# Relearning Flow

A modular skill system for AI-assisted content creation, focusing on journal entries with integrated research and audio narration.

## Overview

This repository contains OpenCode skills and agents that power an end-to-end content pipeline:

```
[Content Idea] → [Research & Draft] → [Fact-Check] → [Audio Script] → [TTS Generation] → [Published]
```

## What's Inside

### Skills (`/skills`)

| Skill | Purpose |
|-------|---------|
| **relearning-content** | Creates journal entries with the "Fallacy → Model → Protocol" structure |
| **create-script** | Condenses content ~50% and adds paralinguistic tags for natural speech |
| **voiceover** | Generates audio using Chatterbox TTS with voice cloning |

### Agents (`/agents`)

| Agent | Purpose |
|-------|---------|
| **google-search** | Research subagent for fact-checking, finding sources, and drafting content |

### Documentation (`/docs`)

- **ARCHITECTURE.md** — Full workflow diagrams and technical details

## The Philosophy

**Core Mission:** Apply enterprise-grade engineering discipline to the messy reality of being human. Debug sleep, focus, and decision-making as if they were mission-critical infrastructure.

**The Lens:** Every human problem is reframed as a **systems engineering problem**. We don't moralize; we diagnose. We don't motivate; we architect.

## Key Features

### Two-Step Audio Pipeline

Instead of basic text-to-speech, this system uses LLM intelligence:

1. **`create-script`** — Condenses articles ~50%, adds natural speech markers like `[chuckle]`, `[sigh]`, `[clear throat]`
2. **`voiceover`** — Generates TTS with voice cloning, deploys, and git pushes

### Research Integration

The `google-search` subagent:
- Verifies claims and finds supporting evidence
- Searches for opposing viewpoints
- Integrates citations directly into drafts

### Engineering Voice

All content follows a consistent voice:
- **Titles:** "The [Technical Noun]: [Subtitle]" (e.g., "The Asymptote of Effort")
- **Metaphors:** Decision fatigue → Memory leak, Habits → Cached functions
- **Structure:** Fallacy → Model → Protocol
- **Endings:** Memorable one-liners ("Stop calculating. Start retrieving.")

## Installation

### For OpenCode

1. Copy skills to your OpenCode skill directory:
   ```bash
   cp -r skills/* ~/.opencode/skill/
   ```

2. Copy agents to your OpenCode agent directory:
   ```bash
   cp -r agents/* ~/.config/opencode/agent/
   ```

3. Configure paths in each skill's SKILL.md (replace `~/projects/...` with your paths)

### Dependencies

- **voiceover skill:** Python with torch, torchaudio, chatterbox, pydub, ffmpeg
- **create-script skill:** No external dependencies (LLM-powered)
- **google-search agent:** Requires web search MCP tools

## Quick Start

1. Tell the AI: "Create a new journal entry about [topic]"
2. The system will:
   - Sync your repo
   - Draft content with research
   - Present for your review
   - Generate condensed audio script
   - Create TTS narration
   - Deploy and push

## Example Transformation

**Input (180 words):**
> Research conducted by Baumeister et al. (1998) demonstrated through a series of elegant experiments that our capacity for self-control operates much like a muscle...

**Output (95 words, 47% reduction):**
> Baumeister's research showed that willpower works like a muscle. Use it too much, and it gets tired.
> 
> [clear throat] Here's the thing. By evening, the average person has made thirty-five thousand decisions...
>
> [chuckle] Stop debugging yourself. Debug your environment.

## License

Apache-2.0

## Acknowledgments

Built with:
- [OpenCode](https://opencode.ai) — AI-powered terminal coding assistant
- [Chatterbox TTS](https://github.com/resemble-ai/chatterbox) — Expressive text-to-speech with voice cloning