Skills

inference.sh skills: Let Your Agent Generate Images and Video, Call LLMs, and Search the Web

Best suited for having an agent generate images or video, call LLMs, or search the web via inference.sh; you must install and log in to its CLI `belt` beforehand, otherwise the skills will not work.

Official source · GitHub - 101-skills/superpowers: Agent superpowers

Install

npx skills add inference-sh/skills

Who It's For

  • Agent users on or planning to use the inference.sh platform who want their agent to generate images directly (50+ models such as FLUX, Gemini, Reve), generate video (40+ models such as Veo, Seedance, Wan), call LLMs (Claude, Gemini, Kimi, GLM), or perform web search (Tavily/Exa).
  • Users who install skills via the npx skills or Claude Code plugin mechanism; the repository also provides JS/TS and Python SDK-style skills as well as UI component skills.
  • Prerequisites: these skills depend on the inference.sh CLI tool belt, which must be installed first, followed by belt login to sign in. Not suitable for users without an inference.sh account or who cannot conveniently use belt.
  • The official evidence does not state which text languages the skills support, operating system requirements, or the scope of file read/write access; this information is unknown and will not be guessed at.

First Try in 3 Minutes

The official primary installation method is npx skills add.

  1. Prerequisite: have a Node/npx environment and an inference.sh account ready, and install belt following the official CLI installation instructions.
  2. Install the skills. To install all skills from the repository root:
npx skills add inference-sh/skills

You can also install a single skill only, e.g. npx skills add inference-sh/skills@flux-image. The official docs also offer a Claude Code plugin method: /plugin install inference-sh; after installation, the skills become available as /inference-sh:flux-image, /inference-sh:google-veo, and so on.

  1. Sign in to the runtime:
belt login

After signing in, you can use belt app list to browse the available apps.

  1. For your first task, choose something low-risk and observable, such as generating an image: ask your agent to "generate a simple test image with flux-image". Note: generating an image calls inference.sh's external service, which means network requests and potential consumption of your account quota. Before executing, confirm you are logged in and aware of this.

  2. Success signal: the agent triggers the corresponding image-type skill and returns the generated image result. If the skill is not installed or belt is not logged in, it will fail. If belt app list can list apps, the CLI environment is working properly.

Use Cases

Case 1: Generating an Image with FLUX

Scenario: quickly verifying image generation capability. After installing and running belt login, tell the agent: "generate an image with inference-sh:flux-image". This uses the ai-image-generation tool-type skill listed in the repository (50+ image models). Within the evidence, you can expect: the agent invokes the skill and returns the generated image; if using the Claude Code plugin method, the invocation form is /inference-sh:flux-image.

Case 2: Browsing the Apps Available on the Platform

Scenario: before generating anything, confirm which models/services are callable under your account. After completing belt login, tell the agent or manually run belt app list to browse apps. This is the confirmed limit: the official documentation explicitly provides this command for browsing apps; as for the exact output format of each app, the evidence does not specify—defer to actual output.

Boundaries

  • Network and external services: generating images and video, calling LLMs, and web search all invoke inference.sh's external services, producing network requests. Whether this consumes account quota and how billing works are not stated in the collected evidence and remain unknown.
  • Side effects and confirmation points: installation writes skill files to your machine; for the first attempt, start with low-risk tasks whose results are disposable.
  • Runtime environment: the belt CLI and login are required—this is a hard prerequisite explicitly stated by the official docs.
  • Installation differences are preserved per platform: npx skills add is the general installation method; Claude Code users can additionally install via /plugin install inference-sh. Do not conflate the two.
  • Key information not yet confirmed: the skills' exact compatibility scope across clients, supported text languages, whether they read/write local files, and prerequisites for skills like the X/Twitter API—none have official evidence and cannot be confirmed.

Repository: https://github.com/101-skills/superpowers