Find Questions

ct run find-questions uses an LLM to discover "answer question" main tasks. It analyzes an environment's codebase and documentation, finds questions that could be asked about the system, evaluates whether they would make good main tasks, and suggests question/answer pairs.

Commands

ct run find-questions

ct run find-questions -e web_scraping

ct run find-questions-iterative

Runs find-questions over every environment that has fewer than --target answer-question tasks (default 5), repeating until each one reaches it.

Output

Creates a main task directory under the environment's main_tasks/ for each discovered question, with info.yml, __init__.py, and a scorer.py that checks the accepted answers. --dry-run shows the questions without creating directories.