🔮 PredictFish

Predict project success using MiroFish swarm intelligence.

Simulate market reactions, customer behavior, and competitive dynamics before writing a single line of code. PredictFish connects to MiroFish (a swarm intelligence engine) to run multi-agent simulations of your product idea in the market.

Why This Exists

From AGENTS.md:

Rent Reminder Pro: 6시간 개발 후 근본 결함 발견 (세입자 Telegram 안 씀)
랜딩 카피 먼저 보여줬으면 5분 만에 발견했을 것

PredictFish adds quantitative prediction to the landing-page sanity check. Before PM starts development:

1. Write landing page copy
2. Run PredictFish simulation ← this skill
3. Show Director: copy + prediction score
4. Director approves or kills the project

How It Works

Project Info → Seed Document → MiroFish API → Swarm Simulation → Prediction Report

MiroFish API Flow:

1. Build knowledge graph from seed
2. Prepare simulation environment
3. Run N rounds of multi-agent simulation
4. Generate prediction report

Prediction Output:

• Success score (0-100)
• Risk factors
• Market reaction summary
• Improvement suggestions

Usage

1. Check MiroFish Status

uv run scripts/predict.py status

2. Run Prediction

uv run scripts/predict.py predict \
  --name "Rent Reminder Pro" \
  --description "Telegram bot that reminds tenants to pay rent on time" \
  --target "Property managers in Korea" \
  --market "5000+ property managers, competitors use email/SMS" \
  --rounds 20 \
  --output results.json

3. Retrieve Saved Report

uv run scripts/predict.py report <report_id>

OpenClaw Integration

When PM receives a new project request:

# Generate landing page copy first (existing process)
# Then run prediction:
cd ~/ubik-collective/systems/ubik-pm/skills/predictfish
uv run scripts/predict.py predict \
  --name "$PROJECT_NAME" \
  --description "$DESCRIPTION" \
  --target "$TARGET_CUSTOMER" \
  --market "$MARKET_INFO" \
  --output ~/ubik-collective/systems/ubik-pm/memory/prediction-$(date +%Y%m%d-%H%M%S).json

# Show Director: landing copy + prediction score
# Wait for Go/No-Go

Requirements

• Python 3.11+ (via uv)
• MiroFish API running on http://localhost:5001
  • Typically via Docker: docker-compose up mirofish
• Environment Variable (optional):
  • MIROFISH_URL - override default API endpoint

Architecture

predictfish/
├── SKILL.md              # This file
├── scripts/
│   └── predict.py        # Main CLI
├── lib/
│   ├── mirofish_client.py   # API client
│   ├── seed_generator.py    # Project → seed document
│   └── report_parser.py     # Report → prediction score
├── templates/
│   └── seed_template.md     # Seed document template
├── README.md
└── pyproject.toml

Example Output

🔮 PredictFish: Rent Reminder Pro
📍 MiroFish API: http://localhost:5001