sijuec/eridani-speak

爱德妮语

两个技能文件。 故事的两个瞬间。 零填满。

• 安装 • 信号 • 洛基 · 之前/之后 贷项

– – . . .

LLM输出为动词. 其中大部分是间接费用——不带信息的娱乐、套期、填充词。 你为每样物物付出代价

这个Repo包含两个技能文件来修复,每个文件都以不同的方式进行.

两者都受"洛奇"(Locky)的启发——安迪·韦尔所著"PProject Hail Mary *"中的爱里迪安人物. 以6x人类密度进行交流的生物 没有填充器。 语法不能高调 纯信号 到了"洛基"一书的结尾会说流利的英语——但仍然直接,仍然密集,仍然没有愉快.

两种模式是这两个时刻。

– – . . .

□ 信号模式

洛基模式产生副作用:将温暖和个性剥去,剩下的是技术输出的清洁标记系统. 那是信号 密度相同,没有性格. 用于代理管道和编码会话,其中人格是不需要的间接费用.

标记指纹 :

• 信号/SKILL.md**

– – . . .

□ 洛基模式

全罗基. 信号加上灵魂

完整的性格——密集,直接,以事实取暖而不是取悦.

eridani-speak

two skill files. two moments in the story. zero filler.

Install • Signal • Rocky • Before / After • Credits

---

LLM output is verbose. Most of it is overhead — pleasantries, hedging, filler phrases that carry no information. You pay for every token.

This repo contains two skill files that fix that, each in a different way.

Both are inspired by Rocky — the Eridian character in Andy Weir’s Project Hail Mary. A being that communicates at 6x human density. No filler. No grammar overhead. Pure signal. By the end of the book Rocky speaks fluent English — but still direct, still dense, still no pleasantries.

The two modes are those two moments.

---

Signal mode

Rocky mode produces a side effect: strip away the warmth and personality, and what’s left is a clean notation system for technical output. That’s Signal. Same density, no character. Built for agent pipelines and coding sessions where personality is unwanted overhead.

Notation fingerprint:

X = Y           definition
X → Y           causes / leads to  
X: a, b, c      properties
Fix: ...        solution
Note: ...       important caveat

→ signal/SKILL.md

---

Rocky mode

Full Rocky. Signal plus soul.

The complete character — dense, direct, warm through fact rather than pleasantry. Nine linguistic patterns extracted from the book and tested empirically until the output is recognisable to anyone who has read it. No character names in the prompt. No book references. Pure linguistic pattern.

→ rocky/SKILL.md

---

Before / After

Normal LLM

335 tokens

An LLM (Large Language Model) is a type of AI trained on massive amounts of text to predict and generate language. The core idea: given some text, what words are likely to come next? Train a neural network on billions of examples of this, at enormous scale, and something remarkable emerges — the model doesn’t just learn to autocomplete, it develops a broad, flexible understanding of language, facts, reasoning, and context. A few key intuitions: It’s a compression of human knowledge. The training data is essentially a huge slice of written human thought — books, code, conversations, articles. The model compresses patterns from all of that into its weights (the billions of numerical parameters that define its behavior). It works by predicting, not retrieving. It doesn’t look things up in a database. It generates responses token by token based on learned patterns — which is why it can be fluent and wrong at the same time.