v1.3.03 loop iterationsFlagship

Prompt Sharpener

AI outputs feel 'generic' because most prompts are vague. The Prompt Sharpener takes an underperforming prompt and rewrites it into a precise, high-performing one — without just making it longer.

What this skill does

The gap between a mediocre prompt and an excellent one isn't length or cleverness — it's specificity. Vague prompts force the model to guess what you want, and it guesses toward the average of its training data. That's why AI outputs feel "generic." Specific prompts constrain the output space to exactly what you need. This skill encodes the six dimensions of specificity that consistently move output quality.

The process starts with a triage. Not every prompt needs sharpening. A factual question doesn't need a persona, format spec, or examples — the skill says so and doesn't rewrite it. An already-strong prompt gets one or two refinement notes, not a wholesale rewrite (respecting the user's work is part of the discipline). Only the actually-weak prompts get the full pass.

For those, the skill diagnoses which of six dimensions are missing — persona, format, constraints, context, examples, or "doing too much" — and rewrites with the missing pieces filled in. It's also opinionated about the "doing too much" case. A prompt that tries to research competitors, analyse their pricing, write a comparison, and draft an email about it is four prompts in a trench coat. The skill identifies these and suggests the chain decomposition, with each step's output feeding the next.

Crucially, it explains what changed and why. You don't just get a longer prompt back; you get a diff with rationale, so the next prompt you write yourself is sharper without needing the skill. The goal is to make the skill less necessary over time — the sign of a tool that actually teaches.

When this triggers

·You wrote a prompt and the output is vague, generic, or technically wrong
·You're getting different-quality results from the same prompt and don't know why
·You want to know if your prompt is actually good before sending it again
·You're handing a prompt off to a teammate and need it to work reliably
·You've never written a 'real' prompt and want to learn the framework, not just copy a template

Example

Trigger

User pastes: 'Write me marketing copy for my new product.'

Output

Diagnosis: missing 5 of the 6 sharpening dimensions. This prompt is forcing the model to guess everything — who's writing, who's reading, what format, what tone, what to emphasise. The model will guess the *average* of its training data, which is why the output feels generic. Sharpened version: "You are a senior B2B copywriter writing for a SaaS company that sells [product]. The target reader is a CTO at a 50-500-person company who's evaluating tools and skeptical of marketing claims. Write the homepage hero section: one headline (under 12 words), a one-sentence subhead, and one CTA. The headline should make a specific, falsifiable claim — not 'transform your workflow' but a concrete outcome a real customer reported. Use the voice of these example pages: [link 1, link 2]. Output three variants. After each, in 1-2 sentences, explain what angle that variant is taking and why a CTO might respond to it." What changed: · Added persona (senior B2B copywriter) · Added audience (CTO, 50-500 ppl, skeptical) · Added format (headline + subhead + CTA + 3 variants) · Added constraint (under 12 words, specific claim) · Added voice reference (two URLs) · Asked for reasoning (explains the angle for each variant)

Get this skill + 15 more

Included in the The Solopreneur Stack — run a one-person empire. Save $110+ vs buying individually.

Get The Solopreneur Stack — $129

What you get

126-line SKILL.md, ready to drop into ~/.claude/skills/
Tested through 3 Karpathy-loop iterations (versions v1.0.0 → v1.3.0)
Triggers automatically when relevant — no command to remember
Lifetime updates as the skill is refined further