Get LinkedIn Drafts That Actually Sound Like You (Claude Code Agent)

A voice is a set of counts. Which punctuation you use, how long your first line runs, how often you ask the reader anything. Most "sound like me" attempts fail because nobody ever measured those, so the model guesses and you can feel it in the draft.

This is the exact voice agent we built live in Claude Code. It counts the fingerprint from about 30 of your real posts, writes it into a one-page profile, and points a writer agent at the numbers. Every draft gets checked against your counts before you see it, including the one everyone forgets, your real question rate.

It's the agent we built live, packaged so you can install it and get your first on-voice draft in about 20 minutes.

What's in the download:

• The full project with a README and the eight build steps from the session

• count.py, a zero-install script that prints your punctuation fingerprint and opening lengths

• A voice-profile template you fill from your real posts

• writer.md, the four-line agent job file that enforces the counts

• 14 synthetic sample posts so you can run the whole thing today, before you've exported yours

Built by Will Leatherman and the team at Catalyst, who've built content systems for 150+ B2B companies. It measures your voice instead of describing it, so the drafts read like you wrote them, not like AI.