技能 infsh-cli

📦

infsh-cli

Name: infsh-cli
Author: inference-sh-skills

低風險 ⚙️ 外部命令🌐 網路存取🔑 環境變數📁 檔案系統存取

使用 inference.sh CLI 執行 250+ 個 AI 應用程式

也可從以下取得: qu-skills,skillssh,inference-skills,inferen-sh,infsh-skills,tool-belt

執行 FLUX、Veo 和 Gemini 等 AI 模型需要強大的硬體和複雜的設定。此技能讓您能直接從終端機使用簡單的 CLI 指令，存取 250 多個 AI 應用程式，無需 GPU。

支援: Claude Codex Code(CC)

⚠️ 67 差

下載技能 ZIP

在 Claude 中上傳

前往設定 → 功能 → 技能 → 上傳技能

開啟並開始使用

測試它

正在使用「infsh-cli」。 Generate an image of a cat astronaut floating in space

預期結果:

CLI 從 FLUX 模型返回生成的圖片 URL。圖片顯示一隻穿著太空衣的貓，在繁星點點的太空背景中漂浮的細緻渲染。

正在使用「infsh-cli」。 Search the web for latest AI research papers

預期結果:

Tavily 搜尋助手返回一系列相關搜尋結果，包含最新 AI 研究論文的標題、URL 和內容摘要。

正在使用「infsh-cli」。 Explain quantum computing using Claude

預期結果:

OpenRouter API 從 Claude Sonnet 返回文字回應，以便於理解的方式解釋量子計算概念。

安全審計

低風險

v1 • 5/11/2026

This is a documentation skill for the inference.sh CLI tool (belt command). The skill has allowed-tools restriction Bash(belt *) which constrains execution to only the belt command. All 203 static analysis findings are in markdown fenced code blocks showing CLI usage examples, not executable code. The pipe-to-shell installation pattern (curl | sh) is a real risk but is the standard installation method for this CLI tool. No malicious intent or obfuscation was found. The skill is safe to publish with a low risk advisory for the installation pattern.

已掃描檔案

596

分析行數

發現項

審計總數

中風險問題 (1)

SKILL.md:16 references/authentication.md:6 references/authentication.md:44 references/cli-reference.md:6

Pipe-to-shell installation pattern documented in skill

The skill documents installing the CLI via curl piped to sh (curl -fsSL https://cli.inference.sh | sh). This pattern is standard for CLI tools but carries inherent risk: the script executes without prior review, and a compromised endpoint could serve malicious code. The skill provides a manual install alternative as a safer option.

低風險問題 (4)

SKILL.md:15-127 references/app-discovery.md:5-106 references/authentication.md:5-53 references/cli-reference.md:5-97 references/running-apps.md:5-164

Documentation-only CLI command examples flagged as external commands

All 146 external_commands findings are Ruby backtick executions or shell commands inside markdown fenced code blocks. These are documentation examples showing users how to use the belt CLI. They are not executable code. No command injection risk exists because the skill is restricted to Bash(belt *) via allowed-tools.

SKILL.md:138-144 references/app-discovery.md:110-112

Documentation URLs flagged as network risk

All 32 network findings are hardcoded URLs in documentation linking to inference.sh platform pages (docs, blog, API). These are informational links for users to read more about the platform. The skill does not make network requests; the URLs are documentation references only.

references/authentication.md:30 references/authentication.md:53 references/cli-reference.md:72

Documentation references to environment variable flagged as env_access

The INFSH_API_KEY environment variable is referenced in documentation examples showing users how to configure authentication. The skill itself does not read or access any environment variables. The reference is instructional - telling users to set the variable for the belt CLI tool.

references/cli-reference.md:84 references/running-apps.md:41 SKILL.md:31

Documentation file path examples flagged as filesystem risk

File paths like ~/.config/fish/completions/infsh.fish and ../data/video.mp4 appear in documentation examples showing shell completion installation and file upload features of the CLI. These are example paths, not actual filesystem operations performed by the skill.

風險因素

⚙️ 外部命令 (146)

🌐 網路存取 (32)

🔑 環境變數 (3)

references/authentication.md:30 references/authentication.md:53 references/cli-reference.md:72

📁 檔案系統存取 (5)

references/cli-reference.md:84 references/cli-reference.md:84 references/running-apps.md:41 SKILL.md:31 SKILL.md:31

偵測到的模式

Pipe-to-shell installation command

審計者: claude

品質評分

架構

100

可維護性

內容

社群

安全

規範符合性

你能建構什麼

為內容創作生成圖片與影片

內容創作者可直接從終端機使用 FLUX 模型生成圖片、使用 Veo 生成影片，無需昂貴的 GPU 硬體或雲端設定。

開發者快速原型製作與 AI 模型比較

從命令列快速測試和比較 Claude、Gemini 和 Grok 等不同 AI 模型，無須切換平台或管理 API 整合。

收集與處理網路資料以供研究

使用 Tavily 和 Exa 搜尋工具從網路擷取資訊，並在資料分析管道中以程式化方式處理結果。

試試這些提示

使用 FLUX 生成圖片

Generate an image of [description] using FLUX through the inference.sh CLI. Run: belt app run falai/flux-dev-lora --input '{"prompt": "[description]"}'

根據提示詞生成影片

Create a video of [description] using Veo. Run: belt app run google/veo-3-1-fast --input '{"prompt": "[description]"}'

搜尋網路並總結結果

Search the web for [topic] using Tavily, then summarize the key findings. Run: belt app run tavily/search-assistant --input '{"query": "[topic]"}'

執行多模型管道

First search for [topic] using Tavily search, then use the results as context for Claude Sonnet to generate an analysis report. Chain belt commands to pass data between models.