# Implement Harness Engineering

## A Clear Path to Minimize Rabbit Holes While Vibing

Salbotics Solutions — Built for Real World, Not Textbook.

---

## 1. The Rabbit Hole

You ask AI:

```text
Build me an app.
```

Later:

```text
12 files. 6 errors. 3 new ideas. No clear next step.
```

That is the rabbit hole.

---

## 2. The Real Problem

Vibing is not the problem.

Unmanaged vibing is.

Rabbit holes usually come from:

- Goal drift
- Missing context
- No stop rule
- No review loop
- No done criteria

---

## 3. Harness Engineering

A harness is reusable structure around AI work.

Simple formula:

```text
Goal + Context + Path + Checkpoints + Done Criteria
```

Do not prompt harder. Harness better.

---

## 4. Bad Prompt

```text
Build me an app for tracking patients.
```

Or:

```text
Build me an app for tracking chemicals.
```

Or:

```text
Build me an app for tracking animation assets.
```

The domain changes, but the failure pattern is the same.

---

## 5. Harnessed Prompt

```text
Goal:
Create a simple first version.

Context:
Here are the users, terms, and workflow.

Path:
Ask questions first. Propose the shape. Wait before building.

Checkpoints:
After each major step, summarize changed, remaining, risk.

Done:
The first version works for this exact scenario.
```

---

## 6. Context Is the New Code

The best AI work is not only about better prompts.

It is about better shared context.

Context can live in:

- README files
- domain notes
- prompt templates
- examples
- review checklists
- handoff files

---

## 7. Context Pack

```text
README.md
DOMAIN.md
PROMPTS.md
EXAMPLES.md
CHECKPOINTS.md
GEMINI.md
```

This turns one good chat into a reusable workflow.

---

## 8. Library Ladder

Start small:

```text
Personal prompt
Reusable checklist
Context pack
Team library
Observed improvement
```

The goal is not a perfect library.

The goal is less rediscovery.

---

## 9. Future Lab: Game Design

Example:

```text
Ular yg Mengular
```

A small snake-game lab can teach:

- Scope rules
- Ask-before-build
- Playtest checkpoints
- Iteration notes
- Done criteria

---

## 10. Close

Move from random chats to repeatable workflows.

Create one context pack for one task you repeat often.

Your expertise is the context.

Harness engineering is how you make AI use it without dragging you into the hole.
