審計紀錄

版本	日期	結果	相對上一版變化
v8 最新	2026年7月5日上午10:22	無已確認發現	能力無變化
v7	2026年7月5日上午10:22	無已確認發現	外部命令
v6	2026年6月29日上午09:18	1 已確認	外部命令
v5	2026年1月17日上午12:02	無已確認發現	能力無變化
v4	2026年1月17日上午12:02	無已確認發現	外部命令
v3	2026年1月10日下午01:06	無已確認發現	能力無變化
v2	2026年1月10日下午01:06	無已確認發現	能力無變化
v1	2026年1月10日下午01:06	無已確認發現	基線

2026年7月5日上午10:22

The static backtick findings are Markdown fences, inline examples, and routine npm test snippets, not executable Ruby backticks. The reconnaissance findings are ordinary TDD prose about tests passing and reading the bundled anti-patterns reference. No prompt injection, exfiltration intent, or malicious automation was found.

2

已掃描檔案

672

分析行數

1

複核項目

0

已忽略的誤報

審計單位: codex

審計版本 7

2026年7月5日上午10:22

The static backtick findings are Markdown fences, inline examples, and routine npm test snippets, not executable Ruby backticks. The reconnaissance findings are ordinary TDD prose about tests passing and reading the bundled anti-patterns reference. No prompt injection, exfiltration intent, or malicious automation was found.

2

已掃描檔案

672

分析行數

1

複核項目

0

已忽略的誤報

審計單位: codex

審計版本 6

2026年6月29日上午09:18

The static analyzer flagged many Markdown code fences and common words as external commands, weak cryptography, and reconnaissance. Review found these pattern hits are false positives because the skill contains documentation and test examples, not executable scripts or network logic. A medium data-integrity concern remains because the skill repeatedly instructs agents to delete code when TDD was not followed.

2

已掃描檔案

672

分析行數

1

複核項目

3

已忽略的誤報

已確認的安全問題 (1)

中

SKILL.md:37-43 SKILL.md:288 SKILL.md:340

Destructive Workflow Guidance

The skill tells agents to delete implementation code and start over when tests were not written first. This supports TDD discipline, but it can cause data loss if followed without user confirmation or version-control checks.

The destructive instruction is explicit and repeated in the main skill file. The risk is operational rather than malicious because the surrounding context is TDD process guidance.

已忽略的靜態誤報 (3)

這些靜態命中已被語義複核判定為誤報，或只命中了 schema 定義中的詞；此處保留顯示以提高透明度，但不影響品質評分。

低

False Positive External Command Detections

The external command detections occur inside Markdown fences, diagrams, TypeScript examples, and illustrative test commands. No evidence found of executable scripts, command construction, or user-controlled shell execution.

All reviewed locations are static text or examples in Markdown. There is no runnable Ruby, shell wrapper, or instruction to execute arbitrary commands.

低

SKILL.md:3 SKILL.md:203 SKILL.md:204 SKILL.md:267 SKILL.md:347 SKILL.md:349 testing-anti-patterns.md:67 testing-anti-patterns.md:69 testing-anti-patterns.md:70 testing-anti-patterns.md:76 testing-anti-patterns.md:88 testing-anti-patterns.md:94

False Positive Weak Cryptography Detections

The weak cryptography detections map to prose or ordinary test example text, not cryptographic functions. No evidence found of MD5, SHA-1, DES, custom encryption, or password hashing logic.

The cited lines contain descriptions, table text, or sample cleanup method names. The file has no cryptographic implementation or dependency use.

低

SKILL.md:178 SKILL.md:335 SKILL.md:359 testing-anti-patterns.md:255

False Positive Reconnaissance Detections

The reconnaissance detections are references to testing concepts such as mocks and setup complexity. No evidence found of host discovery, port scanning, network probing, or system inventory collection.

The exact lines are part of a TDD checklist or anti-pattern documentation. They do not request network access, environment enumeration, or system commands.

審計單位: codex

審計版本 5

2026年1月17日上午12:02

This skill contains only markdown documentation with TDD guidelines and code examples. No executable code, scripts, network calls, filesystem access, or external command execution capabilities. Pure prompt-based documentation safe for publication.

3

已掃描檔案

864

分析行數

1

複核項目

0

已忽略的誤報

風險因素

⚙️ 外部命令 (57)

審計單位: claude

審計版本 4

2026年1月17日上午12:02

This skill contains only markdown documentation with TDD guidelines and code examples. No executable code, scripts, network calls, filesystem access, or external command execution capabilities. Pure prompt-based documentation safe for publication.

3

已掃描檔案

864

分析行數

1

複核項目

0

已忽略的誤報

風險因素

⚙️ 外部命令 (57)

審計單位: claude

審計版本 3

2026年1月10日下午01:06

This skill contains only markdown documentation with TDD guidelines and code examples. No executable code, scripts, network calls, filesystem access, or external command execution capabilities. Pure prompt-based documentation safe for publication.

2

已掃描檔案

672

分析行數

0

複核項目

0

已忽略的誤報

此次已完成的稽核未記錄已確認的安全發現。

審計單位: claude

審計版本 2

2026年1月10日下午01:06

This skill contains only markdown documentation with TDD guidelines and code examples. No executable code, scripts, network calls, filesystem access, or external command execution capabilities. Pure prompt-based documentation safe for publication.

2

已掃描檔案

672

分析行數

0

複核項目

0

已忽略的誤報

此次已完成的稽核未記錄已確認的安全發現。

審計單位: claude

審計版本 1

2026年1月10日下午01:06

This skill contains only markdown documentation with TDD guidelines and code examples. No executable code, scripts, network calls, filesystem access, or external command execution capabilities. Pure prompt-based documentation safe for publication.

2

已掃描檔案

672

分析行數

0

複核項目

0

已忽略的誤報

此次已完成的稽核未記錄已確認的安全發現。

審計單位: claude

版本對比

風險因素

風險因素

已確認的安全問題 (1)

風險因素

風險因素