Competitive Programming Agent

Turn draft contest ideas into ready-to-judge problem packages. The agent takes a rough statement and returns a refined statement, IO specification, constraints, test generator, validator, reference solution, and more.

Features

• Python CLI – feed a draft statement and receive Markdown + JSON output
• Web UI – paste the draft in an interactive interface, configure your API key, and generate results
• Mock mode – experiment without an API key using bundled sample output
• Configurable backend – supply your own OpenAI-compatible base URL, model and organization

Installation

pip install -e .

Create a .env file or export environment variables with your API credentials:

OPENAI_API_KEY=sk-your-key
CP_AGENT_MODEL=gpt-4o-mini
OPENAI_BASE_URL=https://api.openai.com/v1

CLI usage

cp-agent path/to/draft.txt --output package.md

Arguments:

• --model – override the model per run
• --tags tag1 tag2 – hint topical tags
• --difficulty – target difficulty label
• --notes – additional free-form instructions
• --mock-response samples/mock_response.json – bypass API calls for dry-runs

The CLI writes a Markdown summary to --output (or stdout) and a matching JSON file next to it.

Web app

flask --app app run --debug

Open http://127.0.0.1:5000. Fill in the form, provide an API key (or use the mock response), and hit Generate Package.

Development

• Formatters: black and isort
• HTTP client: httpx
• Front-end is a single static page (templates/index.html) with vanilla JS (static/app.js)

Contributions and issue reports are welcome!