Examples

📖 loopsmith's documentation lives at bitphill.github.io/loopsmith → This wiki mirrors it so existing links keep working.

← Wiki home · Rendered viewer with search

---

The example loops

Worked loops in config/examples/, each shipped as a .yaml and an equivalent .md. The two are the same model in two grammars, generated from one source by tools/sync-examples.sh, and a round-trip test keeps them honest.

All of them refuse to validate, with exactly one error each, until you tick their intent.prerequisites. That is not an oversight. An example config is a description of a job nobody has done yet, and the tool's single most valuable refusal is the one that says so.

loopsmith loop validate config/examples/research-loop.yaml
#   error  intent.prerequisites: 2 step(s) not marked done: …

Copy one, do the job by hand once, tick the steps with what you actually did, and then run it.

Loop	What it does
research-loop	Research a question against primary sources, every claim cited
refactor-loop	Behaviour-preserving refactor where the test suite is the gate
traffic-loop	Post where an audience already gathers, measured in referred sessions
trend-radar-loop	Track a category across X, Instagram, and TikTok with dated evidence
landing-page-loop	A static landing page gated on Lighthouse, page weight, and working CTAs
sales-leads-loop	Build a lead list from permitted sources, with lawful basis recorded per record
marketing-automation-loop	Turn product docs into scheduled posts, published behind a human checkpoint
blogger-loop	Write on a trending topic, gated on style measurements and an independent read
cold-outreach-loop	Personalised first contact, with suppression and opt-out enforced by the gate
x402-agent-loop	An agent that pays for things, supervised or autonomous, under a hard cap
viral-game-loop	A small Godot game gated on build health and time-to-first-play
idea-radar-loop	Product ideas traced to dated public complaints, checked against what already sells
account-watch-loop	Watch accounts for pre-viral topics, and score yesterday's predictions
container-refactor-loop	Refactor three modules at once, each in its own container over its own worktree
self-tuning-loop	A weekly report that may propose changes to itself, measured against a recorded baseline

Which one to start from

• Learning what a loop is — research-loop. It is the smallest one whose checks are all deterministic.
• You write software — refactor-loop. The test suite is already the gate, so there is nothing to invent.
• Anything that leaves the building — cold-outreach-loop or marketing-automation-loop. Both show human_checkpoint doing real work.
• Anything that spends money — x402-agent-loop. It is the one built around a hard cap.
• Wide parallelism — container-refactor-loop. Three builders in one wave, each in its own container, with the test suite as the gate.
• Self-evolution — self-tuning-loop. It is mostly about the fence: the baseline a proposal is measured against and the components it may never touch.