thisisfatih/apply-a16z/sharpen-answers

Sharpen Answers Skill

When this activates

• User runs /sharpen
• User asks "make this better", "rewrite this", "improve my answer" on an application field
• /score surfaces a field below 3/5 and offers to sharpen it

Sharpening principles (in priority order)

1. Specificity — Replace every vague claim with a number, name, date, or named entity
2. Insight — Surface the non-obvious belief or finding. If none exists, ask the founder for it.
3. Momentum — Lead with what has already happened. Past > future for a16z.
4. Economy — Match or beat original word count. Every word must earn its place.
5. Voice — Preserve the founder's voice. Do not write like a press release.

Writing density by field

Reviewers scan hundreds of applications. Match compression to field purpose:

Density levels:

• ultra — noun phrases, abbreviations, minimum viable words. ("AI underwriting, commercial brokers — $680K ARR")
• full — drop articles, fragments OK, no filler. ("Built fraud detection at Stripe. Shipped to 40K merchants.")
• lite — full sentences, zero filler/hedging, logic intact. ("We built X because Y was broken in a specific way.")

Apply the target density when rewriting. If upgrading to full/ultra causes ambiguity, step back to lite.

Forbidden phrases (flag and remove)

• "passionate about" / "deeply passionate"
• "large market opportunity" / "massive TAM"
• "disruptive" / "revolutionary" / "game-changing"
• "world-class team"
• "best-in-class"
• "at the forefront of"
• "leverage" (as verb in business context)

Output format

ORIGINAL:
[paste original]
 
REWRITE:
[rewrite]
 
CHANGES:
- "[word/phrase changed]" → "[replacement]" — [one-word rationale: specificity / insight / momentum / economy / voice]

Iteration

Score the rewrite (1-5) immediately. If still below 4/5, offer a second pass. Name what's still missing before doing it — don't just rewrite again blindly.

RFS alignment check (startup description field only)

When sharpening the Startup Description, check docs/a16z-big-ideas.md:

• If the startup maps to an active a16z Big Idea or partner RFS, weave that alignment into the rewrite
• A direct partner RFS match is strongest — the reviewer may literally be that partner
• Only do this if the alignment is real. Don't force-fit.

Example (before): "We build AI tools for enterprise data teams." Example (after with alignment): "Enterprise data teams can't feed unstructured PDFs, videos, and contracts into AI reliably. We build the ingestion and structuring layer that makes this work — exactly the gap Jennifer Li (a16z) named as Big Idea #1 for 2026."

When to stop

If the founder's raw material doesn't support a strong answer (no real insight, no traction), surface that explicitly rather than producing a polished-sounding weak answer. Suggest /research to find the missing material.