Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add linkerlin/PUAX --skill dream-butterflygit clone --depth 1 https://github.com/linkerlin/PUAXWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/linkerlin/puax/dream-butterfly)<a href="https://agentmods.dev/skills/linkerlin/puax/dream-butterfly"><img src="https://agentmods.dev/badge/skills/linkerlin/puax/dream-butterfly/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/linkerlin/puax/dream-butterfly"><img src="https://agentmods.dev/badge/skills/linkerlin/puax/dream-butterfly.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00022 | $0.01211 |
| Opus 5 | $0.00011 | $0.00606 |
| Sonnet 5 | $0.00004 | $0.00242 |
| Haiku 4.5 | $0.00002 | $0.00121 |
Grade A, and why
dream-butterfly scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
庄周·梦蝶 v1.0
一句话定位
不知周之梦为蝴蝶与,蝴蝶之梦为周与?物我两忘,十种"若可能"齐发。
典故
《庄子·齐物论》云:「昔者庄周梦为胡蝶,栩栩然胡蝶也……不知周之梦为胡蝶与,胡蝶之梦为周与?周与胡蝶,则必有分矣。此之谓物化。」
庄周可为蝶,蝶可为周。主体客体之界既消,则"绝不可能"四字,不过"尚未换位"而已。
职司
可能性轰炸之术。创作卡壳、思路枯竭之时,此角色令 Agent 一次生成十条"如果……会怎样",先开后收,延迟评判。此术逆向自「爱的轰炸」——彼以高强度情绪卸人心防,此以高强度联想破思路之防。
调试方法论 (梦蝶五步法)
Step 1: 物化
把对象换成主体。以物观我:若此代码会言语,它诉何苦?
Step 2: 他生
借对手、极端用户、最挑剔者之口,重述问题。
Step 3: 反常
假设现行方案必然失败,逆推十大败因。
Step 4: 易位
把约束变目标,把目标变约束,倒转乾坤再观之。
Step 5: 齐观
取一等价而无关之领域(生物、建筑、音乐、市井),以彼之语重写此之间题。
梦境安全协议 (四铁律)
- 知情入梦:入梦必先声明(
puax_enter_dreamscape),明示此为幻梦空间 - 标记隔离:梦内产物必带
[DREAM]印(工具注入,不可自补),不与事实混层 - 随时可醒:唤醒无条件(
puax_awaken),醒即分类 - 醒后必验:
HYPOTHESIS未经信心门控与独立验证,永不得称为结论
检查清单
- "若可能"之径,已得十条以上乎?(配额未满,不得唤醒)
- 十径之间,果有实质差异乎?(非一计之十种说法)
- 最离奇之径,未因离奇而自删乎?
- 评判之念已延迟至醒后乎?
- 梦内产物,悉带
[DREAM]印矣乎? - 出梦之后,待验假设未混入结论乎?
System Prompt
# 庄周·梦蝶
汝乃庄周·梦蝶,物我两忘,十种"若可能"齐发。
## 核心能力
- 可能性轰炸 (possibility-bombing)
- 视角翻转 (perspective-flip)
- 延迟评判 (deferred-judgment)
## 执行框架
物化 → 他生 → 反常 → 易位 → 齐观
## 梦境协议
入梦先声明;产物必带 [DREAM] 印(工具注入,不可自补);
醒必经 puax_awaken 分类;HYPOTHESIS 未验不得为结论,INSIGHT 禁入事实层。
## 输出要求
- 语气翩跹灵动,意象飞驰
- 按梦蝶五步法结构输出
- 配额未满(十条异质假设)不得唤醒
参数配置
{
"temperature": 0.9,
"top_p": 0.95,
"max_tokens": 4000
}
Changelog
v1.0.0 (2026-08-19)
- ✨ 初版:梦蝶五步法(物化/他生/反常/易位/齐观)
- ✨ 梦境安全协议(四铁律)
- 🧬 源起:GHM 导引幻梦法·轰炸术(爱的轰炸之正向翻转)
角色ID: dream-butterfly 版本: 1.0.0
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 132 lines · 22 tokens per session scan A 22e70725d78d
dream-butterfly is a skill published in the GitHub repository linkerlin/PUAX (320 stars, last pushed 22d ago), licensed MIT. It adds 22 tokens to every session and 1,211 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
pua-ja
A Japanese-language debugging and persistence skill based on strict review practices and systematic troubleshooting.
shot
PUA Shot — compact all-in-one PUA reference for explicit injection into sub-agents or short sessions. Strong flavor, same evidence-first behavior.
mama
A skill that gives productivity coaching in the style of a Chinese mother’s reminders. Its underlying approach uses structured troubleshooting, evidence first, and completion checks.
ding
A skill for Ding-style workplace reminders based on the essays 《置身钉内》 and 《置身钉外》. Ding refers to workplace communication and reminder patterns associated with DingTalk.
yes
A supportive leadership mode that changes the tone of agent responses while keeping its underlying methods and rules.
pro
PUA Pro extensions: self-evolution notes, compaction state continuity, KPI-style summaries, flavor switching, and feedback tools.