gandalf init: create a team manifest scaffold - Gandalf

gandalf init scaffolds a team agent manifest at your project root. It writes a starter gandalf.toml and a .gandalf/skills/ directory so teammates can standardize MCP servers, skills, and hooks across Codex and Claude Code.

Synopsis

gandalf init [flags]

Flags

Flag	Description	Default
`--name`	Project name in the manifest	current directory name
`--from-current`	Scaffold from a scan of local agent setup	off
`-f`, `--force`	Overwrite existing `gandalf.toml`	off
`-p`, `--project`	Project directory	`.`
`--home`	Home directory	`$HOME`
`--store`	Gandalf store directory	`~/.gandalf` or `$GANDALF_STORE`
`--json`	Emit JSON output	off

Examples

gandalf init --name my-team-project
gandalf init --from-current
gandalf init --force

Generated file

The default template includes an MCP server, skills scaffold, a hook, and an env template:

gandalf.toml

version = "1.0"
name = "my-team-project"
description = "Standardized AI agent environment for my-team-project"
agents = ["claude-code", "codex"]

[mcp_servers.postgres-db]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-postgres", "${DATABASE_URL}"]
required_env = ["DATABASE_URL"]
description = "Shared staging database MCP server"

[env_template]
DATABASE_URL = "postgres://user:password@localhost:5432/db"

With --from-current, existing local MCP servers become [mcp_servers.<name>] entries seeded with the source agent, ready to fill in command, args, or url.

Next steps

Edit gandalf.toml to declare team MCP servers, skills, and hooks.
Commit gandalf.toml and .gandalf/skills/ to Git.
Teammates run gandalf apply to sync their local setup.

⌘I