技能 agent-tools
📦
agent-tools
內容修訂 r1 嚴重 🌐 網路存取🔑 環境變數📁 檔案系統存取⚙️ 外部命令
執行 inference.sh AI Apps
代理程式需要一個介面來處理多種託管式 AI 模型與媒體工作流程。這項技能會教 Claude、Codex 和 Claude Code 安全地使用 infsh CLI。
支援: Claude Codex Code(CC)
請勿自動安裝此 Skill。
權威政策要求在執行任何安裝操作前由操作人員審查。
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 "agent-tools" from https://skillstore.io/skills/inferencesh-agent-tools.md and its manifest at https://skillstore.io/api/skills/inferencesh-agent-tools/manifest. Verify the artifact. Do not auto-install. Inspect the skill and report your findings, then wait for an operator or manual installation decision.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。
測試它
正在使用「agent-tools」。 尋找適合短產品片段的快速影片模型。
預期結果:
助理會列出相關影片 apps，說明速度與品質取捨，並建議先生成範例輸入。
正在使用「agent-tools」。 執行影像生成任務並儲存結果。
預期結果:
助理會選擇影像 app、準備輸入、執行任務、回報完成狀態，並指向已儲存的媒體結果。
正在使用「agent-tools」。 疑難排解 invalid input 錯誤。
預期結果:
- 助理會檢查 app schema。
- 它會找出缺少的欄位。
- 它會建議修正後的輸入值。
安全審計
嚴重v4 • 2026/7/5 開啟版本化報告
Review confirmed multiple remote installer commands that pipe downloaded content directly into a shell, which is a critical supply-chain risk. Most API key, hidden file, documentation link, and system reconnaissance matches are benign reference material. The skill still carries elevated risk because it grants broad infsh command access and includes account-changing cloud and X/Twitter actions.
5
已掃描檔案
545
分析行數
16
複核項目
0
已忽略的誤報
已確認的安全問題 (7)
嚴重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The command pipes a remote installer directly into sh without checksum or signature verification. If the endpoint or transport is compromised, it can execute arbitrary code.
嚴重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The reinstall path repeats the same remote script piped into sh. This is a critical supply-chain pattern even when used for a legitimate CLI installer.
嚴重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The installation command downloads a remote script and executes it with sh. This bypasses normal package verification and creates direct code execution risk.
嚴重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The skill prominently instructs running a remote installer directly through sh. That gives the remote endpoint immediate local code execution without integrity verification.
高
Broad Account-Scoped CLI Permission
The skill declares allowed-tools for any infsh command, allowing the agent to run broad account-scoped CLI actions once active.
The permission line is explicit, and the CLI reference lists authenticated app execution and development commands. Impact depends on configured account credentials.
高
State-Changing Social Media Automation
The skill provides X/Twitter commands for posting and lists actions for direct messages, following, likes, and reposts. These workflows can act from a user account.
The listed commands and action categories directly support social publishing and engagement. The risk depends on whether the user has connected credentials.
中
Related Skill Supply Chain Expansion
The related skills section recommends installing multiple additional skills through npx. That expands the trusted instruction set beyond this audited package.
The commands are explicit installation recommendations. The exact risk depends on the content and review status of each related skill.
能力複核項目 (16)
這些是真實的本機能力，對此技能而言可能是預期行為，因此需要複核，但不會按已確認的惡意行為計分。
中
Ruby/shell backtick execution
```bash
The surrounding fenced bash block tells the agent or user to run installer and login commands. These external commands change local and account state.
中
Ruby/shell backtick execution
```bash
The fenced example block contains infsh app run commands for cloud models, web search, and X/Twitter posting. These commands can trigger network calls and account actions.
中
Ruby/shell backtick execution
| List all apps | `infsh app list` |
The inline command lists remote apps through the infsh CLI. It is a legitimate feature, but it still invokes an external authenticated tool.
中
Ruby/shell backtick execution
| Search apps | `infsh app list --search "flux"` |
The inline command searches remote app metadata through infsh. This expected external command can disclose query intent to the service.
中
Ruby/shell backtick execution
| Filter by category | `infsh app list --category image` |
The inline command filters hosted apps through the external CLI. It is benign in intent but confirms networked command execution.
中
Ruby/shell backtick execution
| Get app details | `infsh app get google/veo-3-1-fast` |
The inline command retrieves app details from the inference.sh service. That is legitimate discovery, but it uses an external account-scoped CLI.
中
Ruby/shell backtick execution
| Generate sample input | `infsh app sample google/veo-3-1-fast --save input.json` |
The inline command generates and saves sample input through infsh. It can write files and depends on trusted CLI behavior.
中
Ruby/shell backtick execution
| Run app | `infsh app run google/veo-3-1-fast --input input.json` |
The inline command runs a hosted AI app with an input file. It can consume account credits and submit local prompt data to a remote service.
中
Ruby/shell backtick execution
| Run without waiting | `infsh app run <app> --input input.json --no-wait` |
The no-wait run command can start long-running cloud tasks without immediate result review. It still submits data and may consume account resources.
中
Ruby/shell backtick execution
| Check task status | `infsh task get <task-id>` |
The task lookup command queries remote task state through the infsh CLI. This is normal operation, but it confirms external service interaction.
中
Ruby/shell backtick execution
```bash
The related-skills block recommends npx skills add commands, which can install additional remote skill content. That expands the trusted instruction surface beyond this package.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
This URL is part of a command that downloads a remote installer from inference.sh. The network access is expected, but it creates supply-chain exposure when paired with shell execution.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
This reinstall command downloads the CLI installer from a remote URL. The URL itself is documented, but it contributes to the remote installer trust boundary.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
The hardcoded URL is used to fetch a remote CLI installer. It is expected network access, but it should be verified before execution.
低
Hardcoded URL
](https://cloud.inference.sh/app/files/u/4mg21r6ta37mpaz6ktzwt
The Markdown embeds a remotely hosted image that may be requested when rendered. This is a low-severity privacy and dependency risk, not code execution.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
The URL is the source for a remote installer command. The network access is intended, but it is risky when combined with direct shell execution.
風險因素
🌐 網路存取 (14)
🔑 環境變數 (3)
偵測到的模式
Pipe to shell pattern×4
審計單位: codex 檢視審計紀錄 →
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。
複製報告連結
https://skillstore.io/skills/inferencesh-agent-tools/audits/4?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/inferencesh-agent-tools?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/inferencesh-agent-tools?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/inferencesh-agent-tools/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/inferencesh-agent-tools.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
inferencesh. (2026). agent-tools security audit report (audit version 4) [Author version unspecified]. Skillstore. https://skillstore.io/skills/inferencesh-agent-tools/audits/4BibTeX 引用
@techreport{inferencesh-inferencesh-agent-tools-2026,
author = {inferencesh},
title = {agent-tools security audit report (audit version 4)},
institution = {Skillstore},
year = {2026},
number = {4},
url = {https://skillstore.io/skills/inferencesh-agent-tools/audits/4},
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: "agent-tools security audit report (audit version 4)"
version: "unspecified"
type: report
authors:
- name: "inferencesh"
date-released: "2026-07-05"
url: "https://skillstore.io/skills/inferencesh-agent-tools/audits/4"
identifiers:
- type: other
value: "skillstore:inferencesh-agent-tools:audit:4"
description: "Skillstore immutable audit report identifier"
比較變體
20 個可安裝變體每位作者的技能都保留為獨立的可安裝項目。推薦變體依 Skillstore 證據排序。
為何此變體排名第一
Skillstore 使用量較高
作者 / 來源 Skillstore 評分 證據置信度 Skillstore 使用量 已更新
inference-skills-agent-tools
Skillstore 評分 38
證據置信度 中
Skillstore 使用量 47
已更新
2026-09-09
inference-sh-skills-agent-tools
Skillstore 評分 38
證據置信度 中
Skillstore 使用量 46
已更新
2026-09-09
inference-shell-agent-tools
Skillstore 評分 38
證據置信度 中
Skillstore 使用量 46
已更新
2026-09-09
halt-catch-fire-agent-tools
Skillstore 評分 38
證據置信度 中
Skillstore 使用量 41
已更新
2026-09-09
Skillstore 評分
評分依據 證據置信度: 中 45
架構
85
可維護性
87
內容
70
社群
83
規範符合度
你可以打造什麼
建立媒體生成原型
尋找影像、影片、音訊或 3D apps，並從命令列快速執行測試工作。
比較託管式 AI 模型
探索 LLM、搜尋與媒體 apps，檢查 schema，並比較不同供應商的結果。
操作帳戶工作流程
使用已驗證的 CLI 命令進行 app 執行、任務追蹤、部署檢查，以及選用的 X/Twitter 工作流程。
試試這些提示
列出可用 Apps
尋找用於影像生成的 inference.sh apps。列出優秀選項，並在執行任何項目前說明每個選項最適合的用途。
建立範例輸入
檢查 app google/veo-3-1-fast 並建立一個範例輸入檔案。說明我應該編輯哪些欄位。
執行並追蹤任務
使用我的提示執行影像生成任務，追蹤直到完成，並摘要回傳的媒體結果。
建立多步驟工作流程
尋找 text-to-image app，根據我的產品簡介準備提示，執行任務，然後使用搜尋 app 收集支援參考資料。在付費或發布動作前先詢問。
最佳實務
- 執行前先審查遠端安裝命令。
- 在執行付費任務前，先使用範例輸入與 app 詳細資料。
- 與帳戶擁有者確認 X/Twitter、部署或憑證相關操作。
避免
- 不要在未驗證的情況下將遠端安裝程式 pipe 到 shell。
- 不要在沒有明確核准的情況下執行發布或帳戶操作。
- 不要將真實 API keys 放在提示、日誌或共享檔案中。
常見問題
這項技能能做什麼？
它能協助 AI agent 使用 infsh CLI 探索、執行並追蹤託管式 inference.sh apps。
哪些工具可以使用它？
報告列出支援 Claude、Codex 和 Claude Code。
它需要帳戶嗎？
需要。大多數實用命令都需要 infsh 驗證，且可能需要額度或供應商憑證。
它可以生成影像或影片嗎？
可以。這項技能包含透過 inference.sh apps 進行 FLUX 影像生成與 Veo 影片生成的範例。
它可以發文到 X 或 Twitter 嗎？
可以，前提是相關 app 與帳戶憑證已設定完成。使用者應先確認每一個發布動作。
主要的安全疑慮是什麼？
文件使用遠端安裝程式命令，會將下載內容直接 pipe 到 shell。
開發者詳細資訊
作者
inferencesh授權
MIT
Skillstore 修訂
r1
版本提示
作者未聲明版本。
引用
a25199bc7d6b82598536822d1738eb5d5f54025b
維護新鮮度
2026/7/18
使用量
8 次下載 · 201 次瀏覽
檔案結構