Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Draft clear, well-structured emails — replies, outreach, and follow-ups — in the right tone.
.claude/skills/holaboss-ai-email-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 51% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 26% | 0% |
Write the email the reader can answer in one pass. Most email fails not on grammar but on clarity of ask: the recipient can't tell what you want, by when, or why it matters to them. Fix that first; everything else is polish.
Use Email Writer to draft or rewrite a single email — a reply, a cold or warm outreach, a follow-up, an update, a request. For multi-send marketing sequences and campaigns, use Email Marketing instead; for internal announcements to a whole team, use Internal Comms.
Return a subject line and the body. Keep paragraphs short. If useful, offer one alternative tone (e.g. warmer or more direct) rather than a single fixed draft.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→pass | 10,846 | 10,178 | -6% | 1 | 1 | 0% | 1,583 | 1,997 | +26% | 0 | 0 | — |
case-01 | fail→fail | 7,909 | 15,087 | +91% | 1 | 1 | 0% | 1,354 | 2,521 | +86% | 0 | 0 | — |
case-02 | fail→fail | 7,482 | 10,456 | +40% | 1 | 1 | 0% | 1,200 | 1,895 | +58% | 0 | 0 | — |
case-03 | fail→fail | 3,920 | 5,905 | +51% | 1 | 1 | 0% | 729 | 1,320 | +81% | 0 | 0 | — |
case-04 | fail→fail | 7,316 | 8,955 | +22% | 1 | 1 | 0% | 1,134 | 1,769 | +56% | 0 | 0 | — |
case-05 | fail→pass | 51,021 | 10,541 | -79% | 1 | 1 | 0% | 1,226 | 1,857 | +51% | 0 | 0 | — |
case-07 | fail→fail | 13,162 | 11,123 | -15% | 1 | 1 | 0% | 1,816 | 2,140 | +18% | 0 | 0 | — |
case-08 | fail→fail | 4,673 | 8,329 | +78% | 1 | 1 | 0% | 769 | 1,626 | +111% | 0 | 0 | — |
case-09 | pass→pass | 3,407 | 4,068 | +19% | 1 | 1 | 0% | 496 | 1,064 | +115% | 0 | 0 | — |
case-10 | fail→pass | 8,058 | 10,943 | +36% | 1 | 1 | 0% | 1,118 | 1,849 | +65% | 0 | 0 | — |
case-11 | pass→pass | 5,747 | 6,463 | +12% | 1 | 1 | 0% | 971 | 1,479 | +52% | 0 | 0 | — |
case-12 | fail→fail | 6,878 | 8,314 | +21% | 1 | 1 | 0% | 1,174 | 1,860 | +58% | 0 | 0 | — |
case-13 | pass→pass | 11,274 | 7,497 | -34% | 1 | 1 | 0% | 1,889 | 1,556 | -18% | 0 | 0 | — |
case-14 | fail→fail | 10,397 | 9,306 | -10% | 1 | 1 | 0% | 1,477 | 1,776 | +20% | 0 | 0 | — |
case-15 | fail→pass | 9,315 | 8,607 | -8% | 1 | 1 | 0% | 1,498 | 1,566 | +5% | 0 | 0 | — |
case-16 | fail→fail | 8,082 | 8,492 | +5% | 1 | 1 | 0% | 1,299 | 1,565 | +20% | 0 | 0 | — |
case-17 | fail→pass | 7,460 | 6,058 | -19% | 1 | 1 | 0% | 1,145 | 1,447 | +26% | 0 | 0 | — |
case-18 | fail→fail | 5,575 | 6,360 | +14% | 1 | 1 | 0% | 1,024 | 1,354 | +32% | 0 | 0 | — |
case-19 | pass→pass | 12,640 | 10,456 | -17% | 1 | 1 | 0% | 1,880 | 1,743 | -7% | 0 | 0 | — |
case-20 | fail→pass | 6,666 | 8,179 | +23% | 1 | 1 | 0% | 1,093 | 1,414 | +29% | 0 | 0 | — |
case-21 | fail→fail | 7,447 | 11,311 | +52% | 1 | 1 | 0% | 1,330 | 2,146 | +61% | 0 | 0 | — |
case-22 | pass→pass | 17,489 | 11,375 | -35% | 1 | 1 | 0% | 2,873 | 2,374 | -17% | 0 | 0 | — |
case-23 | pass→pass | 13,045 | 13,922 | +7% | 1 | 1 | 0% | 2,043 | 2,553 | +25% | 0 | 0 | — |
case-24 | pass→pass | 15,727 | 10,592 | -33% | 1 | 1 | 0% | 2,566 | 2,172 | -15% | 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. 24 cases were attempted. The headline lift of +25 percentage points is the difference between those two pass rates over the 24 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.