← ClaudeAtlas

code-reviewlisted

Load when a task needs a deep, report-only review across independent Standards and Spec axes, using bounded read-only native Subagents when available and adding risk lenses only when the diff warrants them.
JasonxzWen/harness-hub · ★ 72 · Code & Development · score 72
Install: claude install-skill JasonxzWen/harness-hub
# Code Review Use this Skill for a substantive pre-delivery review. It reviews the current change; it does not own implementation, verification, Git delivery, or remote state. ## Establish Scope Resolve a credible base, changed and untracked files, the accepted spec or task contract, relevant project instructions, and nearby tests. Do not switch a shared checkout. If intent or base cannot be established from repository facts, report the missing evidence instead of reviewing an invented diff. ## Two Required Axes Review these independently so one kind of evidence does not mask the other: ### Standards axis - repository instructions and local conventions - correctness, robustness, observability, and error handling - test value and public-behavior coverage - maintainability, unnecessary complexity, and duplication ### Spec axis - every accepted behavior and non-goal - edge cases, ownership, authorization, and data boundaries - whether implementation and tests demonstrate the requested outcome - contradictions between the diff, current implementation, and durable docs When the Host supports native parallel work, dispatch both axes as bounded, independent, read-only native Subagents. Give each the same base, diff, intent, and evidence boundary. Subagents do not edit, commit, publish, decide user tradeoffs, or own the final verdict. The main Agent integrates and deduplicates their evidence. If native Subagents are unavailable or the diff is tiny, the main Agent applies b