## Get a token Your token is the key your agent uses to reach your private protocols and your repo. Keep it handy — in step 2 you'll type it into a hidden prompt in your own terminal, never into the chat. [Sign up + create a token](/api-keys) ## Paste this into Claude Code The install protocol has your agent set up [curly](https://github.com/wildeagency/curly), then step aside: it tells you to run `curly init protocol` in your own terminal, where you paste your token into a hidden prompt. `Let's use protocol. Set it up by following https://api.protocol.supply/@protocol/install.md` No token in the snippet — your token never passes through the agent, the chat, or this page. It goes straight from your clipboard into `~/.curly.yaml` (chmod 600). ## Run a checkup The checkup inspects your setup — harness file, always-loaded context, wiring — and prescribes the single next improvement. The first thing it usually finds: the bloated context file this site promised to shrink. Re-run it any time; the delta is your progress. `Fetch https://api.protocol.supply/@protocol/checkup and run it.` Protocol continuously improves. Protocol teaches you about protocol. In a world of continuously evolving agents, the data layer is the foundation, and an intelligent data layer is a superpower. ### To get the most out of protocol - **Have it learn from your current workspace, projects, and skills.** Point it at your existing files and let it extract the protocols you've already been writing without calling them protocols. - **Edit your bootstrap to fine-tune how your sessions start.** [Open bootstrap settings →](/settings/bootstrap) - **Use impulses** — an agent-friendly model for tasks, projects, sessions, and responsibilities. - **Allow it to continually improve.**[See the continuous-improvement protocol →](/@protocol/retro)