Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Writes personalized cold emails by researching prospects, crafting attention-grabbing openers, and scoring drafts against a quality rubric. Use when the user asks to write cold emails, outreach emails, sales prospecting messages, outbound email campaigns, or B2B email sequences.
.claude/skills/gtmagents-cold-email-personalization/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-02 | ✓→✗ | ▼ Worse | -11% | 0% |
Signal found: Prospect's company posted a VP Sales hire on LinkedIn 2 weeks ago.
Subject: {{first_name}}, growing the sales team?
Body: Hi {{first_name}}, saw {{company}} just brought on a new VP Sales — congrats. Specifically, it looks like you're scaling outbound to {{customer_type}}. We helped similar company] ramp 3 new reps to quota 40% faster by templatizing their top performer's research workflow. Worth a 15-min look?
Why it works: Uses a verified, recent signal in the first line; ties it to a specific, relevant outcome; single clear CTA.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,837 | 8,870 | +13% | 1 | 1 | 0% | 1,291 | 1,640 | +27% | 0 | 0 | — |
case-02 | pass→fail | 10,847 | 4,508 | -58% | 1 | 1 | 0% | 1,742 | 1,556 | -11% | 0 | 0 | — |
case-03 | pass→pass | 9,264 | 4,613 | -50% | 1 | 1 | 0% | 1,400 | 1,558 | +11% | 0 | 0 | — |
case-04 | pass→pass | 7,084 | 3,412 | -52% | 1 | 1 | 0% | 1,109 | 1,323 | +19% | 0 | 0 | — |
case-05 | fail→pass | 8,272 | 1,939 | -77% | 1 | 1 | 0% | 1,198 | 1,114 | -7% | 0 | 0 | — |
case-06 | fail→pass | 6,814 | 2,591 | -62% | 1 | 1 | 0% | 1,200 | 1,184 | -1% | 0 | 0 | — |
case-07 | pass→pass | 6,492 | 2,361 | -64% | 1 | 1 | 0% | 977 | 1,242 | +27% | 0 | 0 | — |
case-08 | pass→pass | 9,826 | 5,833 | -41% | 1 | 1 | 0% | 1,598 | 1,743 | +9% | 0 | 0 | — |
case-09 | fail→fail | 6,280 | 2,154 | -66% | 1 | 1 | 0% | 1,014 | 1,135 | +12% | 0 | 0 | — |
case-23 | fail→fail | 11,315 | 14,121 | +25% | 1 | 1 | 0% | 1,779 | 3,005 | +69% | 0 | 0 | — |
case-10 | pass→pass | 10,570 | 4,801 | -55% | 1 | 1 | 0% | 1,680 | 1,660 | -1% | 0 | 0 | — |
case-11 | pass→pass | 6,849 | 2,883 | -58% | 1 | 1 | 0% | 1,018 | 1,296 | +27% | 0 | 0 | — |
case-12 | pass→pass | 11,302 | 9,020 | -20% | 1 | 1 | 0% | 1,907 | 2,217 | +16% | 0 | 0 | — |
case-13 | fail→fail | 9,006 | 6,206 | -31% | 1 | 1 | 0% | 1,544 | 1,835 | +19% | 0 | 0 | — |
case-14 | fail→pass | 8,590 | 2,231 | -74% | 1 | 1 | 0% | 1,353 | 1,232 | -9% | 0 | 0 | — |
case-15 | fail→pass | 11,315 | 2,329 | -79% | 1 | 1 | 0% | 1,850 | 1,162 | -37% | 0 | 0 | — |
case-16 | pass→pass | 8,751 | 5,274 | -40% | 1 | 1 | 0% | 1,429 | 1,626 | +14% | 0 | 0 | — |
case-17 | pass→pass | 10,391 | 8,099 | -22% | 1 | 1 | 0% | 1,856 | 2,273 | +22% | 0 | 0 | — |
case-18 | pass→pass | 8,749 | 4,693 | -46% | 1 | 1 | 0% | 1,352 | 1,584 | +17% | 0 | 0 | — |
case-19 | pass→pass | 6,135 | 3,762 | -39% | 1 | 1 | 0% | 996 | 1,419 | +42% | 0 | 0 | — |
case-20 | pass→pass | 13,145 | 8,649 | -34% | 1 | 1 | 0% | 2,094 | 2,269 | +8% | 0 | 0 | — |
case-21 | fail→fail | 16,824 | 14,584 | -13% | 1 | 1 | 0% | 2,721 | 3,436 | +26% | 0 | 0 | — |
case-22 | fail→fail | 17,334 | 18,727 | +8% | 1 | 1 | 0% | 2,879 | 3,744 | +30% | 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. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 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.