菜单
Reviews an implemented change for over-engineering — unnecessary new files and classes, single-caller abstractions, duplication of existing code, unrelated edits, speculative scope. Use after the production code is written and BEFORE any test is written. Report findings only, never edit.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/mollie/Shopware6 --skill review-minimal-diff命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Bring the Shopware version matrix of the CI, PR and nightly pipelines up to date with the versions Shopware has actually released, and check every version against the available dockware/shopware images. Use when asked to update the Shopware versions, the pipeline matrix or the test matrix, to check for new Shopware releases, or when a pipeline fails because a dockware image is missing — "update shopware versions", "neue Shopware Version", "Matrix aktualisieren", "check for new Shopware releases", "gibt es neue Shopware Releases".
Bring the Shopware version matrix of the CI, PR and nightly pipelines up to date with the versions Shopware has actually released, and check every version against the available dockware/shopware images. Use when asked to update the Shopware versions, the pipeline matrix or the test matrix, to check for new Shopware releases, or when a pipeline fails because a dockware image is missing — "update shopware versions", "neue Shopware Version", "Matrix aktualisieren", "check for new Shopware releases", "gibt es neue Shopware Releases".
Set the release version of the Mollie plugin before it goes to internal QA. Bumps composer.json and the MolliePayments::PLUGIN_VERSION constant, and turns the "# Unreleased" changelog section into the new version heading in both languages. Use when asked to increase, bump or set the plugin version, to prepare a release or hand over to QA — "increase version", "bump the version", "release 5.5.0", "prepare the next release", "Version erhöhen", "neue Version".
基于 SOC 职业分类
正在显示 SKILL.md
| name | review-minimal-diff |
| description | Reviews an implemented change for over-engineering — unnecessary new files and classes, single-caller abstractions, duplication of existing code, unrelated edits, speculative scope. Use after the production code is written and BEFORE any test is written. Report findings only, never edit. |
Follow .agents/guidelines/reviewing.md for the shared protocol, then this checklist.
AGENTS.md sections 1 and 2 give the layout and the minimal-change rule. This review only
removes — never propose a change that adds code.
Goal: the smallest honest solution to the problem, reviewable in a few minutes.
Could this have been a parameter? For every new method, class, interface or file: name the existing place it could have lived and say why it does not fit. A new file without that justification is a finding.
Single-caller abstraction. Any interface, factory, strategy, event, subscriber, config option or feature flag with exactly one caller and no second caller in sight. Finding.
Duplication.
Search shopware/ for code that already does this — a service, a struct, a helper, a
calculation. Grep for the domain nouns in the diff, not just for exact strings. If a
near-duplicate exists, that is a finding even if the new code is nicer.
Unrelated changes. Renames, reformatting, import reordering, comment edits, whitespace, or fixes to code the task did not ask about. Each is a finding — it belongs in a separate change.
Spread. Many files touched for a small behavioural effect: would one change at a lower level have done it? Say it concretely — "these 6 call-site edits collapse into one default in X".
Speculative scope. Backwards-compatibility shims, deprecation paths, extra config, null-safety for cases that cannot occur, handling of inputs that no caller produces. Finding.
Dead weight. Code, constants, parameters or struct fields added but never read. Finding.
Order findings by how much code the fix removes.
