Agent skills for Prevu - drop-in instructions that teach Claude Code, Cursor, OpenHands, and other coding agents how to drive the Prevu CLI.
npm install -D @prevu/skills
# Sync into your agent of choice
npx skills experimental_sync -a claude-code
# or: -a cursor / -a openhands / etc.This symlinks the skills from node_modules/@prevu/skills/skills/ into the agent's skills directory.
Alternative - install from the GitHub mirror without npm:
npx skills add prevu-cloud/prevu -a claude-codeTwo skills, designed around how agents actually load context (description always-on; body on trigger; references on-demand):
The reference skill. Triggers when the user mentions Prevu, the CLI, env work, tokens. Body is a short overview + command map; deep detail lives in references/:
references/auth.md- sign-in, token storage, recovering from auth errors.references/env.md- every env subcommand with examples + SSH key resolution rules.references/keys.md- saved SSH key model.references/lifecycle.md- status state machine and destructive-action guidance.references/troubleshooting.md- concrete failure modes and fixes.
The workflow skill. Triggers on multi-step user scenarios:
references/mirror-dev.md- AFK loop: mirror local dev to a Prevu env so the user can preview changes from their phone while iterating with the agent in chat. Edit-locally -> push -> env-pulls -> preview -> review -> repeat.references/share-preview.md- share a WIP branch as a public preview URL for a teammate, designer, PM, or future-self on another device.
Skills compete for "always-on" context budget via their description. Anthropic's own skill-creator guidance pushes domain organization (one skill per domain, references on-demand) over command-per-skill splits. Prevu's auth/env/keys always co-occur - splitting them just multiplies the always-on tax. The flows are scenario-narrative with distinct trigger phrases, so they get their own skill.
Independent SemVer (0.2.0 at launch). Tracks CLI majors (0.x skills with 0.x CLI).
MIT