail; Part 5 builds content_safety_guardrail. They solve overlapping problems in different ways — either one satisfies the checkpoint.
• The knowledge base is built from FAQ.pdf, and the evaluation datasets are one JSON file per case (the layout the ADK's red-teaming commands produce) rather than a single JSONL file.
• agents/customer_support_agent.part4.yaml and .part5.yaml are the final agent with the later parts' sections removed, so each one imports cleanly at that stage of the workshop.

Not included

• Evaluation and red-team results. Run outputs are environment-specific and the generated ADK config embeds a live authentication token, so they are deliberately left out. Generate your own with the commands above.
• Credentials of any kind. No API keys, tokens, or .env files. The Part 1 files use placeholders you fill in yourself.
• Optional modules. Multi-agent orchestration, MCP servers, agentic workflows, AI gateway models, and NL2SQL ship their own files inside their own part folders.