Recursive-Task-Synthesis is a synthetic recursive task dataset specifically designed for reinforcement learning and capability evaluation of command-line AI agents, aiming to provide researchers with executable and validated real-world task examples. This dataset contains 37,484 validated command-line task instances, each with natural language instructions, reference answers, task source data, and a pre-configured Docker runtime environment. The tasks cover multiple technical fields such as software development, system operation and maintenance, and script writing.

Dataset composition:

• metadata/tasks.parquet: Task instance index
• metadata/shard_manifest.jsonl: TAR packet size and SHA256 checksum record
• data/tasks-*.tar: Cleaned and de-identified complete executable task packages