> For AI agents: the complete documentation index is available at /llms.txt, the full documentation bundle is available at /llms-full.txt.

# Meta-Agentic GenAIScript System - Minimal Product Requirements & Architecture

## Executive Summary

This document presents a **minimal viable product (MVP)** for a Meta-Agentic System: a simple AI assistant that helps users create GenAIScript agents by asking specific questions about GenAIScript APIs and documentation, then generating basic working scripts.

**Key Constraint**: The system assumes **zero prior knowledge** of GenAIScript and must learn through targeted API documentation queries via Context7 MCP integration.

## 🎯 MVP Product Requirements

### What We're Building

A **simple GenAIScript learning assistant** that:

1. Takes a basic task description from user
2. Asks Context7 specific questions about GenAIScript APIs
3. Generates a minimal working GenAIScript based on learned information
4. Validates the generated script syntax

### Core User Journey

```
User: "I want to create a script that reads a file and summarizes it"
↓
System: Queries Context7 → "How to read files in GenAIScript?"
System: Queries Context7 → "What is the script() function signature?"
System: Queries Context7 → "How to create prompts with $`...`?"
↓
System: Generates working .genai.js file
```

## 🏗️ Simplified Architecture

### Core Components

#### 1. Task Parser (Minimal)

**Purpose**: Extract basic intent from user request

- **Input**: Natural language task description
- **Output**: Simple task object with basic requirements
- **No Complex Classification**: Just identify if it's file-based, web-based, or API-based

#### 2. Context7 Query Engine

**Purpose**: Ask specific, minimal questions about GenAIScript

- **Targeted Queries**: Ask only what's needed for current task
- **API-Focused**: Query specific function signatures, not generic concepts
- **Progressive Learning**: Ask follow-up questions based on initial responses

**Example Query Strategy**:

```
Task: "read and summarize files"
↓
Query 1: "script function signature and basic usage"
Query 2: "file reading functions in GenAIScript"
Query 3: "prompt template syntax with $`...`"
```

#### 3. Simple Code Generator

**Purpose**: Create minimal working GenAIScript from learned API info

- **Template-Based**: Use very simple templates filled with API info
- **Single File Output**: Generate one .genai.js file
- **Basic Structure Only**: script() + basic prompt, no complex agents initially

#### 4. Validation Layer (Basic)

**Purpose**: Check if generated code looks syntactically correct

- **Syntax Check**: Verify JavaScript syntax
- **GenAIScript Pattern Check**: Ensure it follows basic GenAIScript patterns
- **No Execution**: Just static validation initially

### 🔄 MVP Workflow

```mermaid
flowchart TD
    A[User Task Description] --> B[Task Parser]
    B --> C[Identify Basic Requirements]
    C --> D[Context7 Query Engine]
    D --> E["Query: script() syntax"]
    D --> F[Query: relevant APIs]
    D --> G[Query: prompt templates]
    E --> H[Simple Code Generator]
    F --> H
    G --> H
    H --> I[Generated .genai.js]
    I --> J[Basic Validation]
    J -->|Valid| K[Output Script]
    J -->|Invalid| L[Error Report]
```

## 🎯 MVP Design Principles

### 1. Zero-Knowledge Assumption

- **No Pre-loaded Information**: System starts with zero GenAIScript knowledge
- **Just-in-Time Learning**: Query Context7 only when specific info is needed
- **Minimal Context**: Ask for specific API details, not broad concepts

### 2. Question-Driven Development

- **Specific Queries**: "What is the workspace.readText() function signature?"
- **Not Generic**: Avoid "How does file reading work in GenAIScript?"
- **Progressive**: Ask follow-up questions based on received answers

### 3. Simple Generation

- **Single Template**: One basic script structure template
- **Fill-in-the-Blanks**: Replace placeholders with Context7 responses
- **No Complex Logic**: Minimal conditional generation

### 4. Quick Validation

- **Syntax Only**: Check JavaScript syntax correctness
- **Pattern Matching**: Verify it looks like a GenAIScript
- **Fast Feedback**: Immediate error reporting

## 📋 MVP Implementation Plan

### Phase 1: Core MVP (Week 1)

1. **Basic Task Parser**: Identify file/web/API operations from user input
2. **Context7 Query System**: Create specific API lookup functions
3. **Simple Template**: Single GenAIScript template with placeholders
4. **Basic Generator**: Fill template with Context7 responses

**Deliverable**: Working prototype that can generate simple file-reading scripts

### Phase 2: Enhancement (Week 2)

1. **Expand Templates**: Add 2-3 more basic script patterns
2. **Better Queries**: Refine Context7 question formulation
3. **Basic Validation**: Add syntax checking
4. **Error Handling**: Graceful failure when Context7 queries fail

**Deliverable**: System handles 3 basic use cases (file ops, simple prompts, basic tools)

### Phase 3: Polish (Week 3)

1. **User Interface**: Simple command-line or basic UI
2. **Documentation**: Clear usage instructions
3. **Testing**: Validate generated scripts actually work
4. **Refinement**: Fix issues discovered in testing

**Deliverable**: Usable MVP ready for initial user feedback

## 🔧 MVP Technical Specs

### Core Dependencies

- **Context7 MCP**: Only dependency - for GenAIScript API lookups
- **Basic JavaScript**: For template processing and validation
- **File System**: To output generated .genai.js files

### MVP Output Characteristics

- **Single File**: Generate one .genai.js file per request
- **Basic Structure**: script() function + simple prompt
- **Working Code**: Must be syntactically correct and executable
- **No Complex Features**: No agents, complex tools, or schemas initially

### Success Criteria

- **Functional**: Generated script runs without syntax errors
- **Correct API Usage**: Uses proper GenAIScript function signatures learned from Context7
- **Fast**: \<10 seconds from user input to generated script
- **Simple**: Non-technical users can understand the process

## 📝 Example MVP Usage

### Input

```
User: "Create a script that reads README.md and summarizes it in 2 sentences"
```

### System Process

```
1. Parse Task → "file reading" + "summarization"
2. Query Context7 → "workspace.readText function usage"
3. Query Context7 → "script() function basic syntax"
4. Query Context7 → "how to create prompts with template literals"
5. Generate Script → Fill template with learned API info
```

### Output

```javascript
// Generated: readme-summarizer.genai.js
script({
  title: "README Summarizer",
  description: "Reads README.md and creates a 2-sentence summary"
})

const content = await workspace.readText("README.md")
def("README", content)

$`Summarize the README content in exactly 2 sentences:

${README}

Provide a clear, concise 2-sentence summary.`
```

## 🎯 MVP Success Metrics

### Functional Goals

- Generate working GenAIScript for 3 basic patterns: file reading, simple prompts, basic data processing
- 80%+ success rate for syntactically correct output
- System asks \<5 Context7 queries per generation

### User Experience Goals

- Non-technical users can describe tasks in natural language
- Clear error messages when generation fails
- Generated scripts run without modification

## 🚀 Next Steps

### Immediate Actions

1. **Validate Concept**: Test Context7 queries manually to confirm GenAIScript docs are accessible
2. **Build Basic Parser**: Simple task classification (file/web/api)
3. **Create Template**: Single working GenAIScript template with placeholders
4. **Test Generation**: Generate and validate one working script end-to-end

### Success Definition

**MVP is successful when it can reliably generate working GenAIScript files for file-reading tasks using only Context7 API lookups, with zero hardcoded GenAIScript knowledge.**

<details class="page-source"><summary>Page source</summary>

<a href="/source/docs/decisions/0000-agentic-system.md.txt">Source snapshot</a> · Revision `f928834c8`.

</details>