write-like-adam

A Claude Code skill that rewrites or drafts text in Adam's own personal writing style — not generic AI-polished prose.

Built by mining ~180 sampled ChatGPT conversation exports (2023–2026, spanning personal chats and business/project work) for how he actually types: his prompts and messages, never the AI's replies. The two are easy to conflate in raw exports because the AI's polished paragraphs sit right next to his fragments — this skill is built specifically to not let that polish leak in.

What it does

1. Rewrites any text — a paragraph, an essay, a generic AI draft — into his own phrasing and mechanics.
2. Drafts new content from scratch — a message, a note, a caption — the way he'd actually type it, ready to send as-is.

Install

Drop this folder into your Claude Code personal skills directory:

~/.claude/skills/write-like-adam/
├── SKILL.md
└── references/
    └── voice-profile.md

Once it's there, it's available across all your Claude Code sessions and projects — no per-project setup needed.

Use

Just ask, in any Claude Code session:

• "write this like me"
• "rewrite this in my voice"
• "how would I actually type this"
• "this sounds too AI, make it sound like me"

Claude will pick up the skill automatically from context — see the description field in SKILL.md for the exact trigger phrasing.

How it works

SKILL.md is the entry point Claude reads when the skill triggers. references/voice-profile.md holds the actual distilled style rules:

• Two registers — clipped/transactional for business & technical asks, versus longer/run-on for personal & reflective content — and rules for telling which one a task calls for.
• Mechanical rules — punctuation habits (; to elaborate, - standing in for an em-dash, dropped articles/apostrophes), tag-questions, before→after correction shorthand, occasional Malay code-switching.
• Anti-patterns — the generic-AI tells (em-dashes, hedging throat-clearers, enthusiasm markers, transitional connective tissue) that get stripped out, since removing these matters more than adding quirks.
• Optional "raw mode" — off by default — for genuine syntactic looseness (doubled prepositions, unresolved referents) rather than just efficient compression, reserved for messages that are clearly just for himself or someone who already shares context.

Notes

• This isn't a universal "sound casual" skill — it's built from one specific person's writing sample. Reusing it for someone else's voice would mean re-mining a corpus of their writing, not editing the rules here.
• Company/project names from the source material have been genericized in the examples; nothing in this repo identifies specific businesses or contains personal identifiers.