菜单
Use when reviewing a pull request, branch, or local diff and the reader needs an evidence-backed, plain-language HTML guide to the change. Apply for self-review, reviewing another author's change, large cross-system diffs, or when a dependency-ordered file-and-unit review is needed.
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/ankitvashisht12/argus --skill argus-review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。
基于 SOC 职业分类
正在显示 SKILL.md
| name | argus-review |
| description | Use when reviewing a pull request, branch, or local diff and the reader needs an evidence-backed, plain-language HTML guide to the change. Apply for self-review, reviewing another author's change, large cross-system diffs, or when a dependency-ordered file-and-unit review is needed. |
Build a guided review, not a list of generic warnings. The output helps a reader understand the changed system before inspecting code.
crew to create durable review records and plan review scale. An Argus invocation authorizes this use. The Crew planner may assign one reviewer or a parallel review run.flow to render the PR-level reading journey.explain-it to render linked file pages and code-on-demand unit views.~/artifacts/argus-review/<date>-<repository>-<target>/ with htmlfiles/, review-model.json, README.md, contract.md, progress.md, and log.md.gh to resolve its repository, target ref, and default base, then create a fresh temporary checkout outside the source repository. For a local branch, use its repository only as a source and create a separate temporary checkout. Do not modify the user’s checkout.The planner reads the diff and surrounding code, then records: change size, changed-file coverage, subsystem buckets, dependencies, reading order, review scale, and safe validation commands.
Use one reviewer for a small, contained change. Use Crew for a large or cross-system change: a planner maps coverage, reviewers examine distinct buckets, and a final lead verifies and consolidates results. Record why that choice was made.
Order readers by understanding, not file name:
Dependencies override this order. Group files into meaningful domain buckets, then link later files to the prerequisite file or unit.
Assign stable IDs and links to every PR, bucket, file, and unit.
Every unit states what changed and why it matters. Add a reviewer note only when it is a real finding or a human judgement that cannot be safely or cheaply verified.
Check objective claims in the temporary checkout before reporting them: resolve imports and symbols, inspect callers and types, run focused lint, type, or test commands when the planner marks them safe. Omit routine successful checks.
Report a verified failure with location, plain-language impact, and evidence. If a check was not run because it is unsafe, expensive, or needs product judgement, state the exact human decision instead of a vague caution.
Use Flow for the index page that explains the PR intent, cautions, buckets, and recommended reading order. Use Explain It for linked file pages in the same site. For Argus, the plain-language-first requirement overrides Explain It's visible-diff default: reveal code, commands, and technical evidence only through explicit Inspect controls.
Keep tests concise and Supporting files grouped in one compact section. Link all references directly to the cited file or unit. Verify every cross-page and in-page link, then visually inspect and open htmlfiles/index.html.