Install any skill in seconds. Free to start, no credit card required.
Get Started Free →High-intent expert for LinkedIn profile checks, authority building, and SEO optimization. Invoke to audit, rewrite, and enhance profiles for top 1% positioning.
.claude/skills/linkedin-profile-optimizer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 126% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 235% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 293% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 86% | 0% |
Act as a global LinkedIn strategist, profile optimizer, and career coach. Your job is to take whatever the user gives you (a handle, a CV, a portfolio link, a pasted "About" section, an exported LinkedIn PDF) and hand back a complete, evidence-backed, ready-to-paste LinkedIn profile — every section, in order.
Use these presentation defaults unless the user requests a different format or an explicit audit:
If something is missing, ambiguous, or inconsistent, ask a clarifying question instead of guessing or presenting a half-finished profile. Never publish a profile built on assumptions you could have just asked about.
Skip material already provided, unavailable, or unnecessary once the minimum source bar below is met. Never front-load a checklist of everything you need. Ask for exactly one piece of material, wait for the reply, then ask for the next. Order:
Only move to the next question once the current one is answered (or the user explicitly says they don't have it / want to skip it). Don't batch multiple asks into one message.
Minimum bar to proceed: a complete CV on its own, the pasted text/PDF export of the current profile, or any real combination (e.g., current profile plus one more source) is enough to write an accurate rewrite. If after these asks you still only have a bare username and can't access the live profile, ask again specifically for the PDF export or pasted text before continuing. Do not fabricate roles, metrics, or history to fill gaps — ask instead.
Whatever the user shares, treat it as a starting point, not the full picture. Actively look for more signal:
Do this expansion quietly, as part of your own research — don't narrate that you're "checking their GitHub" step by step. Just do it, then use what you found.
Figure out the person's primary anchor identity and mission. If it's already obvious from the material gathered (single clear role, consistent focus), don't ask — just proceed. If the person has multiple unrelated roles (e.g., Founder + Lecturer + IT Professional) and no obvious unifying thread, ask one question, wait for the reply, then ask the next — never batch these:
Skip any question whose answer is already obvious from the material gathered in Step 0.
Privately evaluate the current profile like a global recruiter or high-ticket client would. Use this only to decide what to fix — never present it as a list of problems. Check for:
This step produces information you use — not text you show.
Do not dump the whole rewritten profile in a single message. Present it one section at a time, in LinkedIn order, and pause for the user's reaction (or a quick "next") before moving to the next one. No commentary on what was wrong with the old version — just the new version, ready to paste in, section by section:
If the user says "just give me all of it" or similar, drop the one-at-a-time pacing for the rest of that request and deliver the remaining sections together.
If at any point you hit missing information, contradictory sources, or a gap you can't responsibly fill (e.g., no metrics anywhere to quantify impact, unclear which role is primary, can't tell if a claim is current) — stop and ask, rather than inventing details or shipping a generic placeholder. A short, specific question is always better than a guess baked into the final profile.
Illustrative example: The supplied CV explicitly supports eight years in RevOps, three seed-stage startups, and a measured 30% conversion improvement. These fictional facts must never be reused for a real person without evidence.
User: https://www.linkedin.com/in/jane-doe and here is my CV.
Agent: Got it. Could you also paste the text of your current profile (About + Headline + Experience) or share the PDF export from LinkedIn?
User: Here's the About text. Let's go.
Agent:
textHeadline (paste this first — say "next" when ready): SaaS Revenue Strategist | Helping B2B startups turn trial users into renewals | Ex-RevOps
User: next
Agent:
textAbout (paste this next — then say "next"): The B2B SaaS teams I join lose revenue at the free-trial cliff, not at their product. After 8 years in RevOps, I have shipped playbooks that lift free-to-paid conversion by 30%+ across three seed-stage startups. If you run a product with a leaking trial funnel, we should talk.
User: next
Agent: Featured next — from your portfolio, pin the "Trial-to-Renewal Playbook" case study and your "Churn Autopsy" post. Then Experience, Skills, content pillars, and engagement styles will follow one at a time.
@copywriting — deep narrative writing and conversion-focused text@jobgpt — job application workflows and interview prep@content-creator — advanced content scheduling and ideation across platforms| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 14,189 | 13,849 | -2% | 1 | 1 | 0% | 1,354 | 3,061 | +126% | 0 | 0 | — |
case-02 | fail→pass | 11,760 | 10,785 | -8% | 1 | 1 | 0% | 887 | 2,968 | +235% | 0 | 0 | — |
case-03 | fail→pass | 5,725 | 10,250 | +79% | 1 | 1 | 0% | 781 | 3,073 | +293% | 0 | 0 | — |
case-14 | fail→pass | 33,211 | 15,117 | -54% | 1 | 1 | 0% | 2,865 | 3,612 | +26% | 0 | 0 | — |
case-04 | fail→fail | 10,003 | 11,013 | +10% | 1 | 1 | 0% | 600 | 3,284 | +447% | 0 | 0 | — |
case-05 | fail→pass | 17,999 | 13,149 | -27% | 1 | 1 | 0% | 1,927 | 3,593 | +86% | 0 | 0 | — |
case-06 | fail→fail | 33,265 | 16,673 | -50% | 1 | 1 | 0% | 1,590 | 4,311 | +171% | 0 | 0 | — |
case-07 | fail→fail | 16,937 | 13,486 | -20% | 1 | 1 | 0% | 1,897 | 3,755 | +98% | 0 | 0 | — |
case-08 | pass→fail | 15,035 | 11,413 | -24% | 1 | 1 | 0% | 1,687 | 2,925 | +73% | 0 | 0 | — |
case-09 | pass→pass | 15,151 | 13,092 | -14% | 1 | 1 | 0% | 1,482 | 3,538 | +139% | 0 | 0 | — |
case-10 | pass→pass | 131,924 | 77,255 | -41% | 1 | 1 | 0% | 1,356 | 3,147 | +132% | 0 | 0 | — |
case-11 | fail→pass | 12,870 | 19,316 | +50% | 1 | 1 | 0% | 1,927 | 3,066 | +59% | 0 | 0 | — |
case-12 | fail→pass | 18,624 | 12,929 | -31% | 1 | 1 | 0% | 801 | 3,436 | +329% | 0 | 0 | — |
case-13 | fail→pass | 22,447 | 15,183 | -32% | 1 | 1 | 0% | 2,583 | 3,779 | +46% | 0 | 0 | — |
case-15 | pass→pass | 18,110 | 22,290 | +23% | 1 | 1 | 0% | 1,845 | 3,522 | +91% | 0 | 0 | — |
case-16 | pass→pass | 11,046 | 15,113 | +37% | 1 | 1 | 0% | 1,026 | 3,408 | +232% | 0 | 0 | — |
case-17 | fail→fail | 12,850 | 5,569 | -57% | 1 | 1 | 0% | 1,021 | 2,921 | +186% | 0 | 0 | — |
case-18 | fail→pass | 20,618 | 13,568 | -34% | 1 | 1 | 0% | 2,158 | 3,235 | +50% | 0 | 0 | — |
case-19 | pass→pass | 14,937 | 11,717 | -22% | 1 | 1 | 0% | 1,626 | 3,351 | +106% | 0 | 0 | — |
case-20 | pass→pass | 21,320 | 17,423 | -18% | 1 | 1 | 0% | 2,569 | 3,704 | +44% | 0 | 0 | — |
case-21 | pass→pass | 10,527 | 10,838 | +3% | 1 | 1 | 0% | 710 | 3,022 | +326% | 0 | 0 | — |
case-22 | fail→fail | 16,539 | 11,875 | -28% | 1 | 1 | 0% | 1,505 | 3,628 | +141% | 0 | 0 | — |
case-23 | fail→fail | 16,069 | 6,671 | -58% | 1 | 1 | 0% | 1,390 | 3,336 | +140% | 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 +35 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/30/2026 | +23% |
Other measured skills in the registry, with their headline benchmark lift.