Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Restructure the user's REAL Google Doc — open it, tighten and reorganise it, and return a clean version — not advice on how to edit it. Use when asked to clean up this doc, restructure my draft in Drive, make this readable, or tighten the doc for review in Cowork. Reads the document via the Google Drive/Docs connector, applies a structure-and-concision pass (BLUF, one idea per section, cut the filler), and produces a restructured-document artifact plus a change summary — as a new copy, never ove
.claude/skills/mohitagw15856-doc-restructure-live/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 297% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 38% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 75% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 325% | 0% |
A rambling doc buries its point. In Claude Cowork this skill opens the real document, restructures it around the reader's decision, tightens the prose, and hands back a clean copy with a summary of what changed — so the author sees the edit, not a lecture about editing.
Ask for these if not provided:
[FILL IN] rather than inventing missing facts.[FILL IN] gaps for the author.Guardrails: never overwrite the source — always a new copy; preserve facts and citations exactly; mark invented-needs as gaps, don't fill them; if the connector is unauthorised, produce the restructured text inline and say the copy couldn't be written.
> …
| Change | From → To | Why | |---|---|---|
[FILL IN] — what's missing and where[FILL IN], not fabricated| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,580 | 23,188 | +142% | 1 | 1 | 0% | 1,587 | 3,411 | +115% | 0 | 0 | — |
case-02 | fail→fail | 4,799 | 11,311 | +136% | 1 | 1 | 0% | 801 | 2,583 | +222% | 0 | 0 | — |
case-03 | fail→fail | 13,237 | 18,999 | +44% | 1 | 1 | 0% | 1,987 | 4,152 | +109% | 0 | 0 | — |
case-04 | fail→fail | 7,971 | 8,343 | +5% | 1 | 1 | 0% | 469 | 2,401 | +412% | 0 | 0 | — |
case-05 | fail→pass | 11,074 | 10,681 | -4% | 1 | 1 | 0% | 2,018 | 2,491 | +23% | 0 | 0 | — |
case-06 | fail→fail | 6,233 | 6,455 | +4% | 1 | 1 | 0% | 1,037 | 2,032 | +96% | 0 | 0 | — |
case-07 | pass→pass | 15,556 | 4,346 | -72% | 1 | 1 | 0% | 3,031 | 1,598 | -47% | 0 | 0 | — |
case-08 | fail→pass | 2,632 | 4,398 | +67% | 1 | 1 | 0% | 404 | 1,602 | +297% | 0 | 0 | — |
case-09 | fail→pass | 5,897 | 3,154 | -47% | 1 | 1 | 0% | 995 | 1,373 | +38% | 0 | 0 | — |
case-10 | pass→pass | 8,325 | 12,565 | +51% | 1 | 1 | 0% | 1,320 | 3,055 | +131% | 0 | 0 | — |
case-11 | pass→pass | 12,663 | 15,822 | +25% | 1 | 1 | 0% | 1,904 | 3,558 | +87% | 0 | 0 | — |
case-12 | fail→fail | 7,402 | 11,665 | +58% | 1 | 1 | 0% | 1,263 | 2,630 | +108% | 0 | 0 | — |
case-13 | fail→pass | 10,350 | 11,489 | +11% | 1 | 1 | 0% | 1,605 | 2,814 | +75% | 0 | 0 | — |
case-14 | fail→fail | 13,158 | 6,129 | -53% | 1 | 1 | 0% | 2,058 | 1,857 | -10% | 0 | 0 | — |
case-15 | pass→fail | 14,085 | 4,884 | -65% | 1 | 1 | 0% | 2,307 | 1,662 | -28% | 0 | 0 | — |
case-16 | fail→fail | 3,594 | 7,782 | +117% | 1 | 1 | 0% | 622 | 2,094 | +237% | 0 | 0 | — |
case-17 | fail→pass | 5,996 | 18,613 | +210% | 1 | 1 | 0% | 961 | 4,087 | +325% | 0 | 0 | — |
case-18 | pass→fail | 14,761 | 6,415 | -57% | 1 | 1 | 0% | 2,211 | 1,903 | -14% | 0 | 0 | — |
case-19 | pass→pass | 13,688 | 20,593 | +50% | 1 | 1 | 0% | 2,225 | 4,296 | +93% | 0 | 0 | — |
case-20 | fail→fail | 1,808 | 5,530 | +206% | 1 | 1 | 0% | 305 | 1,591 | +422% | 0 | 0 | — |
case-21 | pass→pass | 15,910 | 11,787 | -26% | 1 | 1 | 0% | 2,622 | 2,707 | +3% | 0 | 0 | — |
case-22 | pass→fail | 24,890 | 5,384 | -78% | 1 | 1 | 0% | 3,781 | 1,678 | -56% | 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 +9 percentage points is the difference between those two pass rates over the 21 comparable cases. 3 cases got worse with the skill loaded, and they are 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.