Reference

Workflow

For your agentCopy this and give it to the AI you're building with.

Run `cubek skill`, read all of it and follow it exactly (no CLI: download https://cubek.dev/SKILL.md verbatim with curl; don't summarize it); when you change this project, follow its "Workflow" section.
Run `cubek check` after every edit until it is clean.

curl -fsSL https://cubek.dev/install | sh  # macOS, Linux; Windows: WSL
cubek init my-app --empty  # a real app (no --empty: an example to delete)
cubek dev              # http://127.0.0.1:8787, recompiles on save
cubek dev --stop       # this project's server only (never pkill)
cubek check            # compile, after every edit
cubek test             # bun test *.test.ts
cubek run src/lib/x.ts # runs its `main()`
cubek login            # prints a URL: give it to the user; finishes by itself
cubek deploy           # preview https://<name>--r<N>.cubek.app (members only)
cubek promote <N>      # that preview to production (deploy --prod: a new one)

Apply each error's canonical form:. check lists route classes (db writes run on the primary: keep them in few routes). All commands take --json, --help.

Deployed: logs [--follow] [--route /x], sql "SELECT ...", secrets set NAME VALUE, rollback <rev> (deploy.md). Local: sql --dev "..." [--write], secrets --dev set, task run <name> (tasks.md).
MCP (claude mcp add cubek -- cubek mcp): the CLI, sdk_types (the exact API as .d.ts), whoami (on 401/403).