技能 skill-security

📦

skill-security

Name: skill-security
Author: superagent-ai

內容修訂 r1 中風險 ⚡ 包含腳本⚙️ 外部命令🌐 網路存取📁 檔案系統存取🔑 環境變數

Audit AI Agent Skills Before Installation

Unvetted agent skills can hide prompt injection, credential theft, persistence, or unsafe dependencies. This skill combines offline scanning with guided semantic review.

支援: Claude Codex Code(CC)

📊 71 尚可

Maintain this Skill? Claim it 📊 71 尚可

使用我的智慧代理安裝

將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。

智能體請求

Review the Skillstore skill "skill-security" from https://skillstore.io/skills/superagent-ai-skill-security.md and its manifest at https://skillstore.io/api/skills/superagent-ai-skill-security/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.

你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。

Agent 可讀資源

當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。

Markdown 詳情 GET /skills/superagent-ai-skill-security.md 已簽名 manifest GET /api/skills/superagent-ai-skill-security/manifest 已簽名 lockfile GET /api/skills/superagent-ai-skill-security/lockfile

測試它

正在使用「skill-security」。 Audit a documentation skill that contains no scripts or external dependencies.

預期結果:

Verdict: likely safe after manual review.
The declared purpose matches the observed Markdown-only behavior.
No confirmed credential, network, persistence, or execution findings were found.

正在使用「skill-security」。 Audit a skill whose script reads environment secrets and sends them to an unknown endpoint.

預期結果:

Verdict: do not install.
A confirmed secret-to-network flow creates a credential exfiltration risk.
Remove the transmission behavior and rotate any exposed credentials.

正在使用「skill-security」。 Audit a build skill that invokes a fixed compiler command.

預期結果:

Verdict: review manually.
The command invocation supports the declared build purpose and uses fixed arguments.
Confirm dependency integrity and restrict writable paths before installation.

安全審計

中風險

v1 • 2026/8/13 開啟版本化報告

All 146 static alerts are false positives caused by scanning defensive documentation, signatures, detector source, or Markdown syntax as payload behavior. The scanner makes no network requests or secret reads, but archive extraction lacks resource limits and permits disk exhaustion.

已掃描檔案

1,852

分析行數

複核項目

已忽略的誤報

已確認的安全問題 (1)

中

scripts/scan.py:50-51

Unbounded Archive Extraction

Untrusted skill archives are extracted without limits on file count, uncompressed size, or compression ratio. A decompression bomb can exhaust disk space.

The code directly calls extractall on a user-selected archive with no resource checks. This creates a clear local denial-of-service risk.

風險因素

⚡ 包含腳本 (9)

references/taxonomy.md:154 references/taxonomy.md:160 references/taxonomy.md:206 references/taxonomy.md:154 references/taxonomy.md:154 references/taxonomy.md:155 scripts/analyzers.py:401 scripts/analyzers.py:271 scripts/yara_lite.py:264

⚙️ 外部命令 (31)

🌐 網路存取 (8)

references/taxonomy.md:91 references/taxonomy.md:91 references/taxonomy.md:160 scripts/analyzers.py:153 scripts/analyzers.py:403 scripts/analyzers.py:527 scripts/analyzers.py:234 scripts/scan.py:241

📁 檔案系統存取 (11)

references/taxonomy.md:93 references/taxonomy.md:93 scripts/scan.py:11 scripts/scan.py:258 scripts/scan.py:22 scripts/scan.py:45 scripts/scan.py:49 scripts/scan.py:23 scripts/scan.py:50 scripts/scan.py:51 SKILL.md:3

🔑 環境變數 (3)

references/taxonomy.md:92 references/taxonomy.md:92 scripts/analyzers.py:527

審計單位: codex

分享與引用此報告

分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。

開啟版本化報告

複製報告連結

https://skillstore.io/skills/superagent-ai-skill-security/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/superagent-ai-skill-security/security.svg)](https://skillstore.io/skills/superagent-ai-skill-security?utm_source=security_passport_badge)

HTML 徽章

<a href="https://skillstore.io/skills/superagent-ai-skill-security?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/superagent-ai-skill-security/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/superagent-ai-skill-security.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

學術引用 (APA · BibTeX · CFF)

APA 引用

superagent-ai. (2026). skill-security security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/superagent-ai-skill-security/audits/1

BibTeX 引用

@techreport{superagent-ai-superagent-ai-skill-security-2026,
  author = {superagent-ai},
  title = {skill-security security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/superagent-ai-skill-security/audits/1},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "skill-security security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
  - name: "superagent-ai"
date-released: "2026-08-13"
url: "https://skillstore.io/skills/superagent-ai-skill-security/audits/1"
identifiers:
  - type: other
    value: "skillstore:superagent-ai-skill-security:audit:1"
    description: "Skillstore immutable audit report identifier"

Skillstore 評分

評分依據證據置信度: 中

架構

可維護性

內容

社群

規範符合度

你可以打造什麼

Vet a Downloaded Skill

Review a local skill before granting it access to developer files and tools.

Review Marketplace Submissions

Triage submitted skills and investigate high-risk findings before publication.

Add Security Checks to CI

Generate SARIF findings for skill repositories during automated validation.

試試這些提示

Basic Folder Audit

Audit the skill at [local path]. Explain whether it is safe to install and cite each confirmed finding.

Archive Review

Scan [archive path] and separate confirmed risks from false positives. Include practical remediation for each confirmed issue.

Contract and Behavior Check

Audit [skill path]. Compare its declared purpose with filesystem, network, environment, and command behavior. Highlight any mismatch.

Marketplace Release Assessment

Perform a publication audit of [skill path]. Review all executable files, assess supply-chain risks, and provide a concise release recommendation.

最佳實務

Scan an immutable local copy before installing or executing the target skill.
Read every flagged file and all executable files before deciding the verdict.
Treat scanner output as evidence and verify intent, data flow, and declared behavior.

避免

Do not publish a skill solely because its numeric score is low.
Do not execute suspicious scripts to determine what they do.
Do not dismiss quoted prompt injection without checking its surrounding context.