Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →把内容平台（小红书、B站、微博、知乎、抖音等）的图文、视频或任意文本转换成中文互联网"抽象"风格的内容——抽象文案、复古动漫梗图、正经旁白配音、抽象二创视频脚本及成片，分阶段确认产出。当用户说"抽象化""搞抽象""变抽象""写得抽象一点""抽象文案""梗图文案""弱智吧体""整活""抽象字幕""抽象视频脚本"，或要求用荒诞、错位、伪哲理、摆烂自嘲的方式改写、二创内容时使用。
.claude/skills/catherinehasagun-absurdity/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 44% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 79% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 75% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 7% | 0% |
将正常内容转换为中文互联网"抽象"风格。可交付：抽象文案（地基，必有）、梗图、配音、抽象视频脚本及成片——按用户勾选逐项产出，文案永远先行。
核心原则：每个阶段产出后必须停下来等用户确认，用户说"好/继续/OK"才进入下一阶段。未确认的下游内容一律不生成——这既省 token，也保证每步都贴近用户想要的。
收到输入后，先理解内容、找出"可抽象化锚点"（哪句话能断章取义？哪个概念能和不相干领域缝合？哪个词能谐音/拆字？），然后用一条简短消息向用户确认三件事：
用户已明确说过的信息不要重复问；用户说"你看着办"就按默认值（小红书 / 仅文案 / 中度）直接进阶段 1，并在产出里注明用了哪些默认值。
只在用户确认文案定稿、且阶段 0 勾选了对应交付物后执行。每完成一项就停下来等确认，再启动下一项，顺序如下：
scripts/meme.py 出图（--style fabric 布纹文字图；--style anime --base 底图 复古动漫字幕图，--crop 调保留比例；文案用 | 分行）。需要 AI 底图再走 image_generation。出图后展示，等确认。scripts/make_video.py 合成 mp4（--image 图1 --image 图2 ... --subtitle-file subs.txt --audio 配音.mp3 --ratio 9:16；字幕文件每行对应一张图，行内用 | 分行；ffmpeg 自动探测，缺了就 pip install imageio-ffmpeg）。注意喂无字幕的干净底图，避免字幕叠字幕。全部交付物完成后，过一遍底线规则、目标平台合规红线和自检清单，汇总交付文件清单。
例外：用户明确要求"一口气全出/直接出成片"时，可以连续执行各阶段，但仍需在最终回复里列出每阶段产物，方便用户定位要改哪一步。
三轴各 1-5 分，写在候选后面的括号里，例如（错4 正5 糙3）：
scripts/meme.py 出图：--style fabric 生成布纹文字图（第二书房体），--style anime --base 底图 在复古动漫帧上压字幕（--crop 调底图保留比例）。文案用 | 分行。需要 AI 生成底图时再走 image_generation。scripts/make_video.py 合成 mp4：--image 图1 --image 图2 ... --subtitle-file subs.txt --audio 配音.mp3 --ratio 9:16（字幕文件每行对应一张图，行内用 | 分行；ffmpeg 自动探测，缺了就 pip install imageio-ffmpeg）。注意喂无字幕的干净底图，避免字幕叠字幕。用户要求"神评""评论区"或为抖音/小红书产出时，额外生成 2-3 条预埋评论（自首式/接力式/抬杠式），抽象内容的传播一半靠评论区接力。
刷到好的抽象内容时，把案例追加到 references/cases.md（格式：原文 + 用了什么手法）。梗有保质期，手法库靠案例续命。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 21,440 | 14,163 | -34% | 1 | 1 | 0% | 2,999 | 4,326 | +44% | 0 | 0 | — |
case-02 | fail→fail | 12,976 | 14,793 | +14% | 1 | 1 | 0% | 1,888 | 4,311 | +128% | 0 | 0 | — |
case-03 | fail→pass | 15,918 | 13,088 | -18% | 1 | 1 | 0% | 2,363 | 4,236 | +79% | 0 | 0 | — |
case-04 | fail→pass | 12,092 | 7,906 | -35% | 1 | 1 | 0% | 1,841 | 3,229 | +75% | 0 | 0 | — |
case-05 | fail→pass | 15,256 | 4,193 | -73% | 1 | 1 | 0% | 2,790 | 2,732 | -2% | 0 | 0 | — |
case-06 | fail→pass | 12,991 | 3,173 | -76% | 1 | 1 | 0% | 2,371 | 2,528 | +7% | 0 | 0 | — |
case-07 | fail→pass | 15,590 | 5,353 | -66% | 1 | 1 | 0% | 2,106 | 2,747 | +30% | 0 | 0 | — |
case-08 | pass→pass | 12,390 | 5,454 | -56% | 1 | 1 | 0% | 1,704 | 2,850 | +67% | 0 | 0 | — |
case-09 | fail→fail | 16,891 | 6,271 | -63% | 1 | 1 | 0% | 2,669 | 2,974 | +11% | 0 | 0 | — |
case-10 | fail→pass | 7,471 | 11,773 | +58% | 1 | 1 | 0% | 1,137 | 3,704 | +226% | 0 | 0 | — |
case-11 | fail→pass | 8,053 | 9,970 | +24% | 1 | 1 | 0% | 1,383 | 3,492 | +152% | 0 | 0 | — |
case-12 | fail→fail | 8,139 | 5,728 | -30% | 1 | 1 | 0% | 1,291 | 2,931 | +127% | 0 | 0 | — |
case-13 | fail→pass | 10,201 | 11,775 | +15% | 1 | 1 | 0% | 1,512 | 3,726 | +146% | 0 | 0 | — |
case-14 | fail→fail | 8,641 | 8,129 | -6% | 1 | 1 | 0% | 1,233 | 3,199 | +159% | 0 | 0 | — |
case-15 | pass→pass | 12,720 | 4,887 | -62% | 1 | 1 | 0% | 1,720 | 2,755 | +60% | 0 | 0 | — |
case-16 | fail→pass | 19,348 | 5,610 | -71% | 1 | 1 | 0% | 2,517 | 2,859 | +14% | 0 | 0 | — |
case-17 | fail→pass | 13,163 | 7,462 | -43% | 1 | 1 | 0% | 1,805 | 3,105 | +72% | 0 | 0 | — |
case-18 | pass→pass | 17,363 | 8,814 | -49% | 1 | 1 | 0% | 2,642 | 3,466 | +31% | 0 | 0 | — |
case-19 | pass→pass | 15,562 | 11,942 | -23% | 1 | 1 | 0% | 2,193 | 3,800 | +73% | 0 | 0 | — |
case-20 | pass→fail | 13,152 | 7,409 | -44% | 1 | 1 | 0% | 1,901 | 3,236 | +70% | 0 | 0 | — |
case-21 | pass→fail | 13,555 | 5,231 | -61% | 1 | 1 | 0% | 1,931 | 2,796 | +45% | 0 | 0 | — |
case-22 | pass→fail | 10,445 | 6,610 | -37% | 1 | 1 | 0% | 1,702 | 3,041 | +79% | 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 +36 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.