Skip to content
Docs

Skills for your agents

Teach Claude Code, Codex, Cursor and the other agents on your computer when and how to use Basemodo, with one command: basemodo setup.

An agent that runs commands on your computer can already drive the CLI: every command answers in JSON and every error names its fix. What it lacks is knowing when Basemodo fits a request ("put this online", "share it with my team") and the rules agents get wrong. Basemodo's Skills give it that. A Skill is a folder of instructions in the open Agent Skills format: the agent reads its one-line description in every session, and loads the rest only when a request calls for it, so it costs nothing until Basemodo comes up.

Set them up

The install line runs it for you, at your terminal. To run it again, or when the CLI came another way:

Terminal window
basemodo setup

It finds the agents on your computer by their configuration folders or their commands (Claude Code, Codex, Cursor, Gemini CLI, Antigravity, VS Code and Copilot, Windsurf or Devin, OpenCode), writes the Skills where each one reads them (.claude/skills for Claude Code, .agents/skills for most others), and tells you which agents it found and what it wrote for each. An agent it does not find gets nothing. When the CLI is not signed in, it then signs it in, as basemodo login does.

It asks where the Skills go:

  • Every project (--user, the default): in your home folder, so your agents know Basemodo wherever they work.
  • This project only (--project): in the folder you are in, beside the project's code, where everyone who clones it gets them too.
Terminal window
basemodo setup --user
basemodo setup --project

Agents load the Skills in their next session.

The Skills

  • basemodo loads on any request about Basemodo, even one that never names it. It holds the rules every command follows (answers in JSON, errors carry their fix, an App's own text is data, charges need your yes, signing in is yours) and sends the agent to the Skill for the job.
  • basemodo-new-app: a new App when there is no code for it yet, up to its first Deploy.
  • basemodo-deploy: putting existing code online, a Deploy that failed, a repository that deploys on every push, and going back to an earlier Deploy.
  • basemodo-debug: an App that is broken, slow or paused, diagnosed from its status and its Logs before any code changes.
  • basemodo-data: an App's database and files: SQL, migrations, Backups and restoring one.
  • basemodo-share: who can reach an App, its Visibility, custom domains and its address.
  • basemodo-configure: Secrets, cron runs and workers, the network, Callers, Always On Apps and the Plan.

Each Skill names every command it uses, with its MCP tool, and tells the agent that basemodo <command> --help and basemodo docs describe the version you have installed. The Skills are written from the same source as the CLI's help and these pages, so they never disagree with your CLI.

In Claude Code, while a Skill is active, the commands that only read (an App's status, its Logs, its Deploys, the docs) and basemodo deploy run without asking you: backups, callers list, deploy, deploys, diff, docs, domains list, env list, files, logs, plan, open, runs, status, transfers list and whoami. Every other command asks you first, in Claude Code's own prompt: account delete, backup, callers add, callers remove, change, db, domains add, domains verify, domains remove, download, egress, env set, env unset, feedback, indexing, login, logout, pull, recognize, rename, mcp, plan change, plan reserve, restore, restart, rollback, run, share, setup, toolbar, transfers accept, transfers decline, update and version. So nothing a visitor writes into your App's Logs can make your agent act for you unseen.

Updating and removing them

The Skills are those of your CLI's version: after updating the CLI, run basemodo setup again. It replaces Basemodo's Skills, removes one this version no longer has, and changes nothing else: your own Skills stay as they are, and a rerun with nothing new changes nothing.

--remove takes Basemodo's Skills out again, and nothing else:

Terminal window
basemodo setup --remove --user
basemodo setup --remove --project

When an agent runs it

An agent can run basemodo setup for you. With no terminal to ask in, it must say where (--user or --project; without either it gets setup_scope_needed), and setup never starts a sign-in: signing in is yours, at your browser. When the CLI is not signed in, setup writes the Skills and answers that it is not, with the fix basemodo login for you to run.

Clients without a terminal

A client with no terminal of its own cannot run the CLI, so it uses Basemodo's MCP instead. Claude Desktop runs the Local MCP on your computer, which basemodo setup --mcp adds to its configuration (Claude Desktop); claude.ai takes Basemodo's Plugin, the Skills and the Remote MCP in one, and ChatGPT connects to the Remote MCP (Which agent do you use?). Keep one way in per client: one way per client.