📦
qwen-image
77使用 inference.sh 生成 Qwen 图像
创建精致的图像通常需要特定于模型的选项和谨慎的提示词结构。此技能会指导 Claude、Codex 和 Claude Code 使用 inference.sh 进行 Qwen-Image 生成和编辑。
设计与创意 查看
技能 agent-tools
📦
agent-tools
内容修订 r1 严重 🌐 网络访问🔑 环境变量📁 文件系统访问⚙️ 外部命令
从 CLI 运行云端 AI 应用
团队需要一种统一方式来发现和运行托管 AI 模型，而无需管理 GPU。此技能会引导 Claude、Codex 和 Claude Code 完成 inference.sh CLI 工作流。
支持: Claude Codex Code(CC)
请勿自动安装此 Skill。
权威策略要求在执行任何安装操作前由操作人员审查。
使用我的智能体安装
将此请求复制给你的智能体，其中包含权威 Skill 页面和清单。
智能体请求
Review the Skillstore skill "agent-tools" from https://skillstore.io/skills/tul-sh-agent-tools.md and its manifest at https://skillstore.io/api/skills/tul-sh-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.你的智能体仍应展示其计划，并按安全策略请求必要确认。
Agent 可读资源
当 AI Agent、爬虫或脚本需要干净上下文，而不是读取完整页面时，请使用这些链接。
测试它
正在使用“agent-tools”。 查找适用于产品模型图的图像生成应用。
预期结果:
- 推荐了三个图像应用，并说明其优势和输入需求。
- 建议在任何运行之前先生成示例输入。
- 提示在将提示发送到云端之前需要批准。
正在使用“agent-tools”。 根据我批准的提示创建一个短视频。
预期结果:
- 准备了应用输入并提交了视频生成任务。
- 返回了任务标识符并检查了进度。
- 任务完成后报告了最终下载链接。
正在使用“agent-tools”。 搜索最近的 AI 产品发布并进行总结。
预期结果:
- 选择了一个搜索应用，并请求对外部查询进行批准。
- 收集搜索结果并总结关键发布内容。
- 在请求时保存了结果摘要。
安全审计
严重v5 • 2026/7/7 打开版本化报告
The skill is useful but includes critical install guidance that pipes a remote script into a shell and documents dynamic remote binary downloads. Most API-key and markdown backtick detections are false positives, but external CLI calls, cloud data transfer, and X automation remain real risks. Publication should require safer installation guidance and clear approval gates for data uploads and public actions.
5
已扫描文件
554
分析行数
24
复核项
0
已忽略误报
已确认安全问题 (6)
严重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The command pipes a remote HTTPS response directly into sh. If the server, network path, or installer is compromised, arbitrary code can run immediately.
严重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The reinstall instructions repeat the curl-to-sh pattern. This executes remote installer content without a prior pinned checksum or signature check.
严重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The install command executes a remotely downloaded script through sh. That is a high-impact supply-chain risk if the remote content changes or is compromised.
严重
Pipe to shell pattern
curl -fsSL https://cli.inference.sh | sh
The primary installation instructions pipe a remote script directly into sh. This can execute arbitrary code before the user can inspect or verify the installer.
中
Public Social Media Side Effects
SKILL.md includes an example that posts to X through x/post-tweet. This can create public account actions and should require explicit user approval.
The skill explicitly documents X automation commands, including posting text. This is a clear external side effect, though it is also part of the stated feature set.
中
Unverified Installer Safety Claims
SKILL.md reassures users that the installer has no elevated permissions, background processes, or telemetry. The same line links to a remote installer, so the claim should be backed by verifiable source and pinned releases.
The text is not a prompt injection attempt, but it is trust-building language around a remote execution workflow. The risk depends on external installer transparency.
能力复核项 (24)
这些是真实的本地能力，对此技能可能是预期行为，因此需要复核，但不会按已确认恶意行为计分。
严重
Ruby/shell backtick execution
```bash
The fenced block contains install and login commands, including a remote curl-to-sh installer. The static label is imprecise, but the command block is operational and risky.
高
Ruby/shell backtick execution
> ```bash
The manual install block downloads remote artifacts, verifies a checksum, extracts an archive, and moves a binary into the user path. This is legitimate setup guidance but real external command risk.
高
Shell command substitution
> curl -LO $(curl -fsSL https://dist.inference.sh/cli/manifest.json | grep -o '"url":"[^"]*"' | grep
The command uses nested shell substitution to fetch a remote manifest and derive the binary URL. If the manifest or parsing is compromised, the install flow can select an unsafe artifact.
高
Template literal with command substitution
> ```bash
The manual install block includes dynamic command substitution and shell utilities. An agent following it would execute generated shell commands from remote metadata.
高
Hidden file in home directory
> mv inferencesh-cli-* ~/.local/bin/inferencesh
The manual install flow moves a downloaded binary into ~/.local/bin. That is a standard user install path, but it adds executable code to the user environment.
中
Ruby/shell backtick execution
```bash
The quick examples run infsh commands for media, LLM, search, X, and 3D apps. These commands send user input to cloud services and can create public side effects.
中
Ruby/shell backtick execution
| List all apps | `infsh app list` |
The inline command invokes the infsh CLI to list remote apps. It is read-only, but it still depends on an external CLI and network service.
中
Ruby/shell backtick execution
| Search apps | `infsh app list --search "flux"` |
The command searches remote app listings through infsh. This is expected behavior, but it is still external command execution with network access.
中
Ruby/shell backtick execution
| Filter by category | `infsh app list --category image` |
The command filters app listings through the external infsh CLI. It is low impact but confirms reliance on networked command execution.
中
Ruby/shell backtick execution
| Get app details | `infsh app get google/veo-3-1-fast` |
The command fetches app details from the inference.sh service. It is legitimate discovery behavior but still a real external command invocation.
中
Ruby/shell backtick execution
| Generate sample input | `infsh app sample google/veo-3-1-fast --save input.json` |
The command calls infsh and writes a sample input file. The file write is expected, but it is still external command execution with local filesystem effects.
中
Ruby/shell backtick execution
| Run app | `infsh app run google/veo-3-1-fast --input input.json` |
The command runs a remote app using a local input file. This can upload user-provided file contents to an external service.
中
Ruby/shell backtick execution
| Run without waiting | `infsh app run <app> --input input.json --no-wait` |
The no-wait run command submits work to a remote service and returns later. It can upload input data and make tracking less immediate.
中
Ruby/shell backtick execution
| Check task status | `infsh task get <task-id>` |
The command checks task status through the external infsh service. The action is mostly read-only, but it confirms networked CLI use.
中
Ruby/shell backtick execution
```bash
The related-skills block uses npx skills add commands to fetch and install additional skills. That introduces remote package and supply-chain risk.
中
Hidden file access
> mv inferencesh-cli-* ~/.local/bin/inferencesh
The command writes into a hidden home directory path as part of installing the CLI. It is expected setup behavior but creates persistent local filesystem impact.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
The URL is used in a curl-to-shell installer command, so it fetches remote content from a third-party domain. The network use is intentional but security-relevant because the fetched content is executed.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
The URL appears in a reinstall command that fetches a remote installer before shell execution. This is expected setup behavior but remains a supply-chain network dependency.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
The hardcoded URL is part of the CLI installation command. It is a real external dependency because setup downloads installer content from inference.sh.
低
Hardcoded URL
](https://cloud.inference.sh/app/files/u/4mg21r6ta37mpaz6ktzwt
The markdown image references a remote cloud.inference.sh asset. Rendering it can create a network request and tracking surface, though it is not code execution.
低
Hardcoded URL
curl -fsSL https://cli.inference.sh | sh
The URL is used by an install command that downloads remote content. The paired pipe-to-shell finding captures execution, while this confirms the external dependency.
低
Hardcoded URL
> **What does the installer do?** The [install script](https://cli.inference.sh) detects your OS and
The line links to the installer and describes downloads from dist.inference.sh. It is installer documentation tied to remote executable distribution.
低
Hardcoded URL
> curl -LO https://dist.inference.sh/cli/checksums.txt
The command downloads a checksum file from an external domain. This is expected for manual installation but still a real network dependency.
低
Hardcoded URL
> curl -LO $(curl -fsSL https://dist.inference.sh/cli/manifest.json | grep -o '"url":"[^"]*"' | grep
The command fetches a remote manifest to choose a binary download URL. This is an operational network fetch in the installation path.
风险因素
🌐 网络访问 (17)
🔑 环境变量 (3)
检测到的模式
Pipe to shell pattern×4
审计者: codex 查看审计历史 →
分享与引用此报告
分享版本化评估报告、中性徽章、嵌入卡片和引用。Skillstore 只报告证据，不判断该 Skill 是否安全。
复制报告链接
https://skillstore.io/skills/tul-sh-agent-tools/audits/5?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/tul-sh-agent-tools?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/tul-sh-agent-tools?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/tul-sh-agent-tools/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/tul-sh-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>学术引用 (APA · BibTeX · CFF)
APA 引用
tul-sh. (2026). agent-tools security audit report (audit version 5) [Author version unspecified]. Skillstore. https://skillstore.io/skills/tul-sh-agent-tools/audits/5BibTeX 引用
@techreport{tul-sh-tul-sh-agent-tools-2026,
author = {tul-sh},
title = {agent-tools security audit report (audit version 5)},
institution = {Skillstore},
year = {2026},
number = {5},
url = {https://skillstore.io/skills/tul-sh-agent-tools/audits/5},
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 5)"
version: "unspecified"
type: report
authors:
- name: "tul-sh"
date-released: "2026-07-07"
url: "https://skillstore.io/skills/tul-sh-agent-tools/audits/5"
identifiers:
- type: other
value: "skillstore:tul-sh-agent-tools:audit:5"
description: "Skillstore immutable audit report identifier"
比较变体
20 个可安装变体每位作者的技能都保持为独立的可安装项。推荐变体按 Skillstore 证据排序。
为什么此变体排名第一
Skillstore 使用量更高
作者 / 来源 Skillstore 评分 证据置信度 Skillstore 使用量 已更新
inference-skills-agent-tools
Skillstore 评分 38
证据置信度 中
Skillstore 使用量 44
已更新
2026-09-09
inference-sh-skills-agent-tools
Skillstore 评分 38
证据置信度 中
Skillstore 使用量 43
已更新
2026-09-09
inference-shell-agent-tools
Skillstore 评分 38
证据置信度 中
Skillstore 使用量 43
已更新
2026-09-09
halt-catch-fire-agent-tools
Skillstore 评分 38
证据置信度 中
Skillstore 使用量 38
已更新
2026-09-09
Skillstore 评分
评分依据 证据置信度: 中 45
架构
85
可维护性
85
内容
69
社区
91
规范符合性
你能构建什么
原型化媒体工作流
查找图像或视频应用，创建示例输入，并运行已批准的生成任务。
比较模型选项
搜索可用应用，检查 schema，并测试托管 LLM 或多模态模型。
自动化研究任务
使用搜索和语言应用来收集、总结并保存研究输出。
试试这些提示
查找应用
使用 agent-tools 列出 inference.sh 图像应用。推荐三个选项，并解释每个应用所需的输入。
运行示例任务
使用 agent-tools 为 falai/flux-dev-lora 创建示例输入。在使用我的提示运行它之前先询问我。
跟踪长时间任务
使用 agent-tools 运行 google/veo-3-1-fast 且不等待。检查任务状态，直到结果准备就绪。
规划多应用工作流
设计一个 inference.sh 工作流，用于搜索网页、使用 LLM 总结结果并保存输出。每次外部调用前都先询问。
最佳实践
- 在运行任何任务之前检查应用详情和输入 schema。
- 在进行网络调用、文件上传或社交发布之前请求明确批准。
- 使用示例输入和版本固定来实现可重复运行。
避免
- 不要在无人值守会话中将远程安装脚本通过管道传给 shell。
- 不要将密钥、私有文件或受监管数据作为应用输入发送。
- 不要在没有人工批准的情况下自动执行公开社交媒体操作。
常见问题
这个技能有什么作用？
它帮助 AI 助手使用 inference.sh CLI 来发现和运行托管 AI 应用。
它需要账户吗？
是的。大多数工作流需要 inference.sh 账户、身份验证、额度和网络访问。
它能生成图像和视频吗？
是的。它记录了图像和视频应用发现、示例输入创建、任务执行和结果检索。
它能调用语言模型吗？
是的。它包含通过 OpenRouter 和 inference.sh 应用调用 Claude 及其他托管模型的示例。
我的数据会被发送到外部吗？
是的。当你运行命令时，提示、输入文件和任务数据会发送到 inference.sh 应用。
它能发布到 X 吗？
该技能记录了 X 自动化应用。任何公开操作都应先获得明确批准。
开发者详情
作者
tul-sh许可证
MIT
Skillstore 修订
r1
版本提示
作者未声明版本。
引用
36e07d5e13068e5be64447e8f20b427cf2cbd21a
维护新鲜度
2026/7/18
使用量
23 次下载 · 146 次浏览
文件结构
tul-sh 的更多技能
查看全部📦
ai-image-generation
75使用 Inference.sh 生成 AI 图像
图像团队需要快速访问许多托管生成模型。此技能会指导 Claude、Codex 和 Claude Code 完成 inference.sh 图像工作流。
设计与创意 查看
📦
qwen-image-pro
70创建专业级 Qwen 图像
团队需要可靠的图像生成能力，用于海报、横幅和文本密集型设计。此技能通过 inference.sh 指导 Claude、Codex 和 Claude Code 使用 Qwen-Image Pro。
设计与创意 查看
📦
nano-banana
70使用 inference.sh 生成 Gemini 图像
用户需要一种从 AI 助手快速运行 Gemini 图像模型的方法。此技能提供用于生成、编辑、比例设置和基于事实依据的提示词的 CLI 模式。
设计与创意 查看
📦
remotion-render
50从 TSX 渲染 Remotion 视频
团队需要可重复的视频渲染，而无需构建完整的渲染流水线。此技能通过 inference.sh 将 Remotion 组件代码转换为 MP4 输出。
设计与创意 查看
📦
ai-video-generation
50使用 inference.sh 创建 AI 视频
团队需要可重复使用的视频模型提示词，而不必记住每个 CLI app。此技能会指导 Claude、Codex 和 Claude Code 完成 inference.sh 视频工作流。
设计与创意 查看
相关技能
查看全部📦
writingmate-mcp
80Compare AI Models and Generate Media
由 writingmate
Choosing and testing AI models across separate services takes time. Writingmate connects an MCP host to model discovery, text comparison, image generation, and video generation.
效率 查看
📦
prompt-optimization
82优化 LLM 提示词和智能体指令
由 89jobrien
薄弱的提示词会导致模型输出不一致，并拖慢 AI 功能开发。此技能可帮助为 Claude、Codex 和 Claude Code 构建提示词、示例、约束和输出格式。
效率 查看
📦
prompt-engineering
79用经过验证的模式改进 AI 提示词
由 sickn33
薄弱的提示词会导致 AI 结果不一致，并浪费审核时间。此技能提供清晰的模式、模板和测试方法，用于编写可靠的提示词。
效率 查看
📦
runcomfy-cli
78从 CLI 运行 RunComfy 模型
由 runcomfy-com
团队需要一种可重复的方式，通过脚本运行 RunComfy 媒体模型。此技能指导 CLI 安装、身份验证、模型运行、轮询和下载。
设计与创意 查看
📦
native-transparent-imagegen
86Generate Native-Alpha Transparent Images
由 zseven-w
Image generators can return opaque files or painted checkerboards when transparency is required. This skill generates native-alpha assets and validates every untouched output.
设计与创意 查看
📦
generating-storyboards
82Create Consistent Video Storyboards
由 supercmohq
Video concepts can lose continuity when clips are planned separately. This skill creates sequential storyboard sheets that preserve characters, products, settings, and visual states.
设计与创意 查看