wrap-it-up-auto

Logs durable decisions, lessons, and action items to
wrap-it-up's .claude-brain as they're confirmed in conversation -
not just when you run /wrap-it-up.

---

Why this matters

Session context is the thing you can't get back. Close the window before you remember to run `/wrap-it-up`, hit a context limit mid-task, or just get pulled away, and whatever you decided an hour ago is gone unless you reconstruct it yourself later.

`wrap-it-up-auto` removes the "did I remember to log this" step entirely. The moment a decision, a lesson, or an action item actually gets confirmed in conversation, it's already written to `.claude-brain`, before the session has a chance to end badly. You're not relying on your own memory, or on a single batch job at the end, to catch what mattered.

**One honest caveat:** this doesn't replace `/wrap-it-up` itself, and it isn't free, see "The cost, honestly" below for what that tradeoff actually costs on every turn.

What it does

`/wrap-it-up` captures a session's log in one batch, at the end, from a full reconstruction of what happened. This is the real-time companion: after every turn, a cheap background model call asks one question, *did anything durable just get confirmed?*, and if the answer is yes, it's logged immediately, in the same `Decisions` / `Lessons` / `Action items` sections `/wrap-it-up` itself writes to.

Conservative by design. Most turns log nothing: small talk, a question still being explored, a proposal nobody's confirmed yet. Only a decision that was actually *made* (not just floated), a real lesson (a bug, a gotcha, a surprising fact), or a concretely open action item gets written - the same honesty standard `/wrap-it-up` itself holds. When the judge isn't sure, it logs nothing; a missed entry costs nothing, a premature one pollutes the record.

📝 w