warm-handofflisted

# warm-handoff — ride the cache wave, hand off before it breaks 🏄 **One sentence:** Handoff → fresh session → handoff. The user collects everything (test answers, ideas, critique) in ONE handoff file over hours, starts a fresh session, the agent works it all off through a few subagents and writes the next handoff. Full history with dated user quotes and the reasoning behind every rule: `references/historie.md` (read it when a rule here seems odd — do not load it by default). ## Die sparsamste Arbeitsweise in fünf Sätzen For the user, in plain words (asked for on 30.08.2026, 15:06: *"was ist denn jetzt die Moral der Geschichte?"*): 1. **Eine Session pro Welle** — plan, guardians, merge, handoff all in one; a fresh session costs its ~200k start build, so only start one past ~200k context. 2. **Sammeln kostet nichts** — collect test answers, ideas and critique in the handoff file for hours; that is zero requests, while every chat message costs a full context re-read. 3. **Wächter statt Pings** — one guardian per topic starts the workers, so the main session is woken 5–7 times instead of 15–20. 4. **Unter 200k Kontext bleiben** — the main session only plans, merges and reads reports; everything else lives in agents. 5. **Handoff mit Kostentabelle** schließt die Welle ab — measured numbers, not feelings. **Vorsatz (30.08.2026):** measure EVERY session for the next few days — one logbook line each (requests, equivalent, pings, wall clock, jobs) — so this skill get