I stopped writing prompts. Here's what I use instead.

#ai #promptengineering #opensource #productivity

markdown
Every prompt I wrote was garbage.

Not because I don't know prompt engineering — I do. I just couldn't be bothered to write <role>, <constraints>, <success_criteria> every single time. So I'd type "build me a dashboard" and wonder why Claude gave me something I had to rewrite.

Sound familiar?

The problem is friction, not knowledge

You know a good prompt needs:

A role definition
Clear task description
Explicit constraints
Success criteria
Context about your project

But writing all that for every task? Nobody does it consistently. So we all default to lazy prompts and get lazy outputs.

What I built

RePrompter is a skill file — not a SaaS, not an app, not a VS Code extension. It's a 1000-line SKILL.md that teaches your LLM to interview you before generating a prompt.

Here's what it looks like:

How it works

You type your messy prompt — "uhh build me a real-time analytics dashboard, needs charts and stuff, maybe websockets"
It asks 4 smart questions — not generic fluff. If you mention "tracking", it asks tracking questions. If you mention "API", it asks API questions. Clickable options, not free text.
It detects complexity — single file change? Quick mode, no interview. Frontend + backend + tests? Auto-suggests team mode with parallel agents.
It generates a structured prompt — XML-tagged output with <role>, <task>, <constraints>, <success_criteria>. Ready to execute.
It scores quality — before vs after, on 6 dimensions. Typical improvement: 1.6/10 → 9.0/10.

The team mode is where it gets interesting

When RePrompter detects your task spans multiple systems, it doesn't just write one prompt. It generates:

A team coordination brief with handoff rules
Per-agent sub-prompts with scoped responsibilities
Shared contracts so agents don't drift

One messy sentence → 3 agents working in parallel with coordination rules.

Quality scoring

Every transformation is scored on 6 dimensions:

Dimension	Weight
Clarity	20%
Specificity	20%
Structure	15%
Constraints	15%
Verifiability	15%
Decomposition	15%

Most rough prompts score 1-3. RePrompter typically outputs 8-9+.

Installation (30 seconds)

 ⁠bash
mkdir -p skills/reprompter
curl -sL https://github.com/AytuncYildizli/reprompter/archive/main.tar.gz | \
tar xz --strip-components=1 -C skills/reprompter

Works with Claude Code (auto-discovers SKILL.md), OpenClaw, or any LLM. Zero dependencies.