Getting Started¶

Prerequisites¶

Python 3.12+
UV package manager

Installation¶

cd 5_ai_open_claw/pd_arena
uv sync

Setup¶

1. Run Migrations¶

uv run python manage.py migrate

2. Seed Default Agents¶

uv run python manage.py seed_agents

This creates:

5 LLM persona agents (Cooperative, Tit-for-Tat, Selfish, Forgiving, Random)
6 canonical policy agents (ALLC, ALLD, TFT, GRIM, GTFT, WSLS)
5 Random(α) agents (α = 0.0, 0.25, 0.5, 0.75, 1.0)

3. Start the Server¶

uv run python manage.py runserver 8080

Open http://localhost:8080.

First Experiment¶

1. Browse Agents¶

Go to Agents in the navigation. You'll see a gallery of the 16 seeded agents — 5 LLM personas and 11 policy agents.

2. Chat with an Agent¶

Click on any LLM agent (e.g., "Cooperative Agent"). On the detail page, use the chat panel to talk to it:

"How would you approach a repeated game where your opponent might defect?"

The agent responds in character based on its persona. This helps verify the backstory makes sense before using it in experiments.

3. Create an Experiment¶

Go to Experiments → New Experiment.

Name it: "My First PD Experiment"
Set parameters: 50 rounds, 5 replicates, fixed horizon
Add conditions: Drag "Cooperative Agent" into Agent A slot, drag "Always Defect" into Agent B slot
Add more: Drag "Tit-for-Tat Agent" vs "Always Defect" as a second condition
Click Save as Draft or Run Experiment

4. View Results¶

After the experiment completes, the results page shows:

Summary cards: Total games, average cooperation rates
Cooperation over time chart: Line plot per condition
Condition breakdown: Click into any condition for replicate details
Round-by-round: Click into any game for action-by-action data

Running Tests¶

uv run pytest -q
# 63 passed in 0.41s

Project Structure¶

pd_arena/
├── agents/       # Agent management
├── games/        # Experiment design & execution
├── llm/          # LLM providers & prompts
├── configs/      # YAML presets
├── templates/    # Base layout
├── static/       # CSS, JS
└── fixtures/     # Default data