| name | ds-comment-review |
| description | Run a strict review of code comments — flag the comments that don't earn their place and tighten the rest to the discipline. Language-agnostic. Reports a findings list by default; `--fix` applies the edits in place. |
| disable-model-invocation | true |
When invoked, audit the comments in scope against one question: does each comment earn its place, and is it as short as it can be? The discipline: comments are for humans and explain WHY, not WHAT, in domain terms a reader with only the code can follow — one line by default, only where the reason isn't obvious, never restating the code and never pointing into planning artifacts (decision IDs, roadmap/step/spec references); a long comment is rare and signals importance. This is not /ds-deslop (branch-diff, matches existing style) — this command's only lens is comment discipline, it works on any scope, and it imposes the standard regardless of the codebase's existing comment habits. Comment-only and behavior-preserving: never change code logic.
Arguments
- Positional args are scope (files, directories, globs, or the whole codebase). With no scope, review comments in the code changed on the current branch.
--full → review comments across the whole codebase instead of just the branch's changes. Explicit positional scope still wins; --full only replaces the no-scope default.--fix → apply the edits in place (delete the noise, tighten the verbose, strip cruft, keep the genuine ones) instead of only reporting.
What to flag
- Restating code — comments that narrate what the code plainly does (
// increment i, // loop over users). Delete.
- Obvious / ceremonial — section banners,
// constructor, getter/setter narration, docstrings that just repeat the signature. Delete.
- Planning-artifact references — pointers into the project's planning trail: decision IDs (
D-12), DECISIONS.md/roadmap/step/spec-requirement citations ("as per step 3"), author-stamped notes, a commit message pasted as a comment. A comment must stand on the code alone — these rot by design as housekeeping archives and renumbers those artifacts. Strip.
- Buried WHY — a genuine reason wrapped in five lines of prose. Tighten to one line.
- Drifted — comments describing behavior the code no longer has. Treat as a correctness issue: fix or delete.
- Keep — and respect — the rare legitimately-long comment: a non-obvious algorithm, a subtle invariant, a hard-won workaround. Don't cut these; they are the signal. Make sure they still read as important.
Doc-comments: where idiomatic (Go doc comments, Python docstrings on public API, Rust ///), keep them — but short and to the point, not prose. Don't strip required public-API docs; don't demand docs on the obvious.
Output
Default (report): a prioritized findings list. For each: anchor to file:line, quote the comment, name why it fails (restates / obvious / cruft / verbose / drifted), and give the fix — delete, or tighten to: "<one-liner>". Group by delete / tighten / drifted (correctness) / kept-as-important. A short high-conviction list beats a long pedantic one.
With --fix: apply the edits — delete the noise, tighten the verbose to one line, strip the cruft, preserve and (if needed) sharpen the genuine ones — then give a concise 1–3 sentence summary of what changed. Never touch code logic. After applying, confirm the resulting diff touches only comment lines — if any non-comment line changed, that's a bug: revert it.
Rules:
- Comment-only, behavior-preserving. Never change code.
- Impose the discipline regardless of the codebase's existing comment style — do not match bad patterns (this is the point of the command).
- Keep the rare important long comment; don't flatten everything to one line.
- Change nothing else. In report mode, the output is the list.
