agent-tools
Access 150+ Cloud AI Apps via CLI
こちらからも入手できます: inference-sh-7,inference-sh-6,inference-sh-3,inf-sh,inference-sh-0
Running AI models locally requires expensive GPU hardware. The inference.sh CLI enables access to 150+ cloud AI applications for image generation, video creation, LLM calls, and web search without any local setup.
スキルZIPをダウンロード
Claudeでアップロード
設定 → 機能 → スキル → スキルをアップロードへ移動
オンにして利用開始
テストする
「agent-tools」を使用しています。 infsh app run falai/flux-dev-lora --input '{"prompt": "a red bicycle in a parisian street"}'
期待される結果:
Task ID: abc123xyz
Image URL: https://cloud.inference.sh/images/def456.png
The generated image is ready for download.
「agent-tools」を使用しています。 infsh app run tavily/search-assistant --input '{"query": "Claude Code AI features"}'
期待される結果:
Results:
1. Claude Code - Anthropic's CLI for AI-assisted development
2. Features include: code generation, refactoring, debugging
3. Available via npm: npx @anthropic/claude-code
「agent-tools」を使用しています。 infsh task get abc123xyz --json
期待される結果:
{
"status": "completed",
"output": {
"url": "https://cloud.inference.sh/video.mp4"
}
}
セキュリティ監査
低リスクDocumentation-only skill containing markdown reference files. Static findings are false positives - patterns detected are CLI command examples in documentation, not executable code. All curl pipe patterns use security-enhancing flags (-fsSL). No malicious intent, credential exfiltration, or actual code execution risks detected.
リスク要因
🌐 ネットワークアクセス (2)
⚙️ 外部コマンド (2)
品質スコア
作れるもの
Generate marketing images with FLUX
Create professional product images and marketing visuals using FLUX models through simple CLI commands without local GPU requirements.
Build video workflows with Veo 3
Generate short video clips and animations using Google Veo models for social media, presentations, or prototypes.
Research with web search and LLM chaining
Combine Tavily web search with Claude or other LLMs to research topics and generate summaries automatically.
これらのプロンプトを試す
Use the inference.sh CLI to generate an image with FLUX. First list available image apps, then run falai/flux-dev-lora with a prompt of "${1:a cozy coffee shop interior}".Create a drone-shot style video using Veo 3. Generate a sample input file first, then run google/veo-3-1-fast with the prompt "${1:aerial view of ocean waves at sunset}".Use Tavily search to find information about "${1:latest developments in quantum computing}", then summarize the results using Claude Sonnet via OpenRouter.Post a promotional tweet using the x/post-tweet app. The tweet text should be: "${1:Check out our new AI-powered tool!}"ベストプラクティス
- Generate sample inputs first with 'infsh app sample' to understand required parameters before running apps
- Use version pinning (@version) for production scripts to ensure reproducible results
- Store task IDs when running with --no-wait to retrieve results later
回避
- Hardcoding API keys directly in command lines - use environment variables instead
- Running multiple apps sequentially without checking task status between runs
- Forgetting to check available apps before assuming a model is available