vercel-labs / ai-cli

The one skill it ships is thin by design — its own description is a single generic sentence, and the real detail lives only in the README: piping output between commands (an image straight into a video prompt), reference images and multi-shot comparisons across models specified as creator/model-name, a models command that prints live pricing and per-provider latency, and inline previews in terminals that support the Kitty graphics protocol. All of it runs through the Vercel AI SDK and AI Gateway rather than separate per-provider SDKs.

This is for someone scripting from a terminal, or an agent with shell access, who wants one command for generation across providers instead of switching SDKs per model; it needs Node.js 22+ and either an AI Gateway key or a specific provider key such as OPENAI_API_KEY, and produces nothing without one.

READ THE FULL ANALYSIS＋

The one skill's own description gives an agent very little to decide on. An agent choosing whether to reach for this skill sees only "generate text, images, video, and audio from the terminal" — the piping, multi-model, and pricing-lookup behavior that would actually justify picking it over a single provider call is not in the description at all, only in the README a reader has to open separately. Playing back generated speech or audio in formats other than WAV also depends on finding a local decoder (ffmpeg, mpg123, sox, or afconvert) that the skill does not install.