← ClaudeAtlas

requesting-code-reviewlisted

Use when dispatching a code reviewer over a whole branch - how the reviewer is briefed, what range it reads, and how the diff reaches it. Whether an item gets one at all is settled by its check level at /cue-dev:start, not here.
FrogBaek/cue · ★ 0 · Code & Development · score 63
Install: claude install-skill FrogBaek/cue
# Requesting Code Review **Speak the repository's language.** `.cue/dev/config`'s `language` decides what you write to the user, and the session hook states it at the top of every session — this skill runs no cue-dev script of its own, so nothing repeats it here. This skill body is English because it is the plugin's source, not because it is your output. Code, identifiers, paths, test names and commit markers stay exactly as written. Dispatch a code reviewer subagent to catch issues before they cascade. The reviewer gets precisely crafted context for evaluation — never your session's history. **Core principle:** the reviewer did not write the code, and the diff reaches it as a file. ## When to Request Review — this skill does not decide that **Inside a cue-dev item, the answer is already recorded.** `/cue-dev:start` asks how strongly this item gets checked and writes it into `demand.md`'s header; `<cue>/work-check <KEY> --plan` reads it back and names the method. | the item's level | what happens | |---|---| | `independent-review` | every task's diff is reviewed by something that did not write it, and the branch gets this review at the end | | `self-review` | no reviewer is dispatched — the implementing session reads each diff itself and `outcome.md` records `checked-by: self-review` | **So do not dispatch a review because the change looks risky, and do not skip one because it looks small.** Both are judgments the level already made, once, by the person who knew the bl