code-review-mode
Use for a code review of a branch or code area, an architecture review, a test audit, a thermonuclear review, or a sweep for bugs and unfinished code outside any diff, and whenever writing or changing tests.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- kind
- dev
- name
- code-review-mode
- description
- Use for a code review of a branch or code area, an architecture review, a test audit, a thermonuclear review, or a sweep for bugs and unfinished code outside any diff, and whenever writing or changing tests.
02 资源文件 14
- 参考文档playbooks/architecture-review.md6.0 KB
- 参考文档playbooks/sweep.md2.7 KB
- 参考文档playbooks/test-audit.md7.3 KB
- 参考文档playbooks/thermo-quality-review.md12.1 KB
- 参考文档references/architecture-review/ADR-FORMAT.md2.7 KB
- 参考文档references/architecture-review/codebase-design.md6.0 KB
- 参考文档references/architecture-review/CONTEXT-FORMAT.md2.2 KB
- 参考文档references/architecture-review/DEEPENING.md2.5 KB
- 参考文档references/architecture-review/DESIGN-IT-TWICE.md2.6 KB
- 参考文档references/architecture-review/HTML-REPORT.md6.5 KB
- 许可证references/architecture-review/LICENSE1.0 KB
- 参考文档references/routing-cases.md3.3 KB
- 参考文档references/test-audit/CAMPAIGN.md5.4 KB
- 许可证references/test-audit/LICENSE1.1 KB
03 SKILL.md 正文
Code review mode
Route by the objects under review. Combined reviews return one plan.
| Playbook | Aliases | Reviews |
|---|---|---|
test-audit |
Tests, and production code that exists only for tests | |
thermo-quality-review |
thermonuclear, thermo-nuclear-code-quality-review |
Production code: structure, file size, branching, types, and layering |
architecture-review |
improve-codebase-architecture |
Module shape across an area: depth, seams, and locality |
sweep |
Code that no diff touches: bugs, hard-coded values, and unfinished code |
Pick the route
Take the first rule that fits:
- A test is being written or changed outside a review. Apply the authoring gate in
test-audit, then return to the calling task. No review plan. - A name follows the mode, as in
code-review-mode ; test-audit. Run that playbook alone, through its own procedure and report. Compare names with case, spaces, hyphens, and underscores ignored; an alias counts as its playbook's name. If no name matches, list the choices and stop. - The request limits the review to tests, to code quality, or to a sweep. A test audit, including a sweep of tests, runs
test-auditalone. A thermonuclear review runsthermo-quality-reviewalone. Any other sweep runssweepalone. Each follows its own procedure and report. - Any other review follows the steps below.
For every review route, find the target below before loading a playbook. It replaces the playbook's default scope. Reviews are read-only unless the request asks for changes, including changes to domain documentation.
Before running a review, announce the chosen playbooks, those skipped, and why.
1. Find the target
- A branch: its changes since it left the default branch, plus uncommitted changes.
- An area the request names, such as a directory or a module: the code in that area.
- An architecture review of a branch: the modules the branch touches.
- A sweep: the area the request names, or the whole repository when it names none, even on a branch with changes.
- No changes and no area: ask which area to review, and wait for the answer.
2. Pick the playbooks
Choose one playbook per object present in the target:
test-auditfor tests and test supportthermo-quality-reviewfor production codearchitecture-reviewwhen the target is an area or the request names architecture; all three review that area, including its existing tests and production code
A target with only documentation or configuration has nothing to review: say so and stop.
3. Run the reviews
Give each chosen playbook its own subagent, with the playbook's path, the target, and the instruction to return findings without editing. For this combined review, architecture-review stops at candidates in text. Without subagents, run the playbooks one after another. The step is complete when every chosen playbook has returned its findings.
4. Return one plan
Order the plan's PRs as they must be implemented, each with its reason:
test-auditfirst: a refactor is proven only by tests it leaves untouched.architecture-reviewsecond: the chosen candidate reshapes the modules on that test net.thermo-quality-reviewlast: polish spent on code the architecture PR removes is wasted.
Include only PRs with work to do. Each lists its findings and the check that proves it. When architecture-review returns candidates, recommend one and ask which to build. Mark findings in code a candidate would remove as depending on that candidate. In the architecture PR, delete an old test only after recording the candidate evidence that test-audit requires. Before the quality PR, rerun thermo-quality-review on the code the architecture PR leaves. When the target has almost no tests, the first PR adds behavior tests at the outer boundary, under the authoring gate.
The plan is the deliverable. When the request asks for changes, build one PR at a time in the plan's order.
Use routing cases when changing a playbook name, an alias, or these rules.
04 同仓库更多 Skills
writing-great-skills
Use when creating, modifying, evaluating a skill.
writing-for-agents
Use when creating, editing, or reviewing a skill, AGENTS.md, CLAUDE.md, or another document agents read.
why
Use for questions about design rationale, the history of regressions or incidents, or the evidence behind thresholds and tradeoffs. Use `how` for runtime behavior. Do not use for plain commit or date lookups.
verify-video-archive
Use when validating the macOS or Linux archive workflow reached by `just convert-video`, including real media, prerequisites, terminal progress, source safety, locking, and recovery.
verify-transcript
Use when validating transcript CLI behavior, locating its verification features, or running the paid YouTube end-to-end check.
verify-skills
Use when verifying `just compile-skills`, `just remote-skills`, `just install-skills`, or `just skills-discover` in the skills repository, such as after changing their scripts or adding, renaming, or moving a skill.