- name
SKILL.md
技能文件保留在发布者的仓库中。
在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add OyadotAI/oya-browser --skill code-review -a cursor -y在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add OyadotAI/oya-browser --skill code-review -a claude-code -y在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add OyadotAI/oya-browser --skill code-review -a codex -y在终端中粘贴复制的命令。它会把这个 SKILL.md 安装到对应的智能体中。
npx skills add OyadotAI/oya-browser --skill code-review -a grok -y在 Grok Bot 聊天中发送复制的消息。Bot 会从 GitHub 获取这个 SKILL.md,并添加到你所有 Bot 共享的技能中。
Install this skill: https://github.com/OyadotAI/oya-browser/tree/HEAD/.agents/skills/code-review来自发布者
Review two distinct questions: whether the change follows the project's documented standards, and whether it implements the requested behavior. Perform both passes within the assigned role. Delegate only when explicitly authorized and the assigned tools support delegation. Return the report to the coordinator; remote comments or issue publication require existing authorization.
Read the assignment and identify the requested files or comparison. For branch work, resolve the supplied ref and inspect the appropriate merge-base diff and commit list. For an exact commit comparison, use the requested endpoints. For work in progress, inspect unstaged and staged changes and relevant untracked files; a committed-only diff misses that assignment. Name the range reviewed, and report invalid refs or empty scope plainly.
Read applicable repository instructions, surrounding code, callers, and tests. Include CONTRIBUTING.md and CODING_STANDARDS.md when present. Recover acceptance criteria from the user's request, supplied spec, or relevant local issue material. Read linked remote sources through Oya Browser when access is authorized. Use the current request as the spec when no formal document exists. State missing requirements as coverage limits rather than inventing them or requiring tracker setup.
Apply documented rules in their proper scope and account for established exceptions. Trace correctness, ownership, compatibility, failure handling, and concurrency through real callers. Skip mechanical nits already enforced by tooling. Use references/design-signals.md only to guide investigation of maintainability; a code smell alone does not establish a defect.
Support each blocking finding with a reachable case or a concrete documented requirement, source location, impact, and a small repair. Keep optional maintainability advice separate and include it only when requested. Respect a correctness-only review's scope.
Map each requirement to implementation and evidence. Look for missing or partial behavior, incorrect interpretation, scope growth, and apparent success that fails on cancellation, retry, disconnect, restart, or duplicate input. Inspect affected callers beyond the edited hunk. Distinguish a demonstrated defect from an unresolved question and from a preexisting issue.
Lead with the verdict and covered scope. Use separate Standards and Spec sections, each with its result, supported findings, and coverage limits. For each finding give severity, file/line, triggering input or state, expected and actual behavior, evidence, and smallest repair. Cross-reference a shared underlying defect instead of counting it twice. An empty findings list is valid; do not manufacture findings to fill a section.
List exact checks executed and their results. A passing check supports only the behavior it exercises. Write or run tests only within the user's authorization. Preserve implementation unless a repair was assigned. Use the team's findings asset when acting as reviewer; fit these two passes into it without duplicating the report.
Adapted from Matt Pocock's engineering skills; see SOURCE.md and LICENSE.
This skill should be used when reviewing a branch, pull request, working tree, or changes since a commit for repository standards and the user's requested behavior.
---name: code-reviewdescription: This skill should be used when reviewing a branch, pull request, working tree, or changes since a commit for…---# .agents/skills/code-review/SKILL.md常见问题
从发布者处下载 .agents/skills/code-review/SKILL.md,或复制路径后打开仓库。PlugMCP 只收录该技能,不托管文件。
发布者将它存放在 https://github.com/OyadotAI/oya-browser。
Claude、Cursor、Codex 和 Grok Build 可以读取你从发布者仓库添加的 SKILL.md。在聊天中发送技能的 GitHub 文件夹链接,Grok Bot 就会安装它。