此报告尚无所请求语言的翻译，当前显示规范英文报告。

版本化安全评估

报告编号: SA-A5D66E69

10/11/2026, 2:34:18 PM

anatomy 安全评估 v1

Skill 安全认证报告

审计历史

扫描器版本 3.0.0 审计模型: codex 最新发布的报告

技能名称: anatomy
版本: v1
维护者: wheresryan22
来源: github.com/wheresryan22/anatomy/tree/dbf697a527bbf7d358668e1557cb1e35ff725d46/
覆盖范围: 91 已扫描文件 · 57,606 分析行数
策略版本: skillstore-security-audit-policy-v1

已确认问题的最高严重性

中

有 4 个已确认的安全问题需要处理。

安装上下文

请查看当前 Skill 详情页

本页仅汇总报告证据。Skill 详情页提供 canonical install advisory。

打开当前 Skill 详情页

本报告不会阻止或授权 manifest 或 ZIP。

Most presented matches are benign geometry, documentation, local screenshots, and loopback browser control; no evidence found of prompt injection or intentional exfiltration. CLI-selected imports execute host code, and browser checks relax local-file isolation. Three oversized files and 2,518 omitted matches remain review gaps; this audit does not establish complete repository safety. Static review was capped at 400/2918 representative findings; omitted static matches are unconfirmed, so automatic publishing stays disabled until manual review.

报告位置

证据链

证据从来源绑定延伸到安装契约。证据可用于验证，但不是安全保证。

来源

已绑定 commit 与路径
构件

已绑定内容与 tree hash
审计

已完成
安装契约

打开 manifest 进行验证
打开 manifest

观察到的能力

“已观察到”表示本报告记录了支持证据；“未记录”不证明该能力不存在。

包含脚本

可能执行 Skill 随附的代码。

在 24 个证据位置观察到

网络访问

可能连接外部服务。

在 10 个证据位置观察到

文件系统访问

可能读取或写入本地文件。

在 48 个证据位置观察到

环境变量

可能读取进程环境中的值。

在 3 个证据位置观察到

外部命令

可能调用 Skill 之外的命令或程序。

在 50 个证据位置观察到

能力复核项 (2)

中

scripts/turn-bench.mjs:20

Dynamic import() expression

const mod = await import(pathToFileURL(resolve(argv[0])).href);

The CLI-selected build module executes with Node filesystem and network privileges. An untrusted build target can run arbitrary host code.

中

scripts/turn-check.mjs:486

Dynamic import() expression

const { T, P } = await import(pathToFileURL(resolve(target)).href);

The CLI-selected build module executes with Node filesystem and network privileges. An untrusted build target can run arbitrary host code.

风险问题

已确认的安全问题与仍需审核的项目分开展示。

已确认安全问题 (4)

RISK-001 中

scripts/capture.mjs:101-110 scripts/drive.mjs:112 scripts/turn-check.mjs:38 SECURITY.md:5

Browser Checks Enable Cross-File Access

Browser checks enable --allow-file-access-from-files for user-selected pages. A malicious local page can access other readable local files; temporary profiles do not isolate filesystem access.

All three browser launchers enable cross-file access and accept local pages. SECURITY.md acknowledges this trust requirement; no malicious use was observed.

RISK-002 低

examples/raptor-engine/raptor-engine-light.html:1

Unscanned file (too_large) — manual review required

[unscanned: too_large]

The analyzer skipped this oversized executable HTML file. Its full contents remain unreviewed, so this coverage gap is retained without alleging malware.

RISK-003 低

examples/raptor-engine/raptor-engine.html:1

Unscanned file (too_large) — manual review required

[unscanned: too_large]

The analyzer skipped this oversized executable HTML file. Its full contents remain unreviewed, so this coverage gap is retained without alleging malware.

RISK-004 低

examples/turning-dial/turning-stress-light.html:1

Unscanned file (too_large) — manual review required

[unscanned: too_large]

The analyzer skipped this oversized executable HTML file. Its full contents remain unreviewed, so this coverage gap is retained without alleging malware.

补救措施

本次审计记录了建议的修复措施。实施这些措施是维护者的责任。

FIX-001
中
Static review capped

Manually review the omitted 2518 static analyzer matches or reduce bundled generated/vendor/reference content before enabling automatic publication.
FIX-002
中
Browser checks relax local-file isolation.

Remove cross-file access by default. Serve assets from a restricted local directory, or require explicit consent and filesystem isolation for file-mode checks.
FIX-003
中
Verification imports execute user-selected modules with host privileges.

Require trusted build targets and explicit execution consent. Run unknown modules in a sandbox without credentials, unrestricted filesystem access, or network access.
FIX-004
低
Oversized files and capped static results leave incomplete security coverage.

Review all three skipped HTML files and the 2,518 omitted matches. Provide reproducible builds and smaller reviewable artifacts before automatic publication.

专家证据

不可变主体身份、扫描器元数据、已排除匹配项和源码级证据。

Artifact 主体

Marketplace 提交: a5d66e6915301b999741c6b4437f9eacacc9af0d
内容哈希: 8798bed1ba9b1cc5d0e6f77a9613ab61a707ce3fb7fc0750d20b7c0c3ca1d893
目录树哈希: c5094cb108cda96d41403fd91ea4be03d364a73098a1afea8408904e763998c4
Skill 路径: skills/wheresryan22/anatomy
审计载荷哈希: 57e9c2a675f3b2477b5a978cf6500a43

分析元数据

审计模型: codex

分析状态: 已完成

范围仅限于已记录的文件、行数、方法和证据；本报告不声称进行了运行时或沙箱执行。

验证与导出

JSON 报告已签名审计证明 CITATION.cff SARIF 2.1.0 已签名安装清单来源

manifest 和 lockfile 将安装 artifact 绑定到加密哈希。该完整性声明独立于安全评估。

审计证明: active