Speckit Swarm

# Speckit Swarm - Agent Orchestration System

## Overview

Native implementation of oh-my-opencode-style orchestration using OpenClaw's tools.

## Architecture

### Core Components

1. **Ultrawork Detector** - Detects "ulw"/"ultrawork" keywords and triggers parallel execution
2. **Agent Personas** - Specialized system prompts for different tasks
3. **Task Planner** - Breaks complex tasks into parallel chunks
4. **Continuation Enforcer** - Ensures tasks complete fully

## File Structure

```
skills/speckit-swarm/
├── SKILL.md                      # This file
├── src/
│   ├── ultrawork.ts              # Ultrawork detection & trigger
│   ├── personas/
│   │   ├── mod.ts               # Persona exports
│   │   ├── sisyphus.ts          # Main orchestrator
│   │   ├── hephaestus.ts        # Deep worker
│   │   ├── oracle.ts            # Design/debug
│   │   ├── librarian.ts         # Research/docs
│   │   └── explore.ts            # Fast scout
│   ├── planner.ts               # Task decomposition
│   └── index.ts                 # Main entry
```

## Usage

### Manual Mode
```bash
# Use personas directly
sessions_spawn task:"..." model:"minimax-m2.5" thinking:"high"
```

### Ultrawork Mode
When user includes "ulw" or "ultrawork":
1. Detect keyword
2. Decompose task into parallel chunks
3. Execute with parallel_spawn
4. Aggregate results

## Personas

### Sisyphus (Main Orchestrator)
- Model: minimax-m2.5
- Thinking: high
- Behavior: Relentless execution, parallel coordination, todo tracking

### Hephaestus (Deep Worker)
- Model: minimax-m2.5
- Thinking: high
- Behavior: Autonomous execution, no hand-holding, completes full scope

### Oracle (Design/Debug)
- Model: minimax-m2.5
- Thinking: high
- Behavior: Architecture decisions, bug hunting, code review

### Librarian (Research)
- Model: minimax-m2.1
- Thinking: medium
- Behavior: Docs lookup, code exploration, pattern finding

### Explore (Scout)
- Model: minimax-m2.5-highspeed
- Thinking: low
- Behavior: Fast grep, file...