Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write cold outreach and networking messages that actually get replies. Use when asked to write a cold message to a recruiter/hiring manager, a LinkedIn connection note, a referral request, or a networking/coffee-chat ask during a job search. Produces short, specific, reply-worthy messages — tuned to the recipient and the ask — with a clear subject and a low-friction call to action.
.claude/skills/mohitagw15856-outreach-message/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 32% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 120% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 41% | 0% |
Cold outreach fails when it's long, generic, and all about the sender. The ones that get replies are short, specific to the recipient, and ask for one easy thing. This skill writes that — a message tuned to who you're contacting and what you want (a referral, a chat, a recruiter intro), with a hook that proves you didn't blast it to 200 people.
Ask for these only if they aren't already provided:
Produce the message(s) tuned to the channel:
Offer 2 variants when tone is unclear (warmer vs. more direct), and a note on what makes it work.
Cold-outreach / networking practice — specificity, brevity, a single low-friction ask, and graceful follow-up.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 10,382 | 8,983 | -13% | 1 | 1 | 0% | 2,057 | 2,534 | +23% | 0 | 0 | — |
case-02 | fail→pass | 9,767 | 9,477 | -3% | 1 | 1 | 0% | 1,794 | 2,658 | +48% | 0 | 0 | — |
case-03 | pass→pass | 6,960 | 8,617 | +24% | 1 | 1 | 0% | 1,169 | 2,364 | +102% | 0 | 0 | — |
case-04 | pass→pass | 6,671 | 6,943 | +4% | 1 | 1 | 0% | 1,278 | 2,207 | +73% | 0 | 0 | — |
case-05 | fail→pass | 11,651 | 11,440 | -2% | 1 | 1 | 0% | 2,083 | 2,750 | +32% | 0 | 0 | — |
case-06 | fail→pass | 6,568 | 9,729 | +48% | 1 | 1 | 0% | 1,163 | 2,556 | +120% | 0 | 0 | — |
case-07 | pass→pass | 8,137 | 8,120 | -0% | 1 | 1 | 0% | 1,544 | 2,176 | +41% | 0 | 0 | — |
case-08 | fail→pass | 12,175 | 11,700 | -4% | 1 | 1 | 0% | 2,177 | 2,788 | +28% | 0 | 0 | — |
case-09 | fail→pass | 10,787 | 10,678 | -1% | 1 | 1 | 0% | 1,801 | 2,541 | +41% | 0 | 0 | — |
case-10 | pass→pass | 9,787 | 8,351 | -15% | 1 | 1 | 0% | 1,454 | 2,198 | +51% | 0 | 0 | — |
case-11 | pass→pass | 11,222 | 12,844 | +14% | 1 | 1 | 0% | 1,858 | 2,765 | +49% | 0 | 0 | — |
case-12 | pass→pass | 8,376 | 9,267 | +11% | 1 | 1 | 0% | 1,443 | 2,353 | +63% | 0 | 0 | — |
case-13 | fail→pass | 9,418 | 7,425 | -21% | 1 | 1 | 0% | 1,647 | 2,054 | +25% | 0 | 0 | — |
case-14 | fail→pass | 9,310 | 8,173 | -12% | 1 | 1 | 0% | 1,535 | 2,113 | +38% | 0 | 0 | — |
case-15 | fail→pass | 11,798 | 9,496 | -20% | 1 | 1 | 0% | 1,921 | 2,359 | +23% | 0 | 0 | — |
case-16 | fail→pass | 8,664 | 9,150 | +6% | 1 | 1 | 0% | 1,492 | 2,276 | +53% | 0 | 0 | — |
case-17 | fail→pass | 5,699 | 8,529 | +50% | 1 | 1 | 0% | 948 | 2,229 | +135% | 0 | 0 | — |
case-18 | fail→pass | 10,197 | 9,114 | -11% | 1 | 1 | 0% | 1,631 | 2,205 | +35% | 0 | 0 | — |
case-19 | fail→pass | 7,223 | 8,652 | +20% | 1 | 1 | 0% | 1,130 | 2,102 | +86% | 0 | 0 | — |
case-20 | pass→pass | 22,276 | 19,748 | -11% | 1 | 1 | 0% | 3,669 | 3,751 | +2% | 0 | 0 | — |
case-21 | pass→pass | 14,491 | 17,291 | +19% | 1 | 1 | 0% | 2,215 | 3,414 | +54% | 0 | 0 | — |
case-22 | pass→pass | 10,832 | 11,563 | +7% | 1 | 1 | 0% | 1,815 | 2,513 | +38% | 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 +55 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.