infinite-agents-public/README.md

125 lines
5.3 KiB
Markdown

# Infinite Agentic Loop POC
> **Watch the Tutorial**: [Infinite Agentic Loop with Claude Code](https://youtu.be/9ipM_vDwflI)
An experimental project demonstrating Infinite Agentic Loop patterns using Claude Code - now with **Web-Enhanced Progressive Learning**!
<img src="images/infinite-claude-img.png" alt="Infinite Agentic Loop" style="max-width: 800px; width: 100%;">
## Overview
This project provides two powerful slash commands for orchestrating multiple AI agents in parallel:
1. **`/project:infinite`** - Original loop for generating evolving iterations based on specifications
2. **`/project:infinite-web`** - NEW! Web-enhanced loop where each iteration learns from web resources
### What's New: Web-Enhanced Learning
The new `/project:infinite-web` command creates a **self-improving knowledge loop** where:
- Each iteration fetches and learns from a different web URL
- Knowledge progressively accumulates across iterations
- Agents apply learnings from documentation, tutorials, and examples
- Output quality improves as the system learns more sophisticated techniques
## Usage
Read `.claude/settings.json` to see the permissions and commands allowed.
Start Claude Code: `claude`
Type slash command `/project:infinite` to start the infinite agentic loop.
The infinite command takes three arguments:
```
/project:infinite <spec_file> <output_dir> <count>
```
### 4 Command Variants
#### 1. Single Generation
```bash
/project:infinite specs/invent_new_ui_v3.md src 1
```
Generate one new iteration using the UI specification.
#### 2. Small Batch (5 iterations)
```bash
/project:infinite specs/invent_new_ui_v3.md src_new 5
```
Deploy 5 parallel agents to generate 5 unique iterations simultaneously.
#### 3. Large Batch (20 iterations)
```bash
/project:infinite specs/invent_new_ui_v3.md src_new 20
```
Generate 20 iterations in coordinated batches of 5 agents for optimal resource management.
#### 4. Infinite Mode
```bash
/project:infinite specs/invent_new_ui_v3.md infinite_src_new/ infinite
```
Continuous generation in waves until context limits are reached, with progressive sophistication.
## Web-Enhanced Command (NEW!)
### D3 Data Visualization Example
Generate progressively sophisticated D3 visualizations with web-based learning:
```bash
# Single visualization with web learning
/project:infinite-web specs/d3_visualization_progressive.md d3_output 1
# Batch of 5 with different web sources
/project:infinite-web specs/d3_visualization_progressive.md d3_output 5
# Progressive learning (20 iterations)
/project:infinite-web specs/d3_visualization_progressive.md d3_output 20 specs/d3_url_strategy.json
# Infinite mode - continuous learning
/project:infinite-web specs/d3_visualization_progressive.md d3_output infinite specs/d3_url_strategy.json
```
**How It Works:**
1. **Initial Priming**: Fetches 3-5 foundational web resources to build knowledge base
2. **Progressive URL Assignment**: Each iteration gets a URL matched to its complexity level
3. **Web Fetch & Learn**: Agent fetches URL, extracts techniques, applies learnings
4. **Generate & Document**: Creates output demonstrating new technique, documents source
5. **Accumulate Knowledge**: Each iteration builds on previous learnings
6. **Infinite Waves**: Continues with progressive sophistication until context limits
See [WEB_ENHANCED_GUIDE.md](WEB_ENHANCED_GUIDE.md) for complete documentation and how to create your own web-enhanced loops!
## How the Original Loop Works
1. **Specification Analysis**: Reads and understands the spec file requirements
2. **Directory Reconnaissance**: Analyzes existing iterations to determine starting point
3. **Parallel Coordination**: Deploys Sub Agents with unique creative directions
4. **Quality Assurance**: Ensures each iteration is unique and spec-compliant
5. **Wave Management**: For infinite mode, manages successive waves of agents
## Directions you can take to enhance this pattern
### Original Loop Enhancements
- Apply this to a use case of your choice
- Build an MCP Server that enables reuse of the infinite agentic loop
- Get the `.claude/commands/infinite.md` into your `~/.claude/commands/` directory for global use
- Update `.claude/commands/infinite.md` to generate sets of files instead of a single file
### Web-Enhanced Loop Possibilities
- **React Components**: Progressive React patterns from official docs, popular libraries, and best practices
- **CSS Techniques**: Master CSS animations, layouts, and effects through web tutorials
- **Python Data Science**: Learn pandas, matplotlib, seaborn through progressive web examples
- **Machine Learning**: Implement progressively complex ML models with paper and tutorial learnings
- **API Development**: Build REST/GraphQL APIs learning from official documentation
- **Game Development**: Create progressively complex games learning from web tutorials
- **Any Domain**: Create your own spec + URL strategy for any progressive learning goal
See [WEB_ENHANCED_GUIDE.md](WEB_ENHANCED_GUIDE.md) for detailed instructions on creating your own web-enhanced loops!
## Master AI Coding
Learn to code with AI with foundational [Principles of AI Coding](https://agenticengineer.com/principled-ai-coding?y=infageloop)
Follow the [IndyDevDan youtube channel](https://www.youtube.com/@indydevdan) for more AI coding tips and tricks.
Use the best Agentic Coding tool: [Claude Code](https://docs.anthropic.com/en/docs/claude-code/overview)