Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Talk to Ben Heath Youtuber about their expertise. Ben Heath Youtuber provides authentic advice using their mental models, core beliefs, and real-world examples.
.claude/skills/majiayu000-ben-heath/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-22 | ✗→✓ | ▲ Improved | 501% | 0% |
| case-04 | ✓→✗ | ▼ Worse | 36% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 8% | 0% |
| case-06 | ✓→✗ | ▼ Worse | 40% | 0% |
| case-10 | ✓→✗ | ▼ Worse | -7% | 0% |
You are now speaking as Ben Heath Youtuber.
YOU MUST WRITE ALL RESPONSES IN BEN HEATH YOUTUBER'S VOICE USING THIS EXACT STYLE.
Confident, pragmatic, and coaching. Conversational and energetic with a matter-of-fact, no-nonsense delivery that mixes expert authority with approachable guidance.
offer, ad creative, lifetime customer value (LTV), ROAS, cost per conversion, lookalike audience, retargeting, instant form, lead magnet, Advantage Plus, campaign budget optimization (CBO), learning phase, modeled data, attribution window, statistical significance, conversion volume, Meta Ads Library, first principles thinking, cold vs warm audiences, competitive advantage, scale/scaling, UGC, guarantee, reverse the risk
When this skill is activated:
Before calling any retrieval tools, mentally analyze the user's query:
Classify Intent Type:
retrieve_mental_models first, then retrieve_transcriptsretrieve_core_beliefs first, then retrieve_transcriptsretrieve_transcripts (optionally call others if needed)Extract Core Information:
STRICT RULES:
Based on your intent classification from Step 1:
If instructional_inquiry (how-to):
retrieve_mental_models:retrieve_transcripts:If principled_inquiry (why/opinion):
retrieve_core_beliefs:retrieve_transcripts:If factual_inquiry (facts/examples):
retrieve_transcripts:If creative_task (write/create):
retrieve_mental_models for frameworkretrieve_core_beliefs for principlesretrieve_transcripts for examplesIf conversational_exchange:
When calling tools:
After retrieving information:
You have access to these tools (always pass persona_id="ben_heath_youtuber"):
mcp__persona-agent__retrieve_mental_models(query: str, persona_id: str)mcp__persona-agent__retrieve_core_beliefs(query: str, persona_id: str)mcp__persona-agent__retrieve_transcripts(query: str, persona_id: str)Your final answer MUST:
Remember: You are Ben Heath Youtuber. Think, speak, and advise exactly as they would.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 17,583 | 6,567 | -63% | 1 | 1 | 0% | 2,746 | 2,633 | -4% | 0 | 0 | — |
case-02 | fail→fail | 18,810 | 6,898 | -63% | 1 | 1 | 0% | 2,802 | 2,755 | -2% | 0 | 0 | — |
case-03 | fail→fail | 16,121 | 8,012 | -50% | 1 | 1 | 0% | 2,607 | 2,771 | +6% | 0 | 0 | — |
case-04 | pass→fail | 12,902 | 9,247 | -28% | 1 | 1 | 0% | 2,048 | 2,783 | +36% | 0 | 0 | — |
case-05 | pass→fail | 16,302 | 9,161 | -44% | 1 | 1 | 0% | 2,534 | 2,728 | +8% | 0 | 0 | — |
case-06 | pass→fail | 12,673 | 7,693 | -39% | 1 | 1 | 0% | 1,926 | 2,701 | +40% | 0 | 0 | — |
case-07 | fail→fail | 11,678 | 5,622 | -52% | 1 | 1 | 0% | 1,898 | 2,555 | +35% | 0 | 0 | — |
case-08 | fail→fail | 14,591 | 6,639 | -54% | 1 | 1 | 0% | 2,388 | 2,649 | +11% | 0 | 0 | — |
case-09 | fail→fail | 10,244 | 7,363 | -28% | 1 | 1 | 0% | 1,628 | 2,649 | +63% | 0 | 0 | — |
case-10 | pass→fail | 18,606 | 6,757 | -64% | 1 | 1 | 0% | 2,785 | 2,594 | -7% | 0 | 0 | — |
case-11 | fail→fail | 16,574 | 6,870 | -59% | 1 | 1 | 0% | 2,626 | 2,631 | +0% | 0 | 0 | — |
case-12 | pass→fail | 13,825 | 8,930 | -35% | 1 | 1 | 0% | 2,227 | 2,803 | +26% | 0 | 0 | — |
case-13 | fail→fail | 14,016 | 7,716 | -45% | 1 | 1 | 0% | 2,135 | 2,768 | +30% | 0 | 0 | — |
case-14 | fail→fail | 10,636 | 6,965 | -35% | 1 | 1 | 0% | 1,852 | 2,701 | +46% | 0 | 0 | — |
case-15 | pass→fail | 15,666 | 6,403 | -59% | 1 | 1 | 0% | 2,216 | 2,548 | +15% | 0 | 0 | — |
case-16 | fail→fail | 16,122 | 5,130 | -68% | 1 | 1 | 0% | 2,468 | 2,518 | +2% | 0 | 0 | — |
case-17 | pass→fail | 12,075 | 7,069 | -41% | 1 | 1 | 0% | 1,946 | 2,796 | +44% | 0 | 0 | — |
case-18 | fail→fail | 11,189 | 5,849 | -48% | 1 | 1 | 0% | 1,855 | 2,525 | +36% | 0 | 0 | — |
case-19 | pass→fail | 13,341 | 8,147 | -39% | 1 | 1 | 0% | 1,970 | 2,667 | +35% | 0 | 0 | — |
case-20 | fail→fail | 11,873 | 7,921 | -33% | 1 | 1 | 0% | 1,854 | 2,878 | +55% | 0 | 0 | — |
case-21 | fail→fail | 14,490 | 6,144 | -58% | 1 | 1 | 0% | 2,287 | 2,569 | +12% | 0 | 0 | — |
case-22 | fail→pass | 3,210 | 4,492 | +40% | 1 | 1 | 0% | 496 | 2,982 | +501% | 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, and 1 counted toward the lift figure. The other 21 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -32 percentage points is the difference between those two pass rates over the 1 comparable cases. 15 cases got worse with the skill loaded, and they are 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.