determinism-patternlisted

# determinism-pattern **The problem this solves:** you do the same job over and over — the weekly client report, the new-lead intake, the "rewrite this in our voice" pass — and the result is a coin flip. Some days it's sharp. Some days it's thin and you don't catch it until a client does. When you're sitting there watching, you fix it on the fly. But the whole point of an always-on system is that you're *not* watching. A job that only comes out right when you babysit it is not automatable — it's just a manual task with extra steps. This skill fixes the **output-consistency** half of always-on. (Its sibling, `host-your-agent`, fixes the *get-it-running* half — the scheduled, unattended run. This one makes sure that when the run fires at 1am, what comes back is the same quality you'd have produced by hand.) The fix is one idea, borrowed from how software teams stop shipping bugs, applied to your repeatable work: > **One skill per repeatable process. Version it in a shared repo. Put a judge between the run and the "done."** Three moves. That's the whole thing. 1. **Skillify the process** — take the fuzzy job that lives in your head (and comes out different every time) and write it down as a *skill*: a single file that spells out the inputs, the exact steps, the format, and what "good" looks like. Now every run reads the same instructions instead of your mood that day. 2. **Version it in a shared repo** — the skill lives in a repo, not a chat window. It has a version number