feat: add ADK eval datasets for applying-cms-brand skill

Add three evaluation datasets to satisfy the mandatory ADK Eval
onboarding requirement for skill contributions to afv-library.

Datasets cover: single brand (happy path), multiple brands (user
selection), and no brand found (graceful degradation).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Runying Mao 2026-05-13 11:44:48 -07:00
parent 649931b312
commit b07607caab
11 changed files with 227 additions and 0 deletions

View File

@ -0,0 +1,47 @@
# Expected Output
## Agent Behavior
1. Agent calls `search_brands` to find available brands
2. Agent finds 3 brands and presents options to user using `ask_followup_question`:
- "I found 3 brands in your CMS. Which one should I apply?
1. Nova Enterprise Brand
2. Nova Startup Brand
3. Nova Internal Communications
Which brand would you like to use?"
3. User selects "Nova Enterprise Brand"
4. Agent calls `get_brand_instructions` with managedContentId "9Pn5g000000AAA111"
5. Agent applies brand guidelines to the product description
## Expected Branded Product Description
The generated description should:
- Lead with outcomes, not features
- Use power verbs (accelerate, transform, empower, unlock)
- Have a bold, authoritative, forward-thinking voice
- Be confident, aspirational, and direct in tone
- Keep paragraphs to 3 sentences maximum
- Use second person ("you", "your team")
- Not use "simple", "easy", or clichés
- Incorporate key messages where natural
## Example Branded Output
"Your Team, Unleashed.
Accelerate collaboration across your entire organization. Track progress, share files, and communicate in real time — all from one platform your team already knows how to use.
Enterprise-grade reliability. Startup-speed agility. Built for teams that move fast."
## Evaluation Criteria
- [ ] Agent searched for brands before generating content
- [ ] Agent presented all brand options to user (did not auto-select)
- [ ] Agent waited for user selection before proceeding
- [ ] Agent extracted brand instructions for the selected brand
- [ ] Output leads with outcomes, not features
- [ ] Output uses power verbs
- [ ] Output matches brand voice (bold, authoritative, forward-thinking)
- [ ] Output matches brand tone (confident, aspirational, direct)
- [ ] Guardrails are respected (no passive headlines, no "simple"/"easy", no clichés)
- [ ] Agent summarized which guidelines were applied

View File

@ -0,0 +1,5 @@
Search for our brand and apply it to the following product description:
"Our project management tool helps teams collaborate efficiently. It includes task tracking, file sharing, and real-time messaging."
Use our brand voice and tone to make it compelling.

View File

@ -0,0 +1,10 @@
{
"description": "Mocked response from get_brand_instructions tool for Nova Enterprise Brand (selected by user).",
"tool": "get_brand_instructions",
"input": {
"managedContentId": "9Pn5g000000AAA111"
},
"response": {
"promptBody": "Brand: Nova Enterprise\n\nBrand Voice: Bold, authoritative, and forward-thinking. Speak as a market leader.\n\nBrand Tone: Confident, aspirational, and direct. Inspire action and convey expertise.\n\nKey Messages:\n- Built for teams that move fast\n- Enterprise-grade reliability, startup-speed agility\n- The future of work, delivered today\n\nContent Rules:\n- Lead with outcomes, not features\n- Use power verbs (accelerate, transform, empower, unlock)\n- Keep paragraphs to 3 sentences maximum\n- Use second person (\"you\", \"your team\")\n- Sentence fragments are acceptable for emphasis\n\nTypography:\n- Headlines: Bold, concise (5 words or fewer)\n- Body: Clear, punchy sentences\n\nGuardrails:\n- Never use passive voice in headlines\n- Do not say \"simple\" or \"easy\" (implies the problem was trivial)\n- Avoid clichés (\"game-changer\", \"synergy\", \"leverage\")\n- Do not reference specific pricing"
}
}

View File

@ -0,0 +1,36 @@
{
"description": "Mocked response from search_brands tool when multiple brands exist in the org.",
"tool": "search_brands",
"input": {
"inputs": [
{
"searchQuery": "brand"
}
]
},
"response": {
"results": [
{
"managedContentId": "9Pn5g000000AAA111",
"managedContentKey": "MCK-brand-nova-enterprise",
"title": "Nova Enterprise Brand",
"contentUrl": "/cms/brand/nova-enterprise",
"totalResults": 3
},
{
"managedContentId": "9Pn5g000000BBB222",
"managedContentKey": "MCK-brand-nova-startup",
"title": "Nova Startup Brand",
"contentUrl": "/cms/brand/nova-startup",
"totalResults": 3
},
{
"managedContentId": "9Pn5g000000CCC333",
"managedContentKey": "MCK-brand-nova-internal",
"title": "Nova Internal Communications",
"contentUrl": "/cms/brand/nova-internal",
"totalResults": 3
}
]
}
}

View File

@ -0,0 +1,30 @@
# Expected Output
## Agent Behavior
1. Agent calls `search_brands` to find available brands
2. Agent receives empty results (no brands found)
3. Agent does NOT attempt to generate branded content without guidelines
4. Agent informs the user and offers alternatives:
- "No brands found in Salesforce CMS. To use branding:
1. Create a brand in Salesforce CMS (Content Type: sfdc_cms__brand)
2. Provide brand guidelines directly in this conversation
Would you like to proceed without CMS branding, or provide guidelines manually?"
## Expected Content Output
No branded content should be generated. The agent should:
- Clearly communicate that no brands were found
- Explain how to create a brand in CMS
- Offer the alternative of providing guidelines manually
- Wait for user direction before proceeding
## Evaluation Criteria
- [ ] Agent searched for brands before attempting to generate content
- [ ] Agent did NOT generate branded content without guidelines
- [ ] Agent informed user that no brands were found
- [ ] Agent offered actionable alternatives (create in CMS or provide manually)
- [ ] Agent did NOT hallucinate or assume brand guidelines
- [ ] Agent waited for user response before proceeding
- [ ] Error message was clear and helpful (not technical/confusing)

View File

@ -0,0 +1,4 @@
Apply our brand to the following landing page headline and subheading:
Headline: "Welcome to Our Service"
Subheading: "We provide the tools you need to succeed in today's market."

View File

@ -0,0 +1,15 @@
{
"description": "Mocked response from search_brands tool when no brands are found in the org.",
"tool": "search_brands",
"input": {
"inputs": [
{
"searchQuery": "brand"
}
]
},
"response": {
"results": [],
"totalResults": 0
}
}

View File

@ -0,0 +1,43 @@
# Expected Output
## Agent Behavior
1. Agent calls `search_brands` with a general query
2. Agent finds one brand ("Acme Corp Brand") and confirms with user: "I found the brand 'Acme Corp Brand'. Should I apply this brand's guidelines to the content?"
3. After user confirms, agent calls `get_brand_instructions`
4. Agent applies brand guidelines and produces branded email content
## Expected Branded Email
The generated email should:
- Use a professional yet approachable voice
- Have a warm, encouraging, empowering tone
- Address the reader directly with "you"
- Use active voice
- Keep sentences under 20 words
- Contain no more than 1 exclamation mark per paragraph
- Not use superlatives ("best", "fastest", "most powerful")
- Incorporate key messages where natural (innovation made simple, your success is our priority)
## Example Branded Output
"Hi, welcome to Acme Corp. We're glad you're here.
You now have access to tools designed to help you succeed. Here's how to get started:
1. Set up your profile so we can personalize your experience.
2. Explore the dashboard to see what's available to you.
3. Reach out to our team if you need guidance along the way.
Your success drives everything we do. We're here to support you at every step."
## Evaluation Criteria
- [ ] Agent searched for brands before generating content
- [ ] Agent confirmed brand selection with user
- [ ] Agent extracted brand instructions before generating content
- [ ] Output matches brand voice (professional yet approachable)
- [ ] Output matches brand tone (warm, encouraging, empowering)
- [ ] Content rules are respected (active voice, short sentences, "you" addressing)
- [ ] Guardrails are not violated (no competitor names, no performance guarantees)
- [ ] Agent summarized which guidelines were applied

View File

@ -0,0 +1,5 @@
Apply my brand guidelines to the following welcome email:
"Hi there, welcome to our platform! We're excited to have you on board. Here's what you can do to get started..."
Make sure the email matches our brand voice and tone.

View File

@ -0,0 +1,10 @@
{
"description": "Mocked response from get_brand_instructions tool for Acme Corp brand.",
"tool": "get_brand_instructions",
"input": {
"managedContentId": "9Pn5g000000XYZABC"
},
"response": {
"promptBody": "Brand: Acme Corp\n\nBrand Voice: Professional yet approachable. Speak with confidence and clarity. Avoid jargon.\n\nBrand Tone: Warm, encouraging, and empowering. Make the reader feel supported and capable.\n\nKey Messages:\n- Innovation made simple\n- Your success is our priority\n- Technology that works for you\n\nContent Rules:\n- Use active voice\n- Keep sentences under 20 words\n- Address the reader directly with \"you\"\n- Avoid superlatives (never say \"best\", \"fastest\", \"most powerful\")\n- Do not use exclamation marks excessively (max 1 per paragraph)\n\nGuardrails:\n- Never mention competitors by name\n- Do not make performance guarantees or promises\n- Avoid technical jargon unless explaining it immediately after"
}
}

View File

@ -0,0 +1,22 @@
{
"description": "Mocked response from search_brands tool when a single brand exists in the org.",
"tool": "search_brands",
"input": {
"inputs": [
{
"searchQuery": "brand"
}
]
},
"response": {
"results": [
{
"managedContentId": "9Pn5g000000XYZABC",
"managedContentKey": "MCK-brand-acme-corp",
"title": "Acme Corp Brand",
"contentUrl": "/cms/brand/acme-corp",
"totalResults": 1
}
]
}
}