← ClaudeAtlas

review-itlisted

Standalone implementation-review skill in the Agentic Engineering pipeline. Reviews a worktree, branch, or PR against a substantive, evidence-backed checklist and reports verified findings, a scoped clean result, or explicit limitations — never fixing what it finds. Callable standalone against any worktree, branch, or PR with no prior implement-it session or plan-it-authored issue required; also callable by implement-it before Gate 1 and during an authorized delivery correction. Trigger to review implemented code, review a branch or PR, or verify a correction before it's relied on. Reports only — it never fixes application code, applies formatting fixes, commits, pushes, approves a gate, merges, or mutates GitHub or other live/production state; every correction returns to implement-it. Does not own guide review, issue/plan-synthesis review, investigation discipline, or commit-plan review — those stay with document-it, plan-it, lab-it, and implement-it.
elieandraos/agentic-engineering · ★ 0 · Code & Development · score 70
Install: claude install-skill elieandraos/agentic-engineering
# review-it ## What this skill is `review-it` is an independently callable implementation-assurance skill. It reviews a worktree, branch, or PR against a substantive, evidence-backed checklist and reports verified findings, a scoped clean result, or explicit limitations where evidence or diagnostics weren't available. It does not fix what it finds — every correction returns to `implement-it`. It is portable: it runs the same way whether or not `implement-it` or `plan-it` were ever involved. A worktree with no issue, a branch with no plan, or a PR opened outside this ecosystem entirely are all valid, ordinary targets. ## Entry points The same capability, reached from three different places — not three procedures: - **Standalone.** A human asks directly: review this branch, this PR, my current changes. - **Before Gate 1.** `implement-it` invokes this skill once implementation and its own verification are complete, before stopping at Gate 1 (`implement-it/rules/review-gates.md`). - **During an authorized delivery correction.** `implement-it` invokes this skill the same standalone way before that correction's own Gate 1/Gate 2 pass, when performing a correction `ship-it` has handed it (`ship-it/rules/ci-failure-correction.md`'s "CI failure on an open milestone PR"). No entry point requires another skill's context. A standalone invocation proceeds with whatever scope evidence is actually available (see `rules/scope.md`) — it does not wait for, or invent, an `impleme