Sweep the code for cleanliness with a panel of subagents and fix it — comments that narrate history/process or restate the code, redundant/excessive comments, duplicated code, and best-practice nits — then re-verify iteratively until clean. Protects genuinely useful why/units/invariant comments. Use for "clean up the comments", "remove duplicated code", "no useless/excessive comments", "tidy this up", or a pre-merge polish pass.
Iteratively review the current branch/changes with a panel of subagents — several of them blind (no prior knowledge of what changed) to resist bias — adversarially verify every finding, fix all genuine issues regardless of who wrote them, and repeat until a pass is clean. Use for "review this branch", "iterative review", "review until no issues", pre-PR hardening, or when the user asks for a thorough multi-agent code review.