Agent skills
Reviewed
🦾
Let your AI agent figure the details.
fortrabbit agent skills extend coding agents with domain knowledge of the platform — enabling guided setup, deployment, database sync, remote SSH commands, and MCP-based provisioning from a natural language prompt.
The skills work with Claude Code and OpenAI Codex. GitHub Copilot has basic support. The source is published at github.com/fortrabbit/agent-skills.
This is an early preview. The skill set covers the most common workflows already and is useful in day-to-day development. The MCP commands described here ship with version 0.3.
Two skills
The package ships two skills and installs both together. They divide the work between knowing how to do something and being allowed to do it.
| Skill | Covers |
|---|---|
fortrabbit | The workflows: deployment, SSH, database and content sync, and MCP-based discovery and provisioning |
fortrabbit-api-access | The credentials: connecting an agent to the MCP server, or issuing a token for the public API |
Install
Via GitHub CLI
Requires GitHub CLI v2.90.0 or later. Install globally — available across all projects:
gh skill install fortrabbit/agent-skills fortrabbit --agent claude-code --scope user
One command installs one skill, so run it a second time for the access skill:
gh skill install fortrabbit/agent-skills fortrabbit-api-access --agent claude-code --scope user
Install via curl
This installs both skills in one go:
curl -fsSL https://raw.githubusercontent.com/fortrabbit/agent-skills/main/install.sh | sh
See the GitHub repo for per-project installs, other agents, and update options.
Commands
Invoke the skill from an agent with /fortrabbit followed by a command:
| Command | Description |
|---|---|
start | Detect project state and guide to the next step |
connect | First-time setup: account, app, SSH key, connection test |
mcp | Discover, provision and diagnose apps through the MCP server |
deploy | Trigger a deployment via deploy hook or Git push |
sync up | Rsync all project files to the remote environment |
sync down | Rsync all project files from the remote environment |
db up | Upload the local database to the remote environment |
db down | Download the remote database to the local environment |
content up | Rsync CMS content to the remote environment |
content down | Rsync CMS content from the remote environment |
ssh | Run a remote command via SSH (artisan, craft console, Composer) |
update | Check for and install the latest version |
help | Show all available commands |
Connect the MCP server
Connecting is optional — the skills work over SSH without it. With a connection the agent can list apps, create apps and environments, and read deployment logs without a shell. One command, then an approval in the browser; no token is copied or stored:
claude mcp add --transport http fortrabbit https://api.fortrabbit.com/mcp
Codex does not publish a client metadata document, so it needs the client ID passed explicitly:
codex mcp add fortrabbit --url https://api.fortrabbit.com/mcp --oauth-client-id https://api.fortrabbit.com/.well-known/oauth-client/codex
Ask the agent to list the apps on the account to confirm the connection works. Details on the tools behind it are in the MCP server article.
Scripts and CI
Where a browser flow does not fit — a CI job, a shell script, an MCP client without OAuth support — a public API token takes its place. The fortrabbit-api-access skill looks for one in FORTRABBIT_API_TOKEN or a git-ignored .env before asking, and never writes it to a file that gets committed.
Requirements
sshandrsyncavailable locally- On Windows, WSL is required for rsync and the install script; SSH works natively on Windows 10+