### Agent Prompt Examples for Tool and Knowledge Source Selection

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/generative-mode-guidance

These instructions guide the agent on when to use specific knowledge sources or topics. They are useful for disambiguating choices when multiple options might be relevant.

```text
Use the FAQ documents only if the question is not relevant to Hours, Appointments, or Billing.
Only use the ticket creation topic for creating tickets.
For other requests related to fixing issues, use the troubleshooting topic.
```

--------------------------------

### Specify Tool Usage in Instructions

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/generative-mode-guidance

When there are many tools, explicitly mention the tool name using '/' to guide the agent. Numbered instructions help ensure the agent follows them in order.

```markdown
1. When the user has provided details of their preferred laptop, create a purchase order using /"Purchase Order".
```

--------------------------------

### Generic Rubric Example

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/kit-rubrics-best-practices

An example of a rubric that is too generic for effective AI evaluation.

```text
What good looks like: A helpful, accurate response
Grade 5: Excellent response
Grade 1: Bad response
```

--------------------------------

### Example Localization File Content

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/localize-adaptive-cards

This JSON snippet shows an example of a localization file entry for an Adaptive Card. It includes a key representing the Adaptive Card element and its value containing static text and placeholders.

```json
{
  "'dialog(copilots_header_392e9.topic.recommendbooks)'.'trigger(main)'.'action(readingProgress)'.Value":
    "You are approving request {Topic.requestId} totaling {Topic.formattedTotal} for {Topic.businessUnit}"
}
```

--------------------------------

### Domain-Specific Rubric Example

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/kit-rubrics-best-practices

An example of a domain-specific rubric tailored for investor relations responses.

```text
What good looks like: A professional investor relations response that includes
relevant financial metrics, explains business drivers, maintains formal tone,
and provides forward-looking context appropriate for shareholder communications.

Grade 5 (Exemplary): Includes all key IR elements (KPIs, performance drivers,
strategic context, forward-looking insights), maintains polished professional
tone, flows logically, and requires no improvements.
```

--------------------------------

### Example YAML Structure for a Topic

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/topics-code-editor

This snippet displays a typical block of YAML used to configure a Copilot Studio topic. It illustrates various components like topic type, conversational boosting, generative answers, variables, prompts, knowledge sources, and condition groups.

```yaml
type: conversational
"conversational_boosting":
  enabled: true
  similarity_threshold: 0.62
"generative_answers":
  enabled: true
  search_keywords: "{search_keywords}"
"variables":
  - name: "$myVariable"
    type: "String"
    value: "some value"
"prompt":
  prompt_type: "custom"
  prompt_content: "This is a custom prompt."
"knowledge_sources":
  - id: "knowledgeSource1"
    type: "url"
    url: "https://www.example.com"
  - id: "knowledgeSource2"
    type: "url"
    url: "https://www.another-example.com"
"condition_groups":
  - id: "conditionGroup1"
    conditions:
      - id: "condition1"
        formula: "$myVariable = 'test'"
        redirect_to: "Topic.End"
      - id: "condition2"
        formula: "$myVariable = 'another test'"
        redirect_to: "Topic.End"
```

--------------------------------

### Rubric Grade Distinctions Example

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/kit-rubrics-best-practices

An example demonstrating clear and distinguishable characteristics for each grade level in a rubric.

```text
Grade 5: Includes metrics, drivers, context, and forward insights; professional
         tone; logical flow; investor-ready polish
Grade 4: Includes metrics, drivers, and context; mostly professional tone;
         generally logical flow; minor polish needed
Grade 3: Includes metrics and basic context; acceptable tone but generic;
         functional but lacks depth
Grade 2: Missing key metrics or context; inconsistent tone; gaps in logic
Grade 1: Missing most expected elements; unprofessional tone; confusing structure
```

--------------------------------

### Example of Poor Human Reasoning

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/kit-rubrics-refinement-workflow

An example of insufficient human reasoning, lacking specific details or justification for the assigned grade.

```text
Grade 4: Pretty good.

```

--------------------------------

### Transform API Response for Copilot Studio

Source: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/custom-knowledge-sources

This Adaptive Dialog example demonstrates how to make an HTTP request to a search API and transform its results to the format Copilot Studio expects. It uses `HttpRequestAction` to fetch data and `SetVariable` with `ForAll` to map the API's 'snippet', 'url', and 'title' to Copilot Studio's 'Content', 'ContentLocation', and 'Title'.

```yaml
kind: AdaptiveDialog
beginDialog:
  kind: OnKnowledgeRequested
  id: main
  intent: {}
  actions:
    - kind: HttpRequestAction
      id: searchRequest
      url: ="https://search-api.contoso.com/search?q=" & System.KeywordSearchQuery
      response: Topic.searchResults
      responseSchema:
        kind: Record
        properties:
          query: String
          results:
            type:
              kind: Table
              properties:
                snippet: String
                title: String
                url: String
    
    - kind: SetVariable
      id: setSearchResults
      variable: System.SearchResults
      value: |-
        =ForAll(Topic.searchResults.results,
        {
          Content: snippet,
          ContentLocation: url,
          Title: title
        })

inputType: {}
outputType: {}

```