actor.sh
Reference

CLI Reference

Every actor subcommand with its flags and a worked example.

Complete reference for the actor CLI. Each section covers one subcommand: its flags, defaults, and a minimal example. Run actor --help or actor <cmd> --help for the in-terminal version.

actor main

Launch the orchestrator session. Resolves the main role from settings.kdl, takes its prompt as --append-system-prompt, and execs claude with the actor channel enabled. Trailing arguments are forwarded to the agent CLI verbatim.

The built-in main role ships as the main actor. Override it with a role "main" { ... } block in settings.kdl to swap in your own prompt or agent. Today only agent "claude" is supported on this command; other agents fail with a clear error.

actor main                                # open an interactive orchestrator session
actor main "kick off the refactor"        # one-shot
actor main --model opus                   # forward flags to claude

actor new

Create a new actor. If a prompt is given (positional or piped via stdin), the actor is also run immediately after creation.

FlagDescription
nameActor name. Becomes the git branch and worktree directory name.
promptOptional task prompt. If omitted and stdin is piped, stdin is read.
--dir PATHBase directory for the worktree. Defaults to the current working directory.
--no-worktreeSkip worktree creation; run in the directory directly.
--base BRANCHBranch to create the worktree from. Defaults to the current branch.
--agent NAMECoding agent (claude or codex). Defaults to the role’s agent or claude.
--role NAMEApply a role from settings.kdl. See actor roles for available names.
--model NAMEShorthand for --config model=<name>.
--use-subscription / --no-use-subscriptionForce or disable subscription auth. Strips the agent’s API key env var when on.
--config KEY=VALUEAgent-arg override. Repeatable. Saved as the actor’s default.
actor new my-feature                                      # create worktree, no run
actor new my-feature "fix the nav bar"                    # create and run
actor new my-feature --role qa                            # apply a saved role
actor new my-feature --no-use-subscription                # pass API keys through
actor new my-feature --no-worktree                        # use current directory
actor new my-feature --base develop                       # branch off develop
actor new my-feature --config effort=max --model opus     # set agent config at creation
echo "fix it" | actor new my-feature                      # piped prompt

--config keys collide-check against the agent’s ACTOR_DEFAULTS whitelist; passing an actor-key (e.g. use-subscription) through --config is rejected with a hint pointing at the dedicated flag.

actor run

Run an existing actor with a prompt.

FlagDescription
nameActor name.
promptThe task. If omitted and stdin is piped, stdin is read. Required unless -i.
-i, --interactiveResume the actor in interactive mode (TTY passthrough).
--config KEY=VALUEPer-run override only — not saved. Repeatable.
actor run fix-nav "continue fixing"                       # one-shot
actor run fix-nav --config model=opus "one-off"           # temporary config
actor run fix-nav -i                                      # resume interactively
echo "fix it" | actor run fix-nav                         # piped prompt

The CLI prints the agent’s response to stdout and propagates a non-zero exit code if the run failed.

actor list

List actors. Optional status filter.

FlagDescription
--status STATUSFilter by status (running, done, error, etc.).
actor list                                                # all actors
actor list --status running                               # only running

actor roles

List roles defined in settings.kdl (built-in main plus user-level and project-level roles).

actor roles

actor show

Show full details for an actor, including recent runs.

FlagDescription
nameActor name.
--runs NNumber of recent runs to display. Default 5. Pass 0 to omit.
actor show my-feature
actor show my-feature --runs 20
actor show my-feature --runs 0

actor logs

View agent session output.

FlagDescription
nameActor name.
-v, --verboseInclude tool calls, thinking, and timestamps.
--watchStream output live as it’s written.
actor logs my-feature
actor logs my-feature --verbose
actor logs my-feature --watch

actor stop

Kill the running agent for an actor. The actor row stays; only the live process is terminated.

actor stop my-feature

actor config

View or update an actor’s saved config. With no pairs, prints the current config; with one or more KEY=VALUE pairs, updates those keys. Updates take effect on the next run.

actor config my-feature                                   # view
actor config my-feature model=opus                        # update one key
actor config my-feature model=sonnet effort=max           # update several

actor mcp

Start the MCP server over stdio. This is the entrypoint Claude Code uses to spawn the actor MCP — you typically don’t run it by hand.

FlagDescription
--for HOSTCoding-agent host this server is serving (e.g. claude-code).
actor mcp --for claude-code

actor watch

Open the dashboard TUI. Animated splash by default; use --no-animation over slow links.

FlagDescription
--no-animationDisable splash animation.
actor watch                                               # local TUI
actor watch --no-animation                                # skip splash

actor discard

Remove an actor from the database. The worktree directory stays on disk and the underlying git branch is left in place — recover the name later by running git branch -D <name> in the source repo if needed.

FlagDescription
nameActor name.
-f, --forceBypass on-discard hook failure.
actor discard my-feature
actor discard my-feature --force

actor setup

Install (or reinstall) an integration. Idempotent — safe to re-run.

FlagDescription
--for HOSTRequired. Currently claude-code (skill + MCP) or omarchy (theme-set hook).
--scope SCOPEuser (default), project, or local. Ignored for --for omarchy.
--name NAMEMCP registration name. Default actor. Ignored for --for omarchy.
--uninstallRemove a previously installed integration (currently only supported for --for omarchy).
actor setup --for claude-code                             # user-wide install
actor setup --for claude-code --scope project             # project-local
actor setup --for claude-code --name actor-dev            # alternate MCP name
actor setup --for omarchy                                 # theme-set hook
actor setup --for omarchy --uninstall                     # remove the hook

actor update

Refresh deployed skill files after upgrading actor-sh. Use setup for a fresh install; update is a lightweight refresh that keeps existing registration in place.

FlagDescription
--for HOSTCoding-agent host. Default claude-code.
--scope SCOPEWhich install to refresh. Default user.
--name NAMEMCP name used at setup time. Default actor.
actor update                                              # refresh user-wide install
actor update --scope project                              # refresh project-local

actor –version

Print the installed actor-sh version.

actor --version
actor -V

See also