larkcowork/lark-cowork-plugins/calendar-optimizer

calendar-optimizer

"5 hours/day in meetings" → "audit them, cut 30%." 30-day pattern analysis → specific decline/merge/async recommendations.

Verbs (the installed lark-* skills): calendar +agenda (--start/--end/--calendar-id), vc +notes (--meeting-ids). Atomic shapes + token flags: connectors/LARK-RECIPES.md.

Workflow

1. Read memory. working-hours.md (deep-work blocks),

priority-people.md (VIP weight), delegation-rules.md (exec).

1. Fetch 30-day history. `calendar +agenda --start <t-30d>

--end <today> --jq '.data[]|{summary,start:.start_time.datetime,end:.end_time.datetime,attendees}' (--jq and --format are mutually exclusive — --jq projects; start_time/end_time are nested {datetime,timezone}` objects).

1. Classify each meeting: recurring vs one-off (repeat pattern);

user role (chair/required/optional/FYI); decision vs status (check minutes for "decided/approved/chốt" via vc +notes --meeting-ids <id>); outcome quality (action items vs none); duration.

1. Value score: chair +3 · VIP attendee +2 · decision keyword in

minutes +3 · has action items +2 · zero outcome 3+ in a row −3 · replaceable status sync −2 · conflicts deep-work block −2.

1. Bucket + report: DECLINE (score <0) · SHORTEN (60-min finishing

in 30) · CONVERT-TO-ASYNC (status sync) · KEEP. Include numbers header (total meeting h / %, recurring count, deep-work h left) + projected savings + action checklist.

Hard rules

1. Never auto-decline — drafts only, user confirms.
2. Decline politely — include "thanks, please share minutes", never ghost.
3. Don't decline VIP meetings even if low value — flag for manual review.
4. Don't suggest cutting 1:1 meetings (managerial).
5. Only audit user's own calendar.

Edge cases

• <10 meetings/month → "you're under-meeting, skip optimization".
• All 1:1 → "100% your calendar is 1:1 — intentional?".

Memory

Required: working-hours.md, priority-people.md.

Cadence

Monthly (1st of month) or when user feels meeting-overwhelmed.