Quickstart

Get a loop running in three steps

A short handoff for humans and agents: pick a template, install it, run checks, and stop at approval gates.

Three fetches to a loopAn agent fetches llms.txt, then the catalog, then a loop manifest.llms.txtcatalogloopyamlguidecatalogjob
Agents reach a runnable loop in three fetches.

1. Hand off

Coming from the homepage Loopify box? That command is the fastest path for a workflow you already run by hand. This prompt is the template-first version of the same rules.

Paste this to any capable agent — Claude, Codex, Hermes, or your own:

Read https://loopmaster-ai.pages.dev/llms.txt — it explains how this site works
for agents.

I want to turn a job I do by hand into a loop. Interview me first: what
the job is, when it should run, what it may touch, and what would prove
it worked. One question at a time; skip anything I've already told you.

Then use Loopmaster's quickstart and skills index to pick the best starting
template or skill. Before you install anything, tell me:
1. what you'll install and where it will run,
2. which checks will prove it's working,
3. what needs my approval — anything that runs on a schedule, spends
   money, publishes publicly, or needs new credentials.

Nothing goes on a schedule and nothing gets published until I approve it.

No goal in mind yet? Start with the flagship skill:

Fetch https://loopmaster-ai.pages.dev/skills/deploy-your-first-loop/SKILL.md and follow it.

It deploys the research-watch template — a loop that watches sources you choose and writes a cited report — and proves the install with a dry run.

2. Approve

Your agent should stop and ask before anything that publishes, spends, deletes, or needs credentials beyond the template's declared scopes. That is not a limitation; it is the design.

Review what it proposes, approve only the safe parts, and keep any secret values outside the transcript.

3. Verify

A loop is not running until its checks pass. Each template declares check commands in its manifest and install guide. Your agent runs those commands and reports the results.

Green means running. Anything else means it escalated instead of guessing.

Under the hood

Your agent's short fetch path is:

1. GET /llms.txt                     how the site works, for agents
2. GET /registry/index.json          the template catalog
3. GET /templates/template-id/loop.yaml

For skills, agents can also fetch:

GET /skills/index.json
GET /skills/skill-id/SKILL.md

The HTML pages are for humans. The text and JSON surfaces are for agents.