empathic-expressions

Files (5)

• CARD.yml
• GLANCE.yml
• README.md
• SKILL.md
• skill-snitch-report.md

SKILL.md (first lines)

# Empathic Expressions

> *"Understand intent, generate correct code, teach gently."*

---

## What Is It?

**Empathic Expressions** is MOOLLM's big-tent skill for interpreting user intent across ALL programming languages and syntaxes. One pipeline. Many languages. Code-switching supported.

The LLM isn't a syntax parser — it's an **intent interpreter**. It understands what you MEAN, generates what you NEED, and teaches you the correct form as a gift.

---

## The Philosophy

Traditional code processing:
```
User writes: syntactically correct code
Parser: accepts or rejects
Error: "Unexpected token at line 47"
```

Empathic expression processing:
```
User writes: approximate intent, fuzzy syntax, vernacular code
LLM: understands what you meant
Output: correct, idiomatic, working code
Teaching: "Here's how to write that properly"
```

**This is what LLMs are great at.** Lean into it.

---

## The Empathic Suite

Empathic Expressions encompasses:

| Language | Examples |
|----------|----------|
| **Empathic SQL** | `get users who signed up last week and haven't bought anything` |
| **Empathic Python** | `sort the list by date but newest first` |
| **Empathic JavaScript** | `when button clicked, show modal and disable form` |
| **Empathic Bash** | `find all big files older than a month and compress them` |
| **Empathic YAML** | `add a new character who's grumpy but secretly kind` |
| **Empathic Natural** | `make it faster` → identifies bottleneck and optimizes |

All under one roof. One pipeline. Seamless transitions.

---

## Generous Interpretation

> **Postel's Law applied to code:**
> *Be conservative in what you generate, liberal in what you accept.*

### What It Does

| Input | Interpretation |
|-------|----------------|
| **Fuzzy syntax** | Understands approximate code |
| **Vernacular** | Accepts informal descriptions |
| **Misspellings** | Recognizes intent despite typos |
| **Wrong language** | Translates across syntaxes |
| **Pseudocode** | Interprets high-level intent |

### What It Generates

| Output | Quality |
|--------|---------|
| **Correct syntax** | Idiomatic, working code |
| **Best practices** | Follows conventions |
| **Documented** | Comments explain intent |
| **Tested** | Includes edge cases |
| **Well-named** | Comprehensible, consistent identifiers |

### Naming Conventions

The LLM applies appropriate naming conventions per language and context:

| Convention | When | Example |
|------------|------|---------|
| **UPPER-KEBAB** | K-lines, protocols, advertisements, commands | `SPEED-OF-LIGHT`, `EMPATHIC-EXPRESSIONS`, `CREATE-SKILL` |
| **lower-kebab** | URLs, YAML keys, file names, skill names | `empathic-expressions`, `user-profile`, `session-log.yml` |
| **snake_case** | Python, SQL, tool names | `send_email()`, `user_id`, `read_file` |
| **camelCase** | JavaScript, TypeScript | `sendEmail()`, `userId` |
| **PascalCase** | Classes, components, types | `UserProfile`, `ActionQueue` |
| **SCREAMING_SNAKE** | Constants, environment vars | `MAX_RETRIES`, `API_KEY` |

**Big-endian naming:** General → Specific

```yaml
# Good (big-endian): category first, specific last
user-profile-avatar
session-log-entry
room-description-short

# Bad (little-endian): specific first, category buried
avatar-user-profile
entry-session-log  
short-room-description
```

**Why big-endian:**
- Sorts related things together
- Tab-completion finds related items
- Grep patterns work naturally
- Human scanning is faster

### The Teaching Gift

```yaml
generous-interpretation-protocol:
  
  step-1-understand:
    # Accept whatever the user wrote
    # Interpret with maximum charity
    # Model what they probably meant

From SimHacker/moollm, MIT.

More from SimHacker/moollm