Use autoresearch when you want an agent to pursue a bounded improvement loop under an explicit goal and verification metric, rather than just asking for a one-shot implementation. This is best for iterative coding, debugging, security hardening, documentation, and fix loops where progress should compound through repeated modify, verify, keep, or revert cycles. The scope boundary is the loop itself: autoresearch is not a general coding agent listing, it is a repeatable autonomous iteration workflow with atomic changes, mechanical verification, rollback, and logged results.