intent based workflow

This commit is contained in:
Brian Madison
2025-10-16 07:58:28 -05:00
parent a28a350e14
commit 516fa1a917
4 changed files with 276 additions and 317 deletions

View File

@@ -91,6 +91,27 @@ Work with user to outline the workflow steps:
- Which steps should repeat?
- What variables/outputs does each step produce?
<ask>What instruction style should this workflow favor?
**1. Intent-Based (Recommended)** - Guide the LLM with goals and principles, let it adapt conversations naturally
- More flexible and conversational
- LLM chooses appropriate questions based on context
- Better for complex discovery and iterative refinement
- Example: `<action>Guide user to define their target audience with specific demographics and needs</action>`
**2. Prescriptive** - Provide exact wording for questions and options
- More controlled and predictable
- Ensures consistency across runs
- Better for simple data collection or specific compliance needs
- Example: `<ask>What is your target platform? Choose: PC, Console, Mobile, Web</ask>`
Note: Your choice will be the _primary_ style, but we'll use the other when it makes more sense for specific steps.</ask>
<action>Store instruction_style preference (intent-based or prescriptive)</action>
<action>Explain that both styles have value and will be mixed appropriately</action>
Create a step outline with clear goals and outputs.
</step>
@@ -188,6 +209,122 @@ Example usage in instructions:
<output>Hello {user_name}, the workflow is complete!</output>
```
<critical>Applying instruction style preference:</critical>
Based on the {{instruction_style}} preference from Step 3, generate instructions using these patterns:
**Intent-Based Instructions (Recommended for most workflows):**
Focus on goals, principles, and desired outcomes. Let the LLM adapt the conversation naturally.
**Good Examples:**
```xml
<!-- Discovery and exploration -->
<action>Guide user to define their target audience with specific demographics, psychographics, and behavioral characteristics</action>
<action>Explore the user's vision for the product, asking probing questions to uncover core motivations and success criteria</action>
<action>Help user identify and prioritize key features based on user value and technical feasibility</action>
<!-- Validation and refinement -->
<action>Validate that the technical approach aligns with project constraints and team capabilities</action>
<action>Challenge assumptions about user needs and market fit with thought-provoking questions</action>
<!-- Complex iterative work -->
<action>Collaborate with user to refine the architecture, iterating until they're satisfied with the design</action>
```
**Avoid (too prescriptive):**
```xml
<ask>What is your target audience age range? Choose: 18-24, 25-34, 35-44, 45+</ask>
<ask>List exactly 3 key features in priority order</ask>
```
**When to use Intent-Based:**
- Complex discovery processes (user research, requirements gathering)
- Creative brainstorming and ideation
- Iterative refinement workflows
- When user input quality matters more than consistency
- Workflows requiring adaptation to context
**Prescriptive Instructions (Use selectively):**
Provide exact wording, specific options, and controlled interactions.
**Good Examples:**
```xml
<!-- Simple data collection -->
<ask>What is your target platform? Choose: PC, Console, Mobile, Web</ask>
<ask>Select monetization model: Premium, Free-to-Play, Subscription, Ad-Supported</ask>
<!-- Compliance and standards -->
<ask>Does this comply with GDPR requirements? [yes/no]</ask>
<ask>Choose documentation standard: JSDoc, TypeDoc, TSDoc</ask>
<!-- Binary decisions -->
<ask>Do you want to generate test cases? [yes/no]</ask>
<ask>Include performance benchmarks? [yes/no]</ask>
```
**Avoid (too rigid for complex tasks):**
```xml
<ask>What are your product goals? List exactly 5 goals, each 10-15 words</ask>
<ask>Describe your user persona in exactly 3 sentences</ask>
```
**When to use Prescriptive:**
- Simple data collection (platform, format, yes/no choices)
- Compliance verification and standards adherence
- Configuration with finite options
- When consistency is critical across all executions
- Quick setup wizards
**Mixing Both Styles (Best Practice):**
Even if user chose a primary style, use the other when appropriate:
```xml
<!-- Intent-based workflow with prescriptive moments -->
<step n="1" goal="Understand user vision">
<action>Explore the user's vision for their game, uncovering their creative intent and target experience</action>
<action>Ask probing questions about genre, themes, and emotional tone they want to convey</action>
</step>
<step n="2" goal="Capture basic metadata">
<ask>What is your target platform? Choose: PC, Console, Mobile, Web</ask> <!-- Prescriptive for simple choice -->
<ask>Select primary genre: Action, RPG, Strategy, Puzzle, Simulation, Other</ask>
</step>
<step n="3" goal="Deep dive into gameplay">
<action>Guide user to articulate their core gameplay loop, exploring mechanics and player agency</action> <!-- Back to intent-based -->
<action>Help them identify what makes their game unique and compelling</action>
</step>
```
**Guidelines for the chosen style:**
If user chose **Intent-Based**:
- Default to goal-oriented <action> tags
- Use open-ended guidance language
- Save prescriptive <ask> tags for simple data/choices
- Focus on "guide", "explore", "help user", "validate"
- Allow LLM to adapt questions to user responses
If user chose **Prescriptive**:
- Default to explicit <ask> tags with clear options
- Use precise wording for consistency
- Save intent-based <action> tags for complex discovery
- Focus on "choose", "select", "specify", "confirm"
- Provide structured choices when possible
**Remember:** The goal is optimal human-AI collaboration. Use whichever style best serves the user at each step.
<critical>Save location:</critical>
- Write to {{output_folder}}/instructions.md