Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Turn a document the team keeps rewriting into a template that actually helps — extract the recurring skeleton, mark what varies with real placeholder prompts, keep it lighter than the ceremony it replaces, and pilot it before decreeing it. Use when asked make a template from this doc, we write this same thing every week, standardize our status updates or briefs, or why does nobody use our templates. Produces the extracted template with prompting placeholders, the keep-it-light rules, the example
.claude/skills/mohitagw15856-template-designer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 452% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 120% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 268% | 0% |
| case-12 | ✓→✓ | = Same ✓ | 65% | 0% |
Templates fail in two familiar ways: the empty-headings ghost (## Background — helpfully blank) that guides nobody, and the forty-field bureaucratic form that takes longer than freeform writing did. A working template is extracted from documents the team already writes repeatedly, keeps only the sections that recur with purpose, and — the difference-maker — its placeholders are prompts ("One sentence: what changed since last update, and does the date still hold?") rather than labels ("Update:"). It ships with a filled example, because the example teaches what the skeleton can't.
Ask for these if not provided:
## Anything else escape hatch (its absence is why people abandon templates the first time reality doesn't fit the boxes).The skeleton: sections with prompting placeholders · the escape hatch · fill-time target stated]
A realistic complete example — the length-and-tone teacher]
Cut from the pattern: X (appeared once), Y (always boilerplate) · kept: each section's recurrence + reader-use justification]
Pilot on the next real one → friction revisions → propose-with-evidence · the owner who maintains it]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 81,506 | 40,743 | -50% | 1 | 1 | 0% | 7,640 | 5,539 | -28% | 0 | 0 | — |
case-02 | fail→fail | 35,630 | 29,379 | -18% | 1 | 1 | 0% | 5,632 | 4,313 | -23% | 0 | 0 | — |
case-03 | fail→fail | 31,432 | 31,043 | -1% | 1 | 1 | 0% | 4,201 | 4,453 | +6% | 0 | 0 | — |
case-04 | fail→fail | 11,516 | 24,688 | +114% | 1 | 1 | 0% | 1,000 | 4,361 | +336% | 0 | 0 | — |
case-05 | fail→fail | 46,070 | 22,777 | -51% | 1 | 1 | 0% | 5,977 | 4,475 | -25% | 0 | 0 | — |
case-06 | fail→pass | 10,317 | 29,857 | +189% | 1 | 1 | 0% | 789 | 4,356 | +452% | 0 | 0 | — |
case-07 | fail→pass | 18,513 | 32,555 | +76% | 1 | 1 | 0% | 2,100 | 4,610 | +120% | 0 | 0 | — |
case-08 | fail→fail | 19,312 | 22,683 | +17% | 1 | 1 | 0% | 2,268 | 4,105 | +81% | 0 | 0 | — |
case-09 | fail→fail | 19,157 | 24,872 | +30% | 1 | 1 | 0% | 2,099 | 3,597 | +71% | 0 | 0 | — |
case-10 | fail→fail | 13,799 | 28,102 | +104% | 1 | 1 | 0% | 2,263 | 4,878 | +116% | 0 | 0 | — |
case-11 | fail→fail | 12,421 | 22,881 | +84% | 1 | 1 | 0% | 1,594 | 3,961 | +148% | 0 | 0 | — |
case-12 | pass→pass | 15,553 | 17,621 | +13% | 1 | 1 | 0% | 1,378 | 2,278 | +65% | 0 | 0 | — |
case-13 | fail→pass | 22,957 | 18,413 | -20% | 1 | 1 | 0% | 2,492 | 2,735 | +10% | 0 | 0 | — |
case-14 | fail→fail | 11,349 | 39,108 | +245% | 1 | 1 | 0% | 1,197 | 5,554 | +364% | 0 | 0 | — |
case-15 | fail→fail | 21,430 | 17,616 | -18% | 1 | 1 | 0% | 2,023 | 2,437 | +20% | 0 | 0 | — |
case-16 | pass→pass | 16,436 | 14,554 | -11% | 1 | 1 | 0% | 1,894 | 2,864 | +51% | 0 | 0 | — |
case-17 | fail→fail | 32,704 | 30,030 | -8% | 1 | 1 | 0% | 3,705 | 4,375 | +18% | 0 | 0 | — |
case-18 | fail→pass | 12,555 | 22,091 | +76% | 1 | 1 | 0% | 1,013 | 3,731 | +268% | 0 | 0 | — |
case-19 | pass→pass | 13,172 | 11,325 | -14% | 1 | 1 | 0% | 1,518 | 2,239 | +47% | 0 | 0 | — |
case-20 | pass→pass | 14,629 | 14,033 | -4% | 1 | 1 | 0% | 1,135 | 2,518 | +122% | 0 | 0 | — |
case-21 | pass→pass | 15,275 | 18,309 | +20% | 1 | 1 | 0% | 1,343 | 2,896 | +116% | 0 | 0 | — |
case-22 | pass→pass | 19,064 | 20,510 | +8% | 1 | 1 | 0% | 1,262 | 3,245 | +157% | 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 +18 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.