/build-and-verify
Work from the project directory. Once eve is installed, the full docs are bundled with the installed package and match its version exactly. In most installs, they are at `node_modules/eve/docs/`. In workspaces or local package installs, resolve the installed `eve` package
$ npx -y skills add vercel/eve --agent claude-codeHow it fires
How this command gets triggered: by you, by Claude, or both.
- Fires itselfClaude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/build-and-verify
Context preview
What this command does when you run it.
Work from the project directory. Once eve is installed, the full docs are bundled with the installed package and match its version exactly. In most installs, they are at `node_modules/eve/docs/`. In workspaces or local package installs, resolve the installed `eve` package
Command definition
build-and-verify.mdBuild it out, then verify
Work from the project directory. Once eve is installed, the full docs are bundled with the installed package and match its version exactly. In most installs, they are at `node_modules/eve/docs/`. In workspaces or local package installs, resolve the installed `eve` package location first and read its `docs/` directory. If package docs are unavailable, use https://eve.dev/docs as a fallback. Read `README.md` in the package docs first, then the guide for what you're adding, such as `connections`, `channels/slack`, or `guides/auth-and-route-protection` for the Vercel Connect flow.
Before implementing an integration yourself, use `eve registry search <query>` or `eve registry list` to discover available integrations. Inspect one with `eve registry view <item>`, then install it with `eve add <item>`.
- Put the purpose in `agent/instructions.md` (the always-on system prompt),
replacing the scaffold's placeholder with what the user said the agent should do.
- Add a first typed tool under `agent/tools/` with `defineTool` from `eve/tools`
and a Zod `inputSchema`.
`{{devCommand}}` starts eve's HMR development server and opens the agent's terminal REPL. It does not start or control this coding-agent session, so don't use the bare command as a background verification process. Start eve without the terminal UI in a controllable background process instead:
{{devCommand}} --no-ui
Wait for the server URL, then exercise the HTTP API: create a session with `POST /eve/v1/session`, attach to `GET /eve/v1/session/:id/stream`, and send a follow-up to `POST /eve/v1/session/:id`. Stop the dev process after verification.
When the user is ready to use their agent's REPL, give them the interactive command to run from the project directory:
{{devCommand}}
Verify the project's typecheck passes, adapt the model and provider to the user's data and use case, and don't commit unless the user asks.
Read more
Build it out, then verify
Work from the project directory. Once eve is installed, the full docs are bundled with the installed package and match its version exactly. In most installs, they are at `node_modules/eve/docs/`. In workspaces or local package installs, resolve the installed `eve` package location first and read its `docs/` directory. If package docs are unavailable, use https://eve.dev/docs as a fallback. Read `README.md` in the package docs first, then the guide for what you're adding, such as `connections`, `channels/slack`, or `guides/auth-and-route-protection` for the Vercel Connect flow.
Before implementing an integration yourself, use `eve registry search <query>` or `eve registry list` to discover available integrations. Inspect one with `eve registry view <item>`, then install it with `eve add <item>`.
- Put the purpose in `agent/instructions.md` (the always-on system prompt),
replacing the scaffold's placeholder with what the user said the agent should do.
- Add a first typed tool under `agent/tools/` with `defineTool` from `eve/tools`
and a Zod `inputSchema`.
`{{devCommand}}` starts eve's HMR development server and opens the agent's terminal REPL. It does not start or control this coding-agent session, so don't use the bare command as a background verification process. Start eve without the terminal UI in a controllable background process instead:
{{devCommand}} --no-ui
Wait for the server URL, then exercise the HTTP API: create a session with `POST /eve/v1/session`, attach to `GET /eve/v1/session/:id/stream`, and send a follow-up to `POST /eve/v1/session/:id`. Stop the dev process after verification.
When the user is ready to use their agent's REPL, give them the interactive command to run from the project directory:
{{devCommand}}
Verify the project's typecheck passes, adapt the model and provider to the user's data and use case, and don't commit unless the user asks.
Repo: vercel/eve
Other commands on eve.
- /collect-intent
If the user's intent is not already clear, ask the questions one at a time, use the coding harness's prompt tools when available, and do not guess.
Open command - /intro-handoff
The project at `{{workingDirectory}}` is already scaffolded. Work with the user to complete it, running the commands below from that directory.
Open command - /intro-setup
eve is a filesystem-first framework for durable backend AI agents. The agent is a directory of files that eve compiles and runs: its instructions, tools, connections, and channels. Work through this with the user.
Open command - /scaffold
First settle the target: a new project, or the agent added to an existing directory? For a new project, propose a name and ask the user to confirm it; for an existing one, ask for the directory.
Open command - /vercel-connect
Reach for Vercel Connect for both channels and connections rather than hand-managing tokens. It is the path to a working setup out of the box. See https://vercel.com/docs/connect.
Open command

