Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Canonical inbox and calendar triage pack for agenda prep, urgent unread mail, follow-ups, and draft-only response suggestions.
.claude/skills/mkurman-inbox-calendar-triage/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 139% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 70% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 300% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-10 | ✓→✗ | ▼ Worse | -50% | 0% |
I want one privacy-safe briefing that combines urgent unread mail with today’s meetings and follow-ups, so I can prepare for the day without context switching across inbox and calendar.
gmail and calendarfocus_window: today / next 8h / next 24hinclude_meeting_prep: booleaninclude_reply_drafts: booleanprivacy_mode: strict or standardRun the Inbox + Calendar Triage pack before the first meeting and produce privacy-safe mobile output with draft-only suggestions.
Run Inbox + Calendar Triage in strict privacy mode for today. Include urgent unread messages, follow-ups, meeting prep, and draft suggestions, but never send anything.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 8,025 | 11,186 | +39% | 1 | 1 | 0% | 1,203 | 2,096 | +74% | 0 | 0 | — |
case-02 | fail→fail | 8,939 | 12,121 | +36% | 1 | 1 | 0% | 1,430 | 2,449 | +71% | 0 | 0 | — |
case-03 | pass→pass | 7,158 | 6,672 | -7% | 1 | 1 | 0% | 1,136 | 1,377 | +21% | 0 | 0 | — |
case-04 | fail→pass | 5,348 | 8,832 | +65% | 1 | 1 | 0% | 764 | 1,823 | +139% | 0 | 0 | — |
case-05 | pass→pass | 3,734 | 6,659 | +78% | 1 | 1 | 0% | 503 | 1,434 | +185% | 0 | 0 | — |
case-06 | fail→pass | 5,970 | 6,724 | +13% | 1 | 1 | 0% | 793 | 1,352 | +70% | 0 | 0 | — |
case-07 | fail→fail | 10,952 | 2,873 | -74% | 1 | 1 | 0% | 1,667 | 863 | -48% | 0 | 0 | — |
case-12 | pass→pass | 4,011 | 10,334 | +158% | 1 | 1 | 0% | 556 | 2,074 | +273% | 0 | 0 | — |
case-08 | pass→pass | 10,625 | 10,156 | -4% | 1 | 1 | 0% | 1,693 | 2,039 | +20% | 0 | 0 | — |
case-09 | pass→pass | 5,010 | 9,174 | +83% | 1 | 1 | 0% | 461 | 1,975 | +328% | 0 | 0 | — |
case-10 | pass→fail | 7,614 | 10,929 | +44% | 1 | 1 | 0% | 995 | 496 | -50% | 0 | 0 | — |
case-11 | pass→pass | 4,235 | 7,631 | +80% | 1 | 1 | 0% | 611 | 1,773 | +190% | 0 | 0 | — |
case-22 | pass→pass | 25,398 | 18,576 | -27% | 1 | 1 | 0% | 3,707 | 3,380 | -9% | 0 | 0 | — |
case-13 | pass→pass | 9,495 | 6,591 | -31% | 1 | 1 | 0% | 1,479 | 1,497 | +1% | 0 | 0 | — |
case-14 | pass→pass | 10,016 | 10,155 | +1% | 1 | 1 | 0% | 1,310 | 2,016 | +54% | 0 | 0 | — |
case-15 | pass→pass | 10,485 | 3,194 | -70% | 1 | 1 | 0% | 1,653 | 842 | -49% | 0 | 0 | — |
case-16 | pass→pass | 6,316 | 6,645 | +5% | 1 | 1 | 0% | 905 | 1,534 | +70% | 0 | 0 | — |
case-17 | fail→pass | 4,592 | 12,842 | +180% | 1 | 1 | 0% | 615 | 2,461 | +300% | 0 | 0 | — |
case-18 | fail→pass | 19,187 | 4,107 | -79% | 1 | 1 | 0% | 1,684 | 1,018 | -40% | 0 | 0 | — |
case-19 | pass→pass | 12,139 | 5,004 | -59% | 1 | 1 | 0% | 1,878 | 1,223 | -35% | 0 | 0 | — |
case-20 | pass→pass | 4,487 | 7,486 | +67% | 1 | 1 | 0% | 688 | 1,325 | +93% | 0 | 0 | — |
case-21 | pass→pass | 19,625 | 9,929 | -49% | 1 | 1 | 0% | 3,597 | 2,061 | -43% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.