autorl-retrieve-evidence

Provide a clarified description of your reinforcement learning task, and the retrieval process gathers papers whose methods and environments match that task. The result is a grounded evidence set rather than generic keyword search results.

Task-grounded evidence is prior work—papers, baselines, and code—selected because it directly relates to your specific RL task, environment, and objectives. It contrasts with broad literature surveys that may not transfer to your setting.

Baselines should come from published methods evaluated on tasks similar to yours. This Skill identifies candidate baselines from retrieved evidence so comparisons in your workflow reflect established performance references.

Run evidence retrieval after the task is clarified but before algorithm selection and implementation. Decisions about architectures, reward design, and evaluation protocols are more defensible when grounded in retrieved prior work.

Retrieval quality depends on how precisely the task is clarified; vague task descriptions yield loosely related evidence. Retrieved papers and code still require manual review to confirm relevance and correctness.