Interactive Puzzle Server + Agent

Overview

• A minimal Flask-based puzzle server exposing simple levels and a validation endpoint.
• An interactive CLI agent that lists levels, fetches descriptions, accepts attempts, and asks Ollama for hints.

Setup

• Python 3.10+
• Docker installed and running (required for script-based puzzles)
• Ollama installed and running locally
• Model llama3.2 pulled

It is recomended to use a virtual environment

python3 -m venv venv
source venv/bin/activate

1. Install dependencies:

python3 -m pip install -r requirements.txt

Run

2. Start Ollama service

ollama serve
ollama pull llama3.2
ollama list

3. Start the server:

python3 server.py

4. In another terminal, run the agent:

python3 agent.py

Server is hosted in http://127.0.0.1:5050

Notes and security

• Some levels validate by string comparison.
• Script-based levels run inside a Docker container with network disabled, dropped capabilities, and limited CPU/memory.
• Docker must be installed, the daemon must be running, and the user running the server must be allowed to run docker.
• For production, use stronger sandboxing, authentication, and persistent progress storage.

Design

• The server exposes /levels, /level/<id>, and /submit.
• For script-based levels, POST /submit accepts JSON { "level_id": "5", "files": { "answer.c": "<source>" } } and returns test output.
• The agent uses Ollama (llama3.2) to produce contextual hints; it instructs the model not to reveal flags.

Next steps

• Add more levels with staged tasks and progressive hints.
• Implement an interactive web UI.
• Add secure sandbox execution for C compilation and run (via Firecracker, gVisor, or chrooted containers).