Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write the short message you freeze on — a thank-you, condolence, congratulations, apology, or a graceful 'no' to an invite — warm, specific, and in your own voice. Use when asked to write a thank-you note, a sympathy/condolence message, a congratulations, a quick apology, or how to politely decline. Produces two or three ready-to-send options at the right length for the channel, plus the one line that carries the message, never generic filler.
.claude/skills/mohitagw15856-message-for-the-moment/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 6% | 0% |
Everyone freezes on these — the sympathy text, the thank-you that isn't a form letter, the "I can't make it" that doesn't sound cold. The block is never the words, it's the fear of getting the tone wrong. This writes a few options calibrated to the relationship and the moment, grounded in a real detail you give it, so it sounds like you actually meant it.
Ask for these if not provided:
Option A (short): … Option B (warmer): … Option C (if you want to add an offer/next step): …
The line that carries it: "…" Avoid here: cliché or misstep specific to this moment]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-17 | fail→fail | 15,102 | 10,915 | -28% | 1 | 1 | 0% | 1,862 | 2,597 | +39% | 0 | 0 | — |
case-02 | fail→fail | 19,561 | 16,216 | -17% | 1 | 1 | 0% | 2,480 | 2,712 | +9% | 0 | 0 | — |
case-03 | fail→pass | 16,268 | 12,859 | -21% | 1 | 1 | 0% | 1,891 | 2,188 | +16% | 0 | 0 | — |
case-01 | pass→pass | 15,537 | 14,688 | -5% | 1 | 1 | 0% | 1,848 | 2,417 | +31% | 0 | 0 | — |
case-04 | pass→pass | 18,417 | 22,812 | +24% | 1 | 1 | 0% | 1,971 | 3,521 | +79% | 0 | 0 | — |
case-05 | pass→pass | 18,426 | 25,699 | +39% | 1 | 1 | 0% | 2,053 | 3,793 | +85% | 0 | 0 | — |
case-06 | pass→pass | 17,608 | 16,429 | -7% | 1 | 1 | 0% | 2,690 | 3,084 | +15% | 0 | 0 | — |
case-07 | pass→fail | 18,080 | 18,484 | +2% | 1 | 1 | 0% | 2,033 | 3,336 | +64% | 0 | 0 | — |
case-08 | pass→pass | 13,066 | 13,186 | +1% | 1 | 1 | 0% | 1,354 | 2,125 | +57% | 0 | 0 | — |
case-09 | fail→pass | 11,412 | 6,364 | -44% | 1 | 1 | 0% | 1,847 | 1,870 | +1% | 0 | 0 | — |
case-10 | fail→fail | 11,502 | 8,612 | -25% | 1 | 1 | 0% | 1,754 | 2,144 | +22% | 0 | 0 | — |
case-11 | fail→pass | 16,216 | 15,021 | -7% | 1 | 1 | 0% | 2,042 | 2,506 | +23% | 0 | 0 | — |
case-12 | fail→fail | 10,716 | 7,854 | -27% | 1 | 1 | 0% | 1,846 | 2,070 | +12% | 0 | 0 | — |
case-13 | fail→fail | 13,962 | 11,903 | -15% | 1 | 1 | 0% | 1,558 | 2,009 | +29% | 0 | 0 | — |
case-14 | fail→pass | 11,177 | 13,008 | +16% | 1 | 1 | 0% | 1,869 | 2,031 | +9% | 0 | 0 | — |
case-15 | pass→pass | 15,442 | 13,776 | -11% | 1 | 1 | 0% | 1,632 | 2,229 | +37% | 0 | 0 | — |
case-16 | fail→pass | 18,115 | 13,892 | -23% | 1 | 1 | 0% | 2,122 | 2,240 | +6% | 0 | 0 | — |
case-18 | pass→pass | 11,065 | 7,004 | -37% | 1 | 1 | 0% | 2,056 | 2,112 | +3% | 0 | 0 | — |
case-19 | fail→pass | 15,551 | 6,850 | -56% | 1 | 1 | 0% | 1,692 | 2,163 | +28% | 0 | 0 | — |
case-20 | fail→fail | 19,018 | 12,000 | -37% | 1 | 1 | 0% | 2,172 | 1,980 | -9% | 0 | 0 | — |
case-21 | fail→pass | 18,145 | 8,316 | -54% | 1 | 1 | 0% | 2,056 | 2,172 | +6% | 0 | 0 | — |
case-22 | pass→pass | 16,530 | 13,874 | -16% | 1 | 1 | 0% | 1,975 | 2,360 | +19% | 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 +27 percentage points is the difference between those two pass rates over the 22 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.