Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create iMandalArt 2.2 pure-text 3x3 Mandala index cards with a five-character center axis, eight orthogonal surrounding angles labeled Ⓐ-Ⓗ, compact title cells such as Ⓐ健康復節, hard-line CJK layout, optional の字型 ordering, and 井字號 compass ordering. Use when the user asks for iMandalArt, iMandalart, I MandalArt, の字型, ◊字型, 井字號, 中國城市方位, 64+8+1, 手機方形曼陀羅, 正方形九宮索引卡, 五字定格, ◎中心, Ⓐ-Ⓗ 九宮卡, TheBrain/Cerebro preview-safe hard-line cards, or a square 3x3 card that is not a Markdown table, visible-border PE2 car
.claude/skills/twhsi-imandalart/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 187% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 149% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 74% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 127% | 0% |
Use this skill to turn any source material into one iMandalArt 2.2 pure-text 3x3 card.
iMandalArt 2.2 is:
◎◎◎◎◎, one five-Han-character center line, then ◎◎◎◎◎ again.iMandalArt 2.2 is not:
Default to an iMandalArt 2.2 pure-text hard-line card.
Ⓐ through Ⓗ.◎◎◎◎◎, one five-Han-character center line, then ◎◎◎◎◎.For LLM chat surfaces such as ChatGPT, Claude, Gemini, Codex, or Discord, wrap the card in a fenced text code block so hard line breaks and full-width spaces survive rendering.
For TheBrain/Cerebro direct note insertion or clipboard-only workflows, emit the raw 11-line card without a Markdown fence.
iMandalArt is the compression king in the three-skill publishing chain:
source material -> iMandalArt 2.2 eight angles -> BIRD 2.2 addresses and Routes -> A4 Booklet 2.2
The final card has exactly 11 physical text lines:
textⒶ四字標題 Ⓑ四字標題 Ⓒ四字標題 五字內容一 五字內容一 五字內容一 五字內容二 五字內容二 五字內容二 Ⓓ四字標題 ◎◎◎◎◎ Ⓔ四字標題 五字內容一 中心五字句 五字內容一 五字內容二 ◎◎◎◎◎ 五字內容二 Ⓕ四字標題 Ⓖ四字標題 Ⓗ四字標題 五字內容一 五字內容一 五字內容一 五字內容二 五字內容二 五字內容二
Line rules:
◎◎◎◎◎. between columns.Ⓐ.◎◎◎◎◎.◎◎◎◎◎ unless the user explicitly asks for a second center sentence.Surrounding title rows:
Ⓐ, Ⓑ, Ⓒ, Ⓓ, Ⓔ, Ⓕ, Ⓖ, Ⓗ in that order.Ⓐ輸入來源.Content rows:
◎◎◎◎◎ by default.Center cell:
◎◎◎◎◎ only as the displayed center title.◎◎◎◎◎ in 2.2 default output.Use this order unless the user gives a different map:
Ⓐ source / trigger: Where does this begin?Ⓑ input / material: What is being read or transformed?Ⓒ recognition / method: How is the pattern identified?Ⓓ output / expression: What form does it become?Ⓔ process / thread: What sequence makes it work?Ⓕ storage / place: Where is it saved or remembered?Ⓖ constraint / avoid: What must be avoided?Ⓗ next action: What is the next concrete move?Use this layout when the user asks for の字型, ◊字型, a 64+8+1 index, or explicitly supplies the same spatial order. の字型 is the name of the ordering pattern; ◎ remains the visible center marker.
For a directory or route card, place the lettered cells exactly as follows:
textⒻ Ⓒ Ⓖ Ⓑ ◎ Ⓓ Ⓔ Ⓐ Ⓗ
For the eight keywords inside any child nine-grid, place the numbered cells exactly as follows:
text⑥ ③ ⑦ ② ◎ ④ ⑤ ① ⑧
Apply these rules:
Ⓐ or ① remains the first assigned directory or keyword even though it is displayed at bottom center.Ⓘ as the ninth, center-level directory identity in a 64+8+1 system, but display the center cell as ◎, not Ⓘ.Ⓕ for a directory card or ⑥ for a numbered child grid, not Ⓐ.ⒶⒷⒸ / Ⓓ center Ⓔ / ⒻⒼⒽ or ①②③ / ④ center ⑤ / ⑥⑦⑧.Use this layout when the user asks for 井字號, a Chinese-city compass arrangement, or explicitly describes Ⓒ在上、Ⓑ在左、Ⓐ在下、Ⓓ在右.
Treat the center as the city center and place the four main gates on the cardinal axes:
textⒻ Ⓒ Ⓖ Ⓑ ◎ Ⓓ Ⓔ Ⓐ Ⓗ
Apply these fixed directions:
Ⓒ is north / top.Ⓑ is west / left.Ⓐ is south / bottom.Ⓓ is east / right.Ⓕ, Ⓖ, Ⓔ, and Ⓗ occupy northwest, northeast, southwest, and southeast respectively.Ⓕ, not Ⓐ.Ⓐ as north/top. In 井字號 mode, Ⓒ在上 is the defining rule.Example title rows for a の字型 card:
textⒻ四字標題 Ⓒ四字標題 Ⓖ四字標題 五字內容一 五字內容一 五字內容一 五字內容二 五字內容二 五字內容二 Ⓑ四字標題 ◎◎◎◎◎ Ⓓ四字標題 五字內容一 中心五字句 五字內容一 五字內容二 ◎◎◎◎◎ 五字內容二 Ⓔ四字標題 Ⓐ四字標題 Ⓗ四字標題 五字內容一 五字內容一 五字內容一 五字內容二 五字內容二 五字內容二
When the source is a weekly plan (週計劃, 週計画, Weekly Plan, 2026Wxx, or an eight-domain week card) and the user does not request the default orthogonal map, use this fixed weekly eight-domain layout:
textⒻ內在 Ⓒ財流 Ⓖ學習 ... ... Ⓑ工作 ◎◎◎◎◎ Ⓓ家庭 ... 中心五字句 ... ... ◎◎◎◎◎ ... Ⓔ社群 Ⓐ健康 Ⓗ休閒 ... ...
Weekly labels map to the user's recurring life domains:
Ⓐ health / body / exercise: 健康第一.Ⓑ work / writing / book: 書稿推進.Ⓒ finance / money / subscriptions: 財流降噪.Ⓓ family / partner / children: 家人互動.Ⓔ community / people / relationships: 人脈路徑.Ⓕ inner life / letting go / teacher stance: 去我內在.Ⓖ learning / Japanese / indexing: 日文沉澱.Ⓗ leisure / music / joy / recovery: 開心休閒.This weekly layout overrides the default title-label order in the 2.2 card contract and validation gate. All other constraints still apply: 11 physical lines, blank lines 4 and 8, one full-width separator between columns, compact titles, five-Han-character content phrases when possible, and the center marker on lines 5 and 7.
For non-process topics, keep the spatial order but rename the titles semantically. The eight angles must still be different kinds of attention, not eight synonyms.
Orthogonality test:
Hard line breaks are part of the artifact.
Do not output:
|.+----------+.textⒶ健康復節 Ⓑ書稿收束 Ⓒ財流降噪 羽球慢回身 書章拉中心 訂閱先盤點 跑走養節奏 青蛙先寫書 現流養書路 Ⓓ家人點火 ◎◎◎◎◎ Ⓔ人脈成路 問夢不說教 二九週計劃 約談留三卡 陪走百年路 ◎◎◎◎◎ 好友接技能 Ⓕ內在放下 Ⓖ日文沉澱 Ⓗ休閒充電 少說更點火 句子先編號 合唱鬆心身 不急證明我 接詞成路徑 吉他聽太陽
Before returning a 2.2 card, check:
Ⓐ through Ⓗ and center ◎◎◎◎◎.ⒻⒸⒼ / Ⓑ center Ⓓ / ⒺⒶⒽ.ⒻⒸⒼ / Ⓑ center Ⓓ / ⒺⒶⒽ; numbered child grids must use ⑥③⑦ / ② center ④ / ⑤①⑧.ⒻⒸⒼ / Ⓑ center Ⓓ / ⒺⒶⒽ, with Ⓒ north, Ⓑ west, Ⓐ south, and Ⓓ east.◎◎◎◎◎ by default.If any check fails, rewrite the card instead of explaining the failure.
Use legacy PE2, bordered, HTML, or renderer-based output only when the user explicitly asks for:
For legacy rendering, use scripts/imandalart_card.py and read references/imandalart-style.md only when those legacy modes are relevant. Do not use the PE2 renderer for the default 2.2 pure-text card.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 30,366 | 19,612 | -35% | 1 | 1 | 0% | 6,170 | 7,573 | +23% | 0 | 0 | — |
case-02 | fail→fail | 9,117 | 14,589 | +60% | 1 | 1 | 0% | 1,899 | 7,035 | +270% | 0 | 0 | — |
case-03 | fail→pass | 13,847 | 17,266 | +25% | 1 | 1 | 0% | 2,596 | 7,453 | +187% | 0 | 0 | — |
case-04 | pass→fail | 4,392 | 13,006 | +196% | 1 | 1 | 0% | 644 | 6,421 | +897% | 0 | 0 | — |
case-05 | pass→fail | 12,074 | 14,237 | +18% | 1 | 1 | 0% | 1,655 | 6,706 | +305% | 0 | 0 | — |
case-06 | pass→fail | 11,270 | 12,338 | +9% | 1 | 1 | 0% | 1,962 | 6,345 | +223% | 0 | 0 | — |
case-07 | fail→pass | 18,567 | 13,409 | -28% | 1 | 1 | 0% | 3,182 | 6,621 | +108% | 0 | 0 | — |
case-08 | pass→pass | 6,769 | 13,991 | +107% | 1 | 1 | 0% | 1,171 | 6,951 | +494% | 0 | 0 | — |
case-09 | fail→pass | 14,454 | 12,708 | -12% | 1 | 1 | 0% | 2,600 | 6,466 | +149% | 0 | 0 | — |
case-10 | fail→fail | 19,068 | 16,029 | -16% | 1 | 1 | 0% | 3,570 | 7,003 | +96% | 0 | 0 | — |
case-11 | fail→pass | 19,534 | 13,415 | -31% | 1 | 1 | 0% | 3,728 | 6,480 | +74% | 0 | 0 | — |
case-12 | fail→pass | 18,701 | 16,877 | -10% | 1 | 1 | 0% | 3,100 | 7,044 | +127% | 0 | 0 | — |
case-13 | fail→pass | 9,611 | 10,612 | +10% | 1 | 1 | 0% | 1,496 | 5,865 | +292% | 0 | 0 | — |
case-14 | fail→pass | 15,869 | 18,744 | +18% | 1 | 1 | 0% | 2,627 | 7,574 | +188% | 0 | 0 | — |
case-15 | fail→fail | 8,873 | 15,303 | +72% | 1 | 1 | 0% | 1,530 | 6,977 | +356% | 0 | 0 | — |
case-16 | fail→pass | 13,120 | 12,798 | -2% | 1 | 1 | 0% | 1,922 | 6,330 | +229% | 0 | 0 | — |
case-17 | fail→pass | 10,723 | 13,849 | +29% | 1 | 1 | 0% | 1,789 | 6,565 | +267% | 0 | 0 | — |
case-18 | fail→pass | 9,299 | 13,894 | +49% | 1 | 1 | 0% | 1,576 | 6,489 | +312% | 0 | 0 | — |
case-19 | fail→pass | 15,719 | 14,922 | -5% | 1 | 1 | 0% | 2,447 | 6,849 | +180% | 0 | 0 | — |
case-20 | fail→fail | 9,832 | 14,683 | +49% | 1 | 1 | 0% | 1,761 | 6,721 | +282% | 0 | 0 | — |
case-21 | fail→pass | 8,096 | 15,481 | +91% | 1 | 1 | 0% | 1,278 | 6,959 | +445% | 0 | 0 | — |
case-22 | fail→pass | 12,278 | 16,577 | +35% | 1 | 1 | 0% | 2,241 | 7,236 | +223% | 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 +45 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.