# The commands

At the end, you know every `loopgate` command, and which six your coding agent uses.

## At a glance

**Time:** 5 minutes to read.

**You need:**

- Loopgate on this machine: [On your laptop](laptop.md). `loopgate <command> --help` says every option.

## The six your agent uses

The skill teaches your coding agent these six: [From your coding agent](agent.md).

- `loopgate workflows`: the shelf, every workflow on one line with its purpose.
- `loopgate plan <workflow> "<task>" --out launch.yaml`: the run as a launch file you can edit; it starts nothing. `loopgate plan launch.yaml` checks an edited one.
- `loopgate run launch.yaml`: starts it. `loopgate run <workflow> "<task>"` starts without a launch file.
- `loopgate wait <run>`: returns when the run needs you, replies quietly, stops or ends. `--for <seconds>` returns a snapshot sooner.
- `loopgate decide <run> "<answer>"`: answers the open ask with your words: an option's number or name, Continue, Lock, yes or no.
- `loopgate tell <run> <step> "<text>"`: your words to a step, verbatim: a reply while it waits, a steer while it works, a note before it runs.

The rest are below, each written without its `loopgate`.

## The run

- `runs`: your runs, here and on a runner. `--q <words>` finds tasks by their words.
- `status <run>`: the track, its questions and the workbench link.
- `sent <run> <step>`: every part of the message a step was sent. `--round <n>` reads an earlier round.
- `result <run>`: the latest reply, the commits, the pull request and the cost.
- `inbox`: every run that waits on you.
- `open <run>`: opens the run in the workbench.
- `attach <run>`, also `watch <run>`: follows the run here and answers gates; Ctrl-C detaches.
- `message <run> "<message>"`: a reply to a waiting Interactive step, or to any step after the run ends.
- `note <run> <step> "<text>"`: tells a step something for this run, from its next turn.
- `steer <run> "<message>"`: direction to an agent while it works.
- `pace <run> <step> auto|interactive`: makes a step wait for you after each turn, or move on.
- `set <run> rounds <loop>=<n> budget <usd> engine <step>=<harness>/<model>[/<level>]`: changes a run's setup while it runs and says when each change lands. `set --help` lists every change.
- `retry <run> <step>`: retries a step that failed or was interrupted.
- `cancel <run>`: stops the run and keeps its branch.
- `resume pi <file> --cwd <folder>`: resumes a pi session in that folder, as `bring` prints it.

## Workflows and agents

- `workflows list`: the shelf with versions. `--full` prints every entry whole.
- `workflows show <id>`: one workflow's YAML.
- `workflows check <file>`: checks a workflow you wrote, then draws it.
- `workflows save <file>`: keeps it on your shelf. `--id <id>` revises one you saved.
- `workflows open <id>`: opens a saved workflow in the workbench with a fresh sign-in link.
- `workflows describe <workflow>`: the workflow drawn, then in plain words.
- `workflows validate <workflow>`: the validator's issues, or that it is valid.
- `workflows simulate <workflow>`: walks its scripted routes, spending nothing and changing nothing.
- `workflows schema`: the workflow JSON Schema.
- `guide`: how to write a workflow, block by block. `guide patterns` is the shelf.
- `agents list`: your agents.

## The installation

- `ui`: opens the workbench in your browser. `--no-open` prints the link.
- `daemon`: runs the local daemon in this terminal.
- `doctor`: checks the installation. `--bundle` saves private diagnostics.
- `engines`: the coding agents found and whether each is signed in. `engines install <engine>` installs the tested version; `engines login <engine>` signs in.
- `login`, `logout` and `whoami`: your Loopgate account on this machine.
- `skill install`: installs the skill for Claude Code, Codex and pi.
- `runner`: serves the workbench to your org at the runner's address.
- `team`: your org's members with their ids, for the Slack map.
- `send "<what next>"`, `bring <run>` and `moves`: move a session to your runner, bring it back, and list the moves: [Send a session](send.md).

## Good to know

- Most commands take `--json` for a script or an agent; `<command> --help` says which. `send --json` needs `--yes`.
- `--on <runner>` on `plan` and `run` starts on your team's runner, in its workspace with the same Git origin. Every later command follows the run there.
- A gate is yours. Your agent relays it and runs `decide` only with your words.

## Next

- [From your coding agent](agent.md)
- [Words we use](words.md)