Skill 安全風險審查

來源狀態

可用來源

這是第三方 Community Skill；2lus 收錄不代表已完整安全審計或保證安全。

資源類型

Skill

與類似 Skill 有什麼不同？

• Find Skills 技能探索

  Find Skills 用於探索候選技能；Skill Security 檢查候選來源與權限風險，找到不代表安全。

• Skill Lint 與安全掃描工具鏈

  SkillMD Security Linting 是可執行的掃描工具鏈；Skill Security 是審視第三方技能風險的流程，兩者都不保證安全。

來源描述的能力

這些是來源描述的可能操作，不代表 2lus 已授予權限或已測試。

• 讀取檔案
• Shell 指令

這個 Skill 是什麼？

結合靜態掃描與人工意圖檢查，辨識可疑命令、混淆字元與供應鏈風險；掃描結果是線索，不是安全證明。

可以做什麼？何時適合使用？

• 安裝新的 Community Skill 前。
• Skill 更新後出現新的 scripts 或 dependencies。
• 想知道某個指令是否傳送本機資料。

如何使用

1. 先閱讀掃描器來源與規則，再取得獲准檢查的 Skill 檔案。
2. 不執行目標 Skill，檢閱 SKILL.md、scripts、依賴與可疑資料流。
3. 將發現項目附檔案位置與人工判斷，沒有告警也要記錄檢查範圍與盲點。

使用前你需要準備

• 待審查來源與權限
• 威脅假設及不能操作的範圍

你可以替換：

[scope]、[permissions]

環境與相依需求

• Python 3 與經審查的 scanner scripts；基本掃描可離線，完整 YARA 支援為可選依賴。

使用範例與 Prompt

以下是 2lus 撰寫的示範需求；請替換為你有權處理的檔案與專案，不代表已執行或保證結果。

入門

安裝這個 Community Skill 前，唯讀檢查 SKILL.md 與 scripts 的 shell、network 和 credential access，不執行它。

實務

比較 Skill 新舊版本的 dependencies、下載網址與 Unicode 混淆字元，將每項風險附檔案位置及人工核對方法。

進階

設計離線 Skill 安全審查流程，分開規則掃描、Python AST 線索與人工意圖檢查；說明誤報、漏報與未覆蓋的 runtime 行為。

實用提醒

• 針對網路目的地和秘密來源做人工交叉核對，而非只看風險分數。

限制與注意事項

• 靜態規則可能誤報或漏報，尤其混淆、非英語內容及執行時行為；不能保證 100% 安全。

安全注意事項

• 不要為了分析而執行被審查的 Skill；scanner 本身也是需審查的第三方程式。

使用與設定

此條目不提供已確認的通用安裝指令；請依官方文件與 Agent 版本操作。

支援平台

未確認特定 Agent 相容性

來源與授權

來源查核日期（非安全認證）: 2026-09-29

MIT

相關 Skills

• 安全最佳實務
• Docker 破壞性操作防護
• Skill Creator 技能建立指南

Source status

Active source

This is a third-party community skill. Inclusion by 2lus is not a complete security audit or safety guarantee.

Resource type

Skill

How is this different from similar skills?

• Find Skills

  Find Skills discovers candidates; Skill Security examines their sources and permission risks. Discovery is not a safety verdict.

• Skill Lint & Security Scan

  SkillMD Security Linting is an executable scanning toolchain; Skill Security is a third-party skill review workflow. Neither guarantees safety.

Documented capabilities

These are operations described upstream, not permissions granted or tested by 2lus.

• Read files
• Shell commands

What is this skill?

Combine static scanning with human intent review to identify suspicious commands, confusable characters and supply-chain risks. Findings are leads, not proof of safety.

Use cases and when to use it

• Review a new community skill.
• Inspect added scripts or dependencies in an update.
• Determine whether commands transmit local data.

How to use it

1. Review the scanner and rules before collecting authorized target files.
2. Inspect instructions, scripts, dependencies and data flows without executing the target skill.
3. Record file locations, human judgments and blind spots even when there are no alerts.

What you need

• Source and permissions to review
• Threat assumptions and prohibited actions

You can replace:

[scope], [permissions]

Environment and dependencies

• Python 3 and reviewed scanner scripts; baseline scans can be offline, with full YARA support optional.

Usage and prompt examples

These example requests were written by 2lus. Substitute files and projects you may use; examples are not executed results or guarantees.

Beginner

Before installing this community skill, inspect its instructions and scripts for shell, network and credential access. Execute nothing.

Practical

Compare old and new dependencies, download destinations and Unicode confusables, citing file locations and manual checks for each risk.

Advanced

Design offline review combining rule scanning, Python AST clues and human intent analysis. Explain false positives, false negatives and unobserved runtime behavior.

Tips

• Manually cross-check network destinations and secret sources instead of relying on a score.

Limitations

• Static rules can miss obfuscation, non-English patterns and runtime behavior or produce false alarms; they cannot guarantee safety.

Security notes

• Never execute the target skill to inspect it; the scanner itself is third-party code needing review.

Usage and setup

No verified universal installation command is provided for this entry. Follow the official documentation for your agent version.

Supported agents

Specific agent compatibility unknown

Sources and license

Source check date (not a safety certification): 2026-09-29

MIT

Related skills

• Security Best Practices
• Docker Destructive Guardrails
• Skill Creator