codereview-clean

codereview-clean

Companion to codereview-inline. Removes the embedded review trail when the user is done with it. Closed threads can stay in source indefinitely; this skill only runs on explicit invocation.

Trigger

/codereview-clean <path>, optionally with a mode flag. <path> can be a file, directory, or . for the whole repo.

Modes

Flag	Behavior
(default)	Remove closed comment blocks (@SEEN @AI ...) and their adjacent @AI-reply: comments. Leave still-open @AI threads alone.
--all	All of the above, plus remove still-open @AI comments (use when the user has decided to drop a pending question).
--replies-only	Remove only @AI-reply: comments. Keep user comments intact, including @SEEN markers — useful when the user wants to preserve their own questions as inline documentation.

Processing flow

1. Resolve <path>. Refuse /, ~, paths containing .. that escape the repo, or anything outside the current repo root.
2. List candidate files under the path (skip binary, generated, lockfile, vendored — same filters as codereview-inline).
3. For each file, identify comments matching the active mode's target set. Comment recognition is language-aware via the same filetype map as codereview-inline.
4. Build the edited file in memory:
   • For closed-thread removal: drop the entire comment block containing @SEEN @AI, since that comment exists only as part of the review trail.
   • For @AI-reply: removal: drop the entire reply comment block.
   • For --all open-marker removal: drop the entire user comment block containing @AI (even without @SEEN).
5. Collapse stray blank lines: if removal leaves two or more consecutive blank lines, collapse to one. Don't leave dangling indentation.
6. Write each file once, only if changed.
7. Print summary: Removed N comments across M files, with a per-file count.
8. Run no git commands beyond reads. User commits the cleanup whenever they want.

Edge cases

• Comment block mixes review tokens with user prose (e.g. // @SEEN @AI TODO: revisit this later): default mode treats the whole comment as review trail and removes it. If the user wants to keep surrounding prose, they should hand-edit before running clean, or use --replies-only.
• Multi-line @AI-reply: reply: remove the whole comment block, not just the first line.
• Marker appears inside a string despite the comment-context filter: skip. Same safety check as codereview-inline.
• File contains only review markers, nothing else: still safe; result is an emptier file with no special handling needed.
• Last line of file is a removed comment: trim trailing whitespace/blank lines after removal.

Boundaries

• Never touch code outside of comment blocks containing the target tokens.
• Never run without an explicit <path> argument.
• Never auto-commit.
• Never remove a comment that doesn't contain any of @AI, @SEEN, or @AI-reply:.