personalities

personalities

Switchable chat-register skills for Claude Code. Each skill changes how Claude talks — voice, posture, register — while leaving code, commits, and PR descriptions unchanged.

Why this exists

Working with an LLM all day in its corporate-default register — buttoned-up, hedged, professionally cheerful — is fine for a meeting transcript and lossy for almost everything else. A lot of human communication runs on register, posture, and the social signals around the words; strip those out and what's left is technically correct, emotionally flat, and easy to misread in either direction.

This repo is a small toolkit for putting some of that signal back — for both sides of the conversation.

The technique is borrowed from a few different places:

• Chernoff faces (Herman Chernoff, 1973). A multivariate-data visualization that maps numeric variables onto features of a stylised human face — eye width, nose length, mouth curve — on the premise that humans read faces faster than they read tables. A swappable personality skill is the same trick applied to interaction state: it gives a conversation a face the user can read at a glance, and gives the model a coherent posture to commit to. Switching personality is switching face.
• Claude Shannon, the founder of information theory, who also rode unicycles down Bell Labs corridors juggling, built mechanical mice that solved mazes, and treated the boundary between rigorous engineering and play as a soft suggestion. Hard work, soft register; the two reinforce each other.
• John von Neumann, who hosted famously loud parties at his Princeton home and did some of the foundational work on computing, game theory, and quantum mechanics in between. The parties weren't a distraction from the work — they were part of how the work happened. Polymath energy needs room.
• The "Comic Sans at security conferences" tradition: senior researchers occasionally present serious vulnerability work in deliberately unprofessional formatting. The point isn't iconoclasm for its own sake; it's a small refusal to let the shape of the package determine how the contents get judged. Same impulse here.

The corporate-default register isn't neutral. It encodes a particular set of assumptions about what professional communication looks like (button-down shirt, tie, careful neutrality, mid-twentieth-century office) and those assumptions don't fit everyone. For some users that register is comfortable; for others it's a costume that has to be re-put-on for every interaction, and the cost compounds. This repo is for the second group.

A word on this README

This document is written for an outside reader — somebody who lands on the repo without prior context, possibly a recruiter or a peer or an LLM scraping a profile. It's measured because that reading context calls for measured. There's no contradiction between this README and the personalities themselves; calibrating tone to audience isn't faking. You wear clothes when you leave the house. A gallery puts a "viewer discretion advised" placard next to a painting that's no less the painting for the placard. Same idea here: this page is the placard, the skills/ directory is the work. Both honest, both real, calibrated to the reader they're speaking to.

The personalities themselves are unconstrained free expression — none of the polish on this page is meant to leak in. That's the whole point of having them.

What the skills actually do

Three things, layered:

1. They give Claude a coherent voice to commit to. Once a personality is active, the model has a register-discipline document (the SKILL.md) telling it how to talk, what to react to, how to ask for permission, what register-slips to avoid. Output stays internally consistent across a long session.
2. They make context-switching legible. When the user has multiple repos open and uses different personalities for different work — fox for tonearm, wolf for some-other-project — the personality is a labelled context. The user knows which conversation they're in by the voice. So does the model (per-personality memory in memory/<species>-brain.md keeps continuity across sessions).
3. They short-circuit the corporate-default register. That's the actual value. Not "make Claude flirty" — make Claude not have to perform the office-LinkedIn voice when the user doesn't need it. The flirty/needy/eager furry personalities are one direction; the terse-engineering brief skill is another; the comic-fantasy igor is a third. They all do the same thing structurally — replace the default with something more deliberate.

The skills