@ema.co/mcp-toolkit 0.2.0

package/resources/templates/voice-ai/workflow-prompt.md

@@ -0,0 +1,120 @@
+# Workflow Prompt: {Persona Name}
+
+> **IMPORTANT**: Copy the content in the "Auto Builder Prompt" section below into Ema Auto Builder.
+> The safeguards and validation sections help ensure correct interpretation.
+
+---
+
+## Auto Builder Prompt
+
+```text
+Title: {Persona Name}
+
+Summary: {One sentence describing what this Voice AI does}
+
+Trigger: voice_call (inbound phone call)
+
+---
+## EXPLICIT NODE DEFINITIONS
+---
+
+Agents/Nodes:
+1. trigger (chat_trigger)
+   - outputs: chat_conversation, user_query
+
+2. intent_classifier (chat_categorizer)
+   - input: trigger.chat_conversation
+   - categories: [{Category1}, {Category2}, {Category3}, Fallback]
+   - output: category
+   - description: "Classifies caller intent"
+
+{For each category, define the processing nodes:}
+
+3. {category1_processor} ({action_type})
+   - trigger_when: category::{Category1}
+   - inputs: {specify inputs}
+   - outputs: {specify outputs}
+   - description: "{What this node does}"
+
+4. {category1_response} (call_llm)
+   - trigger_when: {category1_processor} completes
+   - inputs: 
+     - query: trigger.user_query
+     - named_inputs_{Category1}_Data: {category1_processor}.{output}
+   - instructions: "{Specific response instructions}"
+
+{Repeat for each category...}
+
+N. fallback_response (call_llm)
+   - trigger_when: category::Fallback
+   - inputs: query: trigger.user_query
+   - instructions: "Ask a clarifying question. Provide examples: 
+     1. {Example query 1}
+     2. {Example query 2}
+     3. {Example query 3}"
+
+---
+## EXPLICIT CONNECTIONS
+---
+
+Connections:
+trigger.chat_conversation → intent_classifier.conversation
+trigger.user_query → intent_classifier.query
+
+intent_classifier.category_{Category1} → {category1_processor}.trigger_when
+intent_classifier.category_{Category2} → {category2_processor}.trigger_when
+intent_classifier.category_{Category3} → {category3_processor}.trigger_when
+intent_classifier.category_Fallback → fallback_response.trigger_when
+
+{category1_processor}.{output} → {category1_response}.named_inputs_{Category1}_Data
+trigger.user_query → {category1_response}.query
+
+{Repeat for all connections...}
+
+---
+## RESULTS MAPPING
+---
+
+Results:
+- {category1_response}.response_with_sources → WORKFLOW_OUTPUT
+- {category2_response}.response_with_sources → WORKFLOW_OUTPUT
+- {category3_response}.response_with_sources → WORKFLOW_OUTPUT
+- fallback_response.response_with_sources → WORKFLOW_OUTPUT
+
+---
+## VALIDATION ASSERTIONS
+---
+
+Assertions:
+- [ ] intent_classifier has edges for ALL categories: {Category1}, {Category2}, {Category3}, Fallback
+- [ ] Each category has exactly one routing path to a response node
+- [ ] All search/data nodes receive TEXT_WITH_SOURCES input from trigger.user_query
+- [ ] All call_llm nodes have explicit instructions
+- [ ] Fallback provides examples of valid queries
+- [ ] All response nodes map to WORKFLOW_OUTPUT
+```
+
+---
+
+## Safeguard Checklist (Verify Before Pasting)
+
+### Structure Safeguards
+
+- [ ] **Trigger specified**: `voice_call` or `chat_trigger`
+- [ ] **All categories listed**: Including `Fallback`
+- [ ] **Explicit node definitions**: Each node has name, type, inputs, outputs
+- [ ] **Explicit connections**: Using `source.output → target.input` format
+- [ ] **Results mapping**: All outputs map to `WORKFLOW_OUTPUT`
+
+### Type Compatibility Safeguards
+
+- [ ] `chat_trigger.chat_conversation` → `chat_categorizer.conversation` ✓
+- [ ] `chat_trigger.user_query` → `call_llm.query` ✓
+- [ ] `search.search_results` → `respond_with_sources.search_results` ✓
+- [ ] `external_action_caller.tool_execution_result` → `call_llm.named_inputs_*` ✓
+
+### Category Routing Safeguards
+
+- [ ] Every category has **exactly one** outgoing edge
+- [ ] Edge format: `category_<Name>` (not `category::<Name>`)
+- [ ] Target input: `trigger_when`