Review code comments

Review comments in the scoped code with a skeptical independent pass. Remove narration, stale explanations, commented-out code, workaround sermons, and suppressions that hide fixable problems. Preserve comments that carry information code cannot express.

This skill concerns comments in source files. It never reads or writes human review threads, issue comments, chat messages, or other correspondence. It must never post findings or contact anyone.

Use a host-available general subagent with references/reviewer-prompt.md rather than a named custom agent. Subagents may communicate with each other, but no agent may communicate with a person. Repeat that sentence verbatim in every child prompt. If delegation is unavailable, run the same prompt as a separate local review pass before editing.

Model preferences are optional. Read repository-documented model configuration or .agents/agent-models.md when present, and map roles only to models the current host confirms are available. Otherwise omit model selection and inherit the parent or host defaults. Never require the file or hard-code unconfirmed model identifiers. When engineering-mode is installed, match each subagent's reasoning profile to the job by its Delegation rule; otherwise use the configured profile.

Scope and authority

Use the files or diff named by the caller. Otherwise use the smallest current diff implied by context. Local comment edits are allowed only when the user's request authorizes changes. When installed, safe-refactor is authoritative for broader behavior-preserving restructuring. In either case, keep this skill inside comment cleanup and small obvious encodings; report any broader work as a local follow-up instead of expanding scope.

Keep only proven exceptions

Keep: