loopmanager 0.1.0

Validate, visualize and sync Loop Architecture (loopmanager/v1) loops to Claude Code routines.

loopmanager

Preview. loopmanager is an early preview: the YAML format and the commands may still change, and it targets Claude Code routines, itself a research preview.

loopmanager is the CLI for Loop Architecture, an architecture style that makes the agentic loop between systems the first-class unit of design.

It validates, visualizes, and syncs a Loop Architecture, one YAML document that defines many systems and many loops, and turns each loop into a runnable Claude Code routine.

Install

uv tool install loopmanager
# with PNG visualization support (needs native Cairo, e.g. `brew install cairo`):
uv tool install "loopmanager[png]"

Or run without installing: uvx loopmanager --help.

Quick start

loopmanager init your-org                       # scaffold <id>.loopmanager.yaml
loopmanager validate your-org.loopmanager.yaml     # schema + lint
loopmanager view your-org.loopmanager.yaml         # open the interactive diagram in a browser
loopmanager serve your-org.loopmanager.yaml        # run the whole architecture locally + monitor
loopmanager sync your-org.loopmanager.yaml      # every loop → a Claude Code routine

Commands

Command	What it does
loopmanager init <id>	Scaffold a new <id>.loopmanager.yaml.
loopmanager validate <file>	Validate against the schema and lint. Exit non-zero on errors.
loopmanager lint <file>	Best-practice / safety checks.
loopmanager view <file>	Open the interactive diagram in a browser (via the visualizer).
loopmanager serve <file>	Run the whole architecture locally: schedule, trigger agents, live monitoring.
loopmanager sync <file> [loop-id]	Sync all loops (or one) to Claude Code routines.
loopmanager sync <path> --from-claude	Reverse: reconstruct a Loop Architecture YAML from existing routines.

serve

loopmanager serve <file> starts a local server that schedules every loop from the YAML (5-field cron and one-off ISO triggers), triggers the agent for each run, tracks run history, and serves the diagram plus live monitoring (per-loop badges: running / last run / outcome) at http://127.0.0.1:8700.

Runs are dry-run by default (safe, they only simulate a turn). Pass --exec to actually run each agent with claude -p. Trigger a loop by hand from the diagram's detail panel ("Run now").

view writes a small HTML page that embeds the YAML and loads the shared visualizer (visualizer/dist/visualizer.js), which parses the YAML and builds the diagram itself, systems, loops, layout and favicons all live in the visualizer.

What sync produces

Routines are cloud-managed (they live in your claude.ai account, created with /schedule), so cloud is the default. For each loop, sync writes the slash command and prints the /schedule step:

.claude/
  commands/loop-<id>.md    # the prompt Claude Code runs (what to do + the systems it uses)

Run a loop once in Claude Code with /loop-<id>, or register it as a routine on its cadence with /schedule at claude.ai/code, using the printed cron/event.

--local additionally writes a .claude/routines/<id>.json descriptor (schedule/event, model, repositories, connectors), an offline record that sync --from-claude can read back.

The format

See the specification and the JSON schema. A full example lives in examples/your-org.loopmanager.yaml.

Development

uv sync --extra dev
uv run pytest