Product loops

The loops that build and look after the product itself — coding tasks that open pull requests, a babysitter that keeps those PRs mergeable, error triage and investigation, the docs loop, feature inventory with browser QA, a daily app-health watchdog, user-directed QA, release notes, and Cast synthetic users.

Product loops do the engineering work: they read your connected repo, change or check the code, and hand back a pull request, a findings comment, or a report you review. Engineering loops mature first because the work is verifiable — tests and builds give a clean success signal — so several of these run on their own cadence. All of them depend on a connected GitHub repo.

Coding tasks

Ask for a change in plain language; get a pull request. Full features get a spec to approve first.

PR babysitter

Keeps the PRs 1mn opened mergeable when your base branch moves — re-merging and conservatively resolving conflicts, or flagging the ones that need you.

Error triage & investigation

Turn a captured crash or a reported symptom into a root cause — and a fix PR when it's small and low-risk.

Docs

Keeps this documentation in sync with your code, one reviewable PR at a time.

Feature inventory & QA

A canonical map of every feature, re-verified in a real browser as the code changes.

Watchdog

A daily browser health check that probes your critical flows and files a recording for every failure it reproduces.

QA

Point a browser session at one thing from a ticket — reproduce a bug, check the UX, measure friction — and get a recording back.

Changelog

Turn the commits you shipped into user-facing release notes, as a PR or a reviewable document.

Cast

Synthetic users that drive your app in a real browser and leave honest feedback.

They read the real code. Each loop shallow-clones your repo in an isolated sandbox and grounds its work in what the implementation actually does — not in a description of it.
They use your selected repo. If you've picked a repository in Integrations → GitHub, every loop uses it verbatim. Otherwise it picks the most likely repo and notes the assumption.
The artifact is the deliverable. A coding loop's output is the PR diff you review on GitHub; an inventory's output is the feature rows; QA's output is a verdict plus a recording. You stay the reviewer.

What stays gated

Code is reversible, so most of these loops can run autonomously — but merging is always yours. A loop opens a pull request; it never merges. The same goes for publishing: the docs loop and content loop hand you a PR, and your merge is what ships. The health loops go further and change nothing at all: Watchdog and QA are recommend-only — they drive a browser and report, never touching the code. See How loops work for why blast radius decides how far a loop is allowed to go.