the-loop CLI

the-loop is a lightweight, extensible command-line companion to the the-loop plugin. The plugin is the operating model an agent follows inside a session; the CLI is what starts those sessions, keeps them attached to work items, and tells you what happened.

pip install the-loopy-one
the-loop --help

You do not need it

Worth saying plainly, because "there is a CLI" reads like a prerequisite. It is not. Run /the-loop:work-on <ticket> in Claude Code or Cursor and the whole loop works with no CLI installed at all.

You want the CLI when you stop driving the loop by hand:

You want to…	Use
Have a comment on a GitHub issue start an agent, unattended	start with the webhook receiver or polling enabled
Keep one session per work item and route later activity to it	sessions
Watch an agent work, live, and type into it	sessions attach
Answer "why did nothing happen?"	events
Gate CI on a work item's own phase rules	check
Hand a review round to a different model	critic
Run more than one the-loop against a repository, each with its own work items	the instance block — see running several instances

What it is

A Python package — import package the_loop, console script the-loop, published to PyPI as the-loopy-one — with an extensible command registry.

• One runtime dependency, PyYAML. Its whole configuration is YAML, so reading it is not optional (decision-038); everything else is stdlib.
• Python is deliberate. It leaves room to add self-learning / ML capabilities later, which are mostly exposed as Python SDKs.
• Extensible by design. Commands are discovered from a registry — see adding a command.

Two halves

Some commands are daemons: long-running, machine-scoped, watching several repositories at once. Others are repo-scoped: they run once, inside a checkout. Worth knowing which is which, because it decides what you have to configure to run them — a daemon needs a cli-config.yaml, a repo-scoped command needs nothing but the checkout.

Note the arrow that is not there: nothing in the CLI reads a repository's harness config. Since issue #352 that file is the agent's alone; the spec directory, the critic roster and the graph hooks are the CLI config's, and anything else the CLI needs from a repository the agent hands over as a flag. The rule is decision-123; the split into two files is decision-032, and both files are explained in Configuring the-loop.

Next

• Installation — get it on your machine.
• Getting started — from nothing to a work item an agent picks up by itself.
• Concepts — the model the command pages assume.
• State on disk — what it writes, what is inside, and what you can carry to another machine.
• Commands — the full reference.