scan

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Codebase Scan

代码库扫描

Audit the codebase against another skill's criteria using a parallel agent team.

使用并行Agent团队，对照另一项Skill的标准审计代码库。

Workflow

工作流程

1. Parse args & load skill

1. 解析参数并加载Skill

Extract the skill name from the args passed to this skill.

If no skill name provided, list available skills in
```
.claude/skills/
```
and ask the user which to scan against.
If skill doesn't exist, list available skills and tell the user.

Read

.claude/skills/<name>/SKILL.md

plus any files in

references/

and

rules/

subdirectories.

Distill the skill's content into a numbered criteria checklist: a flat list of concrete, testable rules labeled C1, C2, C3, etc. Each criterion should be a single sentence describing what to check for.
If the skill has no evaluable code criteria (e.g., workflow-only skills like
```
why
```
that don't define code patterns or rules), tell the user it's not scannable and stop.

从传递给本Skill的参数中提取Skill名称。

如果未提供Skill名称，列出
```
.claude/skills/
```
目录下的可用Skill，并询问用户要对照哪项进行扫描。
如果指定的Skill不存在，列出可用Skill并告知用户。
读取
```
.claude/skills/<名称>/SKILL.md
```
文件，以及
```
references/
```
和
```
rules/
```
子目录下的所有文件。
将Skill的内容提炼为编号的标准检查清单：一份扁平化的具体可测试规则列表，标记为C1、C2、C3等。每条标准应为描述检查内容的单句。
如果该Skill没有可评估的代码标准（例如像
```
why
```
这类仅涉及工作流、未定义代码模式或规则的Skill），告知用户该Skill不可扫描并终止流程。

2. Discover relevant files

2. 发现相关文件

Use the skill's criteria to infer file scope:

React/frontend criteria →
```
app/**/*.tsx
```
,
```
app/**/*.ts
```
Backend criteria →
```
services/**/*.ts
```
General/mixed → both of the above
CSS/styling →
```
app/**/*.css
```
,
```
app/**/*.tsx
```

Always exclude:

node_modules/

dist/

*.test.*

*.spec.*

**/migrations/**

**/*.d.ts

, generated files.

Count candidate files. If zero, tell the user and stop.

根据Skill的标准推断文件范围：

React/前端标准 →
```
app/**/*.tsx
```
,
```
app/**/*.ts
```
后端标准 →
```
services/**/*.ts
```
通用/混合标准 → 上述两者
CSS/样式标准 →
```
app/**/*.css
```
,
```
app/**/*.tsx
```

始终排除：

node_modules/

dist/

*.test.*

*.spec.*

**/migrations/**

**/*.d.ts

以及生成的文件。

统计候选文件数量。如果数量为0，告知用户并终止流程。

3. Plan team composition

3. 规划团队组成

Split files into chunks by directory subtree so no file is assigned to two agents:

Candidate files	Teammates
< 100	2 (or skip team for < 20)
100–500	3
500+	4–5, split by top-level directory

For very small scans (< 20 files), skip the team — scan inline and jump to step 6.

按目录子树将文件拆分为多个区块，确保没有文件被分配给多个Agent：

候选文件数量	团队成员数
< 100	2个（若少于20个文件则可跳过团队）
100–500	3个
500+	4–5个，按顶层目录拆分

对于非常小的扫描任务（少于20个文件），跳过团队组建——直接进行内联扫描并跳至步骤6。

4. Spawn scan team

4. 启动扫描团队

```
spawnTeam
```
with name
```
scan-<skill-name>
```
Create one
```
TaskCreate
```
per chunk with:
- Subject:
```
Scan <directory-area> against <skill-name> criteria
```
- Description: Include the full criteria checklist, the file scope (glob patterns), and the teammate instructions from
```
references/teammate-instructions.md
```
Spawn teammates as
```
general-purpose
```
subagent_type (they need Read, Glob, Grep + team communication)
Name teammates
```
scanner-1
```
,
```
scanner-2
```
, etc.
Assign tasks via
```
TaskUpdate
```
with
```
owner
```

调用
```
spawnTeam
```
创建名为
```
scan-<skill名称>
```
的团队
为每个区块创建一个
```
TaskCreate
```
任务：
- 主题：
```
扫描<目录区域>以对照<skill名称>标准
```
- 描述：包含完整的标准检查清单、文件范围（glob模式）以及
```
references/teammate-instructions.md
```
  中的团队成员说明
以
```
general-purpose
```
子代理类型启动团队成员（他们需要具备Read、Glob、Grep权限以及团队通信能力）
为团队成员命名为
```
scanner-1
```
、
```
scanner-2
```
等
通过
```
TaskUpdate
```
为任务分配
```
owner
```
（负责人）

5. Collect & synthesize

5. 收集与整合

Wait for all teammates to complete their tasks. Each teammate reports findings in structured format:

FINDING: C<n> | SEVERITY | file/path.ts:LINE | Description

Collect all findings from teammate messages.

等待所有团队成员完成任务。每个成员需以结构化格式报告发现的问题：

FINDING: C<n> | 严重程度 | file/path.ts:行号 | 描述

从团队成员的消息中收集所有发现的问题。

6. Generate report

6. 生成报告

Create

.claude-scan/<skill-name>.md

using the format in

references/report-format.md

Key sections:

Executive Summary (2-3 sentences)
Criteria Evaluated (table with violation counts)
Findings by Severity (Critical → Warning → Info tables)
Patterns Observed
Statistics
Recommended Fix Order (batched by non-overlapping file groups)

使用

references/report-format.md

中的格式创建

.claude-scan/<skill名称>.md

文件。

关键章节：

执行摘要（2-3句话）
已评估的标准（包含违规次数的表格）
按严重程度分类的发现问题（从严重→警告→信息的表格）
观察到的模式
统计数据
推荐修复顺序（按不重叠的文件组批量划分）

7. Cleanup & present

7. 清理与展示

If a team was spawned: send
```
shutdown_request
```
to all teammates, then
```
cleanup
```
Display inline summary: total findings by severity, top violated criteria, scan scope
Tell the user the full report is at
```
.claude-scan/<skill-name>.md
```
Ask: "Want me to spawn a fix team to address these findings?"

如果启动了团队：向所有团队成员发送
```
shutdown_request
```
，然后执行
```
cleanup
```
操作
展示内联摘要：按严重程度统计的总问题数、违规最多的标准、扫描范围
告知用户完整报告位于
```
.claude-scan/<skill名称>.md
```
询问：“是否需要我启动修复团队来解决这些问题？”

8. Fix team (if user says yes)

8. 修复团队（若用户同意）

Read the report's "Recommended Fix Order" section
Group fixes into non-overlapping file batches
```
spawnTeam
```
with name
```
fix-<skill-name>
```
Create tasks per batch — each task description includes:
- The specific violations to fix (from the report)
- The relevant criteria definitions
- Instructions from
```
references/fix-team-instructions.md
```
Spawn
```
general-purpose
```
teammates named
```
fixer-1
```
,
```
fixer-2
```
, etc. (need Edit/Write)
Assign tasks and wait for completion
Shutdown and cleanup fix team
Ask user if they want to re-scan to verify fixes

读取报告中的“推荐修复顺序”章节
将修复任务按不重叠的文件组进行分组
调用
```
spawnTeam
```
创建名为
```
fix-<skill名称>
```
的团队
为每个批次创建任务——每个任务描述包含：
- 需修复的具体违规问题（来自报告）
- 相关的标准定义
- ```
references/fix-team-instructions.md
```
  中的说明
以
```
general-purpose
```
类型启动名为
```
fixer-1
```
、
```
fixer-2
```
等的团队成员（需要具备Edit/Write权限）
分配任务并等待完成
关闭并清理修复团队
询问用户是否要重新扫描以验证修复效果