Coding agents

hanzo code runs a coding agent against the repository you are standing in. Three agents are supported, and naming one gets you that agent or a clear failure — they are separate products, never aliases of each other.

hanzo code                      # Hanzo Dev, the default
hanzo code claude               # Claude Code
hanzo code codex                # Codex

Give it a task as one argument, quoted:

hanzo code "fix the failing test in api/users"
hanzo code claude "port this module to the new store"

Naming the agent

Every spelling below resolves in one place, so they cannot drift apart:

hanzo dev is a top-level shorthand for hanzo code dev. claude-code and cc are accepted spellings of claude.

Naming it twice is refused rather than resolved — hanzo code claude --codex tells you to name it once.

Approvals

The agent auto-approves its own steps by default. Two ways out, in increasing order of how much they give up:

hanzo code --ask            # the backend asks before each step
hanzo code --no-sandbox     # bypass approvals AND drop the sandbox

--safe is a synonym for --ask, and either wins over autoApprove in ~/.hanzo/settings.json.

The other hanzo code

hanzo code context, hanzo code file and hanzo code index are a different thing that shares the word: they are the code intelligence API — search and symbols across your indexed repositories, for you and for agents. They take no backend and run no agent.

The rule is the first argument: a word that names a backend selects one, and anything else is a task or a subcommand.