vanillaskylisted
Install: claude install-skill VanillaSkyAi/skills
# VanillaSky — editable social videos from a prompt
Keep the user experience simple: **Describe → Review → Export.** Turn the brief
into a visually inspected, editable first cut; open it in VanillaSky Studio for
review; let the user export the MP4 there. Internally the working video is a
`VideoConfig` JSON, but do not make config mechanics the user's workflow and do
not make a full MP4 render part of the initial flow.
## Three non-negotiables
1. **Read the catalog; never guess.** After resolving the CLI, run
`vanillasky templates` for the bundled template map and
`vanillasky templates <id> --json` for an exact schema. The live fallback is
`https://vanillasky.ai/llms-components.txt`; repo and standalone installs
carry `registry/llms-components.txt` plus `registry/r/<name>.json`. If no
exact catalog is reachable, stop rather than inventing an id or variable.
2. **Never invent proof.** Names, quotes, numbers, stats, screenshots, and
product claims must come from the brief or supplied assets. If a template
requires evidence the user did not provide, choose another template.
3. **Never hand off an uninspected config.** Validation catches structure;
only looking at rendered frames catches overflow, weak composition, and
accidental defaults.
## Start the session
First resolve the CLI. The official VanillaSky CLI is **not published as an
npm package**. Never search npm, run `npm install vanillasky`, or use a package
that merely owns that name. If `comm