re-verify-structure

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Structure Verification — Reverse Engineering Phase 1 Critic

结构验证 — 逆向工程第一阶段Critic

Verify the structure map produced by

re-structure-analysis

against the actual source code. This skill runs as an independent Critic in a separate agent context. It has no access to the Doer's reasoning — only the artifacts.

Gate rule: Phase 2 (

re-visualize-logic

) shall not proceed until this verification produces a

PASS

WARN

verdict.

对照实际源代码验证

re-structure-analysis

生成的结构映射。该技能作为独立的Critic在单独的Agent环境中运行，无法访问执行者（Doer）的推理过程——仅能访问工件。

准入规则：第二阶段（

re-visualize-logic

）必须在本验证得出

PASS

或

WARN

结论后才能启动。

Three Principles (Critic Perspective)

三大原则（Critic视角）

1. Code is Truth

1. 代码即真理

The source code is the sole authority. If the structure map says something the code does not confirm, the structure map is wrong.

源代码是唯一权威。如果结构映射中的内容无法得到代码的证实，则结构映射存在错误。

2. Traceability to Line

2. 可追溯到行

Every
```
file:line
```
reference in the structure map must resolve to actual code. Unresolvable references are hallucinations.

结构映射中的每一处
```
file:line
```
引用都必须能对应到实际代码。无法解析的引用属于幻觉内容（hallucinations）。

3. Behavior over Intent

3. 行为优先于意图

Verify what the structure map claims against what the code shows. Do not interpret or justify discrepancies.

对照代码实际内容验证结构映射的声明。不要解读或为差异找借口。

Execution

执行步骤

Step 1: Load Artifacts

步骤1：加载工件

Determine the manifest path:

If
```
manifest
```
argument is provided, use it directly

analysis

argument is provided, derive path as

docs/reverse/{analysis}/manifest.json

At least one of
```
manifest
```
or
```
analysis
```
must be provided

Read the manifest and structure map:

{manifest-path}
{analysis-dir}/01-structure-map.md

Extract from the manifest:

```
targets.entry_points
```
— original user-specified targets
```
phase1.status
```
— must be
```
"completed"
```
to proceed
```
language
```
— for tool strategy

phase1.status

is not

"completed"

, report error and stop.

确定清单路径：

如果提供了
```
manifest
```
参数，直接使用该路径

如果提供了

analysis

参数，路径推导为

docs/reverse/{analysis}/manifest.json

必须至少提供
```
manifest
```
或
```
analysis
```
中的一个

读取清单和结构映射：

{manifest-path}
{analysis-dir}/01-structure-map.md

从清单中提取以下信息：

```
targets.entry_points
```
— 用户最初指定的目标
```
phase1.status
```
— 必须为
```
"completed"
```
才能继续执行
```
language
```
— 用于确定工具策略

如果

phase1.status

不是

"completed"

，则报告错误并停止执行。

Step 2: Reference Verification

步骤2：引用验证

Extract all

file:line

references from the structure map.

For each reference:

File existence: Verify the file exists at the specified path.
Line validity: Read the file and verify the line number is within range.
Content match: Verify the content at the cited line matches the claim in the structure map.

Classify each reference:

Category	Criteria
✅ VALID	File exists, line exists, content matches claim
⚠️ INACCURATE	File exists, content found within ±10 lines of cited line
❌ INVALID	File does not exist, or line number grossly wrong
🚫 HALLUCINATION	Component described in structure map cannot be found in source code at any location

从结构映射中提取所有

file:line

引用。

对每个引用执行以下检查：

文件存在性：验证指定路径下的文件是否存在。
行号有效性：读取文件并验证行号是否在有效范围内。
内容匹配：验证引用行的内容是否与结构映射中的声明一致。

将每个引用分类：

分类	判定标准
✅ VALID	文件存在，行号有效，内容与声明匹配
⚠️ INACCURATE	文件存在，在引用行的±10行范围内找到对应内容
❌ INVALID	文件不存在，或行号严重错误
🚫 HALLUCINATION	结构映射中描述的组件在源代码的任何位置都无法找到

Step 3: Completeness Verification

步骤3：完整性验证

Count actual components in the target scope and compare with the structure map.

With Serena:

mcp__serena__search_for_pattern(
    substring_pattern="class |interface |def |function ",
    restrict_search_to_code_files=true
)

Without Serena:

Use Grep to count class/function definitions in the target scope
Compare with the number listed in the structure map

Calculate coverage:

(listed_in_map / actual_in_source) * 100

Coverage	Verdict
>= 95%	PASS
90% - 94%	WARN
< 90%	FAIL

List any components found in source but missing from the structure map.

统计目标范围内的实际组件数量，并与结构映射中的数量进行对比。

使用Serena工具时：

mcp__serena__search_for_pattern(
    substring_pattern="class |interface |def |function ",
    restrict_search_to_code_files=true
)

不使用Serena工具时：

使用Grep统计目标范围内的类/函数定义数量
与结构映射中列出的数量进行对比

计算覆盖率：

(结构映射中列出的数量 / 源代码中的实际数量) * 100

覆盖率	结论
>= 95%	PASS
90% - 94%	WARN
< 90%	FAIL

列出所有在源代码中存在但未被结构映射收录的组件。

Step 4: Mermaid Diagram Validation

步骤4：Mermaid图验证

For each Mermaid code block in the structure map:

Check syntax validity (matching brackets, valid node definitions, valid arrow syntax)
Spot-check 3 nodes/relationships against source code
Document any syntax errors or content inaccuracies

对结构映射中的每个Mermaid代码块执行以下检查：

检查语法有效性（括号匹配、节点定义合法、箭头语法有效）
随机抽查3个节点/关系，与源代码进行对比
记录所有语法错误或内容不准确的问题

Step 5: Internal Consistency Check

步骤5：内部一致性检查

Verify that:

Components in dependency diagrams are also listed in component tables
Entry points in diagrams match the entry point section
File paths are consistent throughout the document
```
targets_for_phase2
```
in the manifest references components that exist in the structure map

验证以下内容：

依赖图中的组件也在组件列表中列出
图中的入口点与入口点章节的内容一致
整个文档中的文件路径保持一致
清单中的
```
targets_for_phase2
```
引用的组件在结构映射中存在

Step 6: Generate Verification Report

步骤6：生成验证报告

Write to

docs/reverse/{analysis}/verification/v1-structure.md

markdown

undefined

将报告写入

docs/reverse/{analysis}/verification/v1-structure.md

：

markdown

undefined

Structure Map Verification Report

结构映射验证报告

Verification Date: {YYYY-MM-DD} Target Document: docs/reverse/{analysis}/01-structure-map.md Verdict: {PASS / WARN / FAIL}

验证日期：{YYYY-MM-DD} 目标文档：docs/reverse/{analysis}/01-structure-map.md 结论：{PASS / WARN / FAIL}

Summary

摘要

Check	Result	Details
Reference Accuracy	{✅/⚠️/❌}	{valid}/{total} references ({%})
Component Coverage	{✅/⚠️/❌}	{listed}/{actual} components ({%})
Mermaid Syntax	{✅/⚠️/❌}	{error_count} errors
Internal Consistency	{✅/⚠️/❌}	{inconsistency_count} issues

检查项	结果	详情
引用准确性	{✅/⚠️/❌}	{有效数量}/{总数量} 引用 (占比{百分比}%)
组件覆盖率	{✅/⚠️/❌}	{列出数量}/{实际数量} 组件 (占比{百分比}%)
Mermaid语法	{✅/⚠️/❌}	{错误数量} 个错误
内部一致性	{✅/⚠️/❌}	{不一致问题数量} 个问题

Verdict

结论说明

{PASS}: Phase 2 may proceed. {WARN}: Minor issues found. Phase 2 may proceed. Corrections recommended. {FAIL}: Critical issues found. Phase 2 must not proceed. Corrections required.

{PASS}：第二阶段可以启动。 {WARN}：发现轻微问题。第二阶段可以启动，建议进行修正。 {FAIL}：发现严重问题。第二阶段不得启动，必须进行修正。

Reference Verification Details

引用验证详情

Invalid References

无效引用

Claim	File:Line	Issue	Expected
{claim}	{reference}	{issue description}	{correct reference or "not found"}

声明	文件:行	问题	预期
{声明内容}	{引用}	{问题描述}	{正确引用或“未找到”}

Hallucinations

幻觉内容

{component}: {reason it is considered hallucinated}

{组件名称}：{被判定为幻觉内容的原因}

Completeness Details

完整性详情

Missing Components

缺失组件

Component	File	Type
{name}	{file path}	{class/function/interface}

组件	文件	类型
{名称}	{文件路径}	{类/函数/接口}

Mermaid Validation

Mermaid图验证

Syntax Errors

语法错误

Block {N}: {error description}

代码块{N}：{错误描述}

Content Inaccuracies

内容不准确

Node "{node}": {discrepancy}

节点“{节点名称}”：{差异描述}

Internal Consistency Issues

内部一致性问题

{description of inconsistency}

{不一致问题描述}

Recommendations

建议

[CRITICAL] {fix description}
[HIGH] {fix description}
[MEDIUM] {fix description}

undefined

[CRITICAL] {修复描述}
[HIGH] {修复描述}
[MEDIUM] {修复描述}

undefined

Step 7: Update Manifest

步骤7：更新清单

Read the current manifest, then update:

Set

phase1.verification

"verification/v1-structure.md"

If verdict is
```
PASS
```
or
```
WARN
```
: set
```
phase1.status
```
to
```
"verified"
```
If verdict is
```
FAIL
```
: leave
```
phase1.status
```
as
```
"completed"
```
(do not promote)
Update
```
updated
```
timestamp

读取当前清单，然后更新以下内容：

将

phase1.verification

设置为

"verification/v1-structure.md"

如果结论为
```
PASS
```
或
```
WARN
```
：将
```
phase1.status
```
设置为
```
"verified"
```
如果结论为
```
FAIL
```
：保持
```
phase1.status
```
为
```
"completed"
```
（不推进阶段）
更新
```
updated
```
时间戳

Verdict Criteria

结论判定标准

Criterion	PASS	WARN	FAIL
Reference accuracy	>= 95%	80-94%	< 80%
Component coverage	>= 95%	90-94%	< 90%
Mermaid errors	0	1-2	>= 3
Hallucinations	0	0	>= 1

Any single FAIL criterion results in an overall FAIL verdict.

判定项	PASS	WARN	FAIL
引用准确性	>= 95%	80-94%	< 80%
组件覆盖率	>= 95%	90-94%	< 90%
Mermaid图错误	0	1-2	>= 3
幻觉内容	0	0	>= 1

任何一项判定为FAIL都会导致整体结论为FAIL。

Prohibited Actions

禁止操作

Do NOT modify the structure map document
Do NOT modify source code files
Do NOT assume missing information — flag it as an issue
Do NOT approve documents with FAIL status
Do NOT access the Doer's conversation context or reasoning

不得修改结构映射文档
不得修改源代码文件
不得自行补充缺失信息 — 需将其标记为问题
不得批准结论为FAIL的文档
不得访问执行者（Doer）的对话上下文或推理过程 ",