Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Sweep the user's REAL mail and calendar for dropped balls — threads awaiting their reply, promises they made, and replies they're owed — then draft the nudges. Use when asked what am I forgetting, what have I not replied to, who owes me a reply, or chase my open threads in Cowork. Reads sent/received mail via the Gmail connector and recent events via Calendar, finds the open loops, and produces a follow-up-list artifact plus ready-to-send draft nudges.
.claude/skills/mohitagw15856-followup-sweep/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 96% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 60% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 222% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 39% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 42% | 0% |
Dropped balls hide in the sent folder: the reply you promised, the question you asked that never came back, the thread that stalled after you. In Claude Cowork this skill scans the real account for those open loops and drafts the nudges so nothing important dies of silence.
Ask for these if not provided:
Guardrails: drafts only, never auto-send; don't nudge someone who replied in a thread you missed — re-check the latest message; never invent a promise the user didn't make; if a connector is unauthorised, produce the list from what's available and say what's missing.
A Follow-up Sweep artifact:
N open loops · X you owe · Y owed to you · Z promises · D drafts ready
| Thread | Loop type | Age | Stakes | Suggested action | |---|---|---|---|---|
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 6,118 | 6,133 | +0% | 1 | 1 | 0% | 951 | 1,866 | +96% | 0 | 0 | — |
case-13 | pass→pass | 11,021 | 5,506 | -50% | 1 | 1 | 0% | 1,615 | 1,873 | +16% | 0 | 0 | — |
case-01 | fail→pass | 7,437 | 7,522 | +1% | 1 | 1 | 0% | 1,414 | 2,267 | +60% | 0 | 0 | — |
case-02 | fail→fail | 14,014 | 9,142 | -35% | 1 | 1 | 0% | 2,383 | 2,401 | +1% | 0 | 0 | — |
case-03 | pass→pass | 16,267 | 13,030 | -20% | 1 | 1 | 0% | 2,833 | 3,267 | +15% | 0 | 0 | — |
case-05 | fail→pass | 4,407 | 6,541 | +48% | 1 | 1 | 0% | 624 | 2,011 | +222% | 0 | 0 | — |
case-06 | pass→pass | 5,843 | 5,134 | -12% | 1 | 1 | 0% | 956 | 1,726 | +81% | 0 | 0 | — |
case-07 | pass→pass | 6,238 | 3,883 | -38% | 1 | 1 | 0% | 1,037 | 1,590 | +53% | 0 | 0 | — |
case-08 | fail→pass | 6,859 | 3,951 | -42% | 1 | 1 | 0% | 1,176 | 1,640 | +39% | 0 | 0 | — |
case-09 | pass→pass | 6,366 | 3,955 | -38% | 1 | 1 | 0% | 1,049 | 1,620 | +54% | 0 | 0 | — |
case-10 | fail→pass | 6,377 | 3,411 | -47% | 1 | 1 | 0% | 1,052 | 1,498 | +42% | 0 | 0 | — |
case-11 | fail→fail | 4,840 | 7,117 | +47% | 1 | 1 | 0% | 778 | 2,095 | +169% | 0 | 0 | — |
case-12 | pass→pass | 12,829 | 6,156 | -52% | 1 | 1 | 0% | 2,023 | 1,752 | -13% | 0 | 0 | — |
case-14 | fail→pass | 6,209 | 10,084 | +62% | 1 | 1 | 0% | 1,004 | 2,513 | +150% | 0 | 0 | — |
case-15 | fail→pass | 6,632 | 4,552 | -31% | 1 | 1 | 0% | 929 | 1,691 | +82% | 0 | 0 | — |
case-16 | fail→pass | 7,446 | 5,068 | -32% | 1 | 1 | 0% | 1,247 | 1,712 | +37% | 0 | 0 | — |
case-17 | fail→fail | 8,359 | 8,930 | +7% | 1 | 1 | 0% | 1,422 | 2,385 | +68% | 0 | 0 | — |
case-18 | pass→pass | 5,662 | 3,649 | -36% | 1 | 1 | 0% | 987 | 1,503 | +52% | 0 | 0 | — |
case-19 | pass→pass | 6,676 | 7,452 | +12% | 1 | 1 | 0% | 1,197 | 2,146 | +79% | 0 | 0 | — |
case-20 | pass→pass | 5,959 | 9,465 | +59% | 1 | 1 | 0% | 1,077 | 2,372 | +120% | 0 | 0 | — |
case-21 | pass→pass | 4,385 | 5,142 | +17% | 1 | 1 | 0% | 653 | 1,678 | +157% | 0 | 0 | — |
case-22 | pass→pass | 11,204 | 9,311 | -17% | 1 | 1 | 0% | 1,884 | 2,366 | +26% | 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. The headline lift of +36 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.