Context Switch Budget

Context Switch Budget Skill

The calendar shows eight meetings totaling four hours and calls the other four "free" — but fifteen-minute gaps between meetings are free the way confetti is paper: technically yes, usable no. Each switch costs re-entry time (the mind reloads the last context for 10–20 minutes), and a day of eight switches can contain zero real work while looking half-empty. The budget discipline: census the week's actual switches (the number shocks reliably), batch the same-kind work (meetings to meeting-blocks, reviews to review-hours, chat to email-triage-system-style windows), defrag the calendar (consolidate the gaps into slabs worth having), and price new fragmentation at the door.

Not quite this? Use context-budget when you mean an LLM's context window and token budget.

What This Skill Produces

• The switch census — the week's transitions counted and typed (meeting↔work, task↔task, interrupt-driven), with the fragmentation map
• The batching plan — same-kind work consolidated: the meeting corridors, the review hour, the comms windows
• The defrag moves — the specific reschedules that convert cheese-holes into slabs (deep-work-blocking receives them)
• The door pricing — the phrases that protect the defragged shape ("Tuesday corridor?" as the default counter-offer)

Required Inputs

Ask for these if not provided:

• A real week's calendar — last week, as it actually ran (including the unscheduled interruptions the user remembers); the census works on evidence
• The work's kinds — what recurs: meetings, reviews, writing, admin, comms; batching needs the categories named
• The movable vs. fixed — which meetings the user controls or can counter-propose (recurring-meeting-pruner verdicts feed this), and the team's collaboration-hours constraints
• The interrupt sources — what punches through ad hoc (chat, drive-bys, "quick calls") — each gets a batching destination or a boundary

Framework: The Budget Rules

1. Census first — the number is the intervention: count last week's switches (every transition between unlike work counts; every meeting-to-meeting gap under 30 minutes counts as a lost slab). The typical knowledge-week runs 30–50; seeing "41 switches, longest unbroken stretch: 52 minutes" reframes the problem from "I'm unproductive" to "my week is structurally incapable of depth" — which is true, and fixable.
2. Batch by kind, on a rhythm: meetings compress into corridors (Tue/Thu afternoons — the counter-offer default) · reviews/approvals into a daily half-hour · comms into 2–3 windows · admin into one Friday hour. Same-kind work shares context; batching converts eight reloads into one.
3. Defrag the residue: the gaps that remain after batching get consolidated — the two 45-minute holes become one 90-minute slab by moving one meeting ("could we do 2pm instead of 3?"— movers rarely mind, they just never think to ask). Slabs go to the deep-work design; the census re-runs in two weeks to verify the shape held.
4. Interrupts get destinations, not resistance: the chat that punches through routes to the comms windows (with the office-hours-design never-waits line honored) · drive-bys get the warm deferral ("in the middle of a thing — grab me at 3?") · the genuinely-urgent keeps its lane, defined. Fighting interrupts case-by-case loses; routing them by class holds.
5. Price fragmentation at the door: every incoming meeting request gets the corridor counter-offer by default ("Could this land Tuesday PM? I batch meetings there") — most acceptors don't care where, so the asker's shape wins by initiative. The switch-cost line, for when justification helps: "a 3pm call splits my only deep block — the 10am corridor keeps me sharp for it anyway."