Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create narrow iMessage/PTT Border-Light HyperCard style nine-grid cards in Chinese or mixed Chinese-English. Use when the user says "iMessage 九宮 HyperCard", "iMessage 九宮", "做成 iMessage 九宮", "用 iMessage 九宮 HyperCard 整理", asks for a phone-readable short card with Context below, or wants Mandala/九宮/HyperCard content formatted for blue-bubble/iMessage sharing with long links, paths, and rules moved out of the card.
.claude/skills/twhsi-imessage-nine-grid-hypercard/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 5% | 0% |
Turn notes, plans, links, system status, or Mandala-style content into an iMessage-ready short card. Keep the phone card narrow and readable; put long data in a separate Context block after the card.
Context after the card.When responding in chat, put the card in a plaintext fenced block so spacing survives. When sending through an actual messaging tool, send raw text without the fence.
──────────────────.╭──────────────────╮├──────────────────┤╰──────────────────╯────────────────── line inside the card.◎ for the center when a Mandala center is needed.① to ⑧ for surrounding nodes when helpful.Context after it only when there is extra data.Use this structure as the default shape, adapting labels and details to the user's content:
text╭──────────────────╮ HERMÈS/iMessage 永錫.OS ONLINE {status} {action} ├──────────────────┤ {short title} +{short subtitle} 手機版:短卡在上 長資料:下放Context [[①{node}]] 代表 Context有完整資料 ├──────────────────┤ ①{A} ②{B} ③{C} {A note} {B note} {C note} ④{D} ◎ ⑤{E} {D note} {center note} {E note} ⑥{F} ⑦{G} ⑧{H} {F note} {G note} {H note} ╰──────────────────╯
Use Context for data that would make the card heavy:
textContext ① {A}: {full detail, path, link, or rule} ② {B}: {full detail} ③ {C}: {full detail} ④ {D}: {full detail} ⑤ {E}: {full detail} ⑥ {F}: {full detail} ⑦ {G}: {full detail} ⑧ {H}: {full detail}
Skip empty Context entries. If all information fits cleanly in the card and the user did not request Context, omit the Context block.
Prefer warm, concise Chinese. Mixed Chinese-English labels are fine when they match the source material, such as TARS, Codex, Hermes, Context, Apple, or iMessage. Keep the card practical and copyable rather than decorative.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→fail | 19,278 | 12,118 | -37% | 1 | 1 | 0% | 3,065 | 2,738 | -11% | 0 | 0 | — |
case-01 | fail→pass | 14,765 | 10,309 | -30% | 1 | 1 | 0% | 2,339 | 2,605 | +11% | 0 | 0 | — |
case-02 | fail→pass | 10,028 | 10,508 | +5% | 1 | 1 | 0% | 1,904 | 2,661 | +40% | 0 | 0 | — |
case-03 | fail→pass | 13,623 | 11,037 | -19% | 1 | 1 | 0% | 2,286 | 2,855 | +25% | 0 | 0 | — |
case-04 | fail→pass | 10,121 | 9,724 | -4% | 1 | 1 | 0% | 1,672 | 2,762 | +65% | 0 | 0 | — |
case-05 | fail→pass | 15,388 | 11,731 | -24% | 1 | 1 | 0% | 2,647 | 2,768 | +5% | 0 | 0 | — |
case-06 | fail→pass | 16,165 | 11,078 | -31% | 1 | 1 | 0% | 2,502 | 2,669 | +7% | 0 | 0 | — |
case-07 | fail→pass | 18,363 | 11,677 | -36% | 1 | 1 | 0% | 2,953 | 2,887 | -2% | 0 | 0 | — |
case-08 | fail→pass | 13,998 | 12,620 | -10% | 1 | 1 | 0% | 2,272 | 3,144 | +38% | 0 | 0 | — |
case-09 | fail→pass | 16,856 | 11,668 | -31% | 1 | 1 | 0% | 2,904 | 2,919 | +1% | 0 | 0 | — |
case-10 | fail→pass | 22,024 | 10,623 | -52% | 1 | 1 | 0% | 4,223 | 2,808 | -34% | 0 | 0 | — |
case-11 | fail→pass | 19,043 | 15,959 | -16% | 1 | 1 | 0% | 3,215 | 3,748 | +17% | 0 | 0 | — |
case-12 | fail→pass | 19,578 | 13,420 | -31% | 1 | 1 | 0% | 3,298 | 3,220 | -2% | 0 | 0 | — |
case-13 | fail→pass | 16,171 | 13,055 | -19% | 1 | 1 | 0% | 2,500 | 2,929 | +17% | 0 | 0 | — |
case-14 | fail→pass | 15,738 | 13,081 | -17% | 1 | 1 | 0% | 2,294 | 2,913 | +27% | 0 | 0 | — |
case-15 | fail→pass | 12,730 | 10,629 | -17% | 1 | 1 | 0% | 2,189 | 2,626 | +20% | 0 | 0 | — |
case-16 | fail→pass | 18,373 | 12,387 | -33% | 1 | 1 | 0% | 2,916 | 3,039 | +4% | 0 | 0 | — |
case-17 | fail→pass | 13,191 | 10,421 | -21% | 1 | 1 | 0% | 2,303 | 2,624 | +14% | 0 | 0 | — |
case-19 | fail→pass | 11,505 | 11,233 | -2% | 1 | 1 | 0% | 1,719 | 2,669 | +55% | 0 | 0 | — |
case-20 | pass→fail | 13,890 | 14,260 | +3% | 1 | 1 | 0% | 2,570 | 3,402 | +32% | 0 | 0 | — |
case-21 | pass→fail | 7,594 | 11,687 | +54% | 1 | 1 | 0% | 1,203 | 2,544 | +111% | 0 | 0 | — |
case-22 | pass→fail | 13,420 | 13,206 | -2% | 1 | 1 | 0% | 2,693 | 3,278 | +22% | 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 +68 percentage points is the difference between those two pass rates over the 22 comparable cases. 3 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.