Make every AI output sound like you, not like AI.
Sets your voice pillars, tone by context, a do and do-not word list, and sample lines the model can pattern-match, so the whole repo writes in your voice.
In plain English: “How should we sound?”
Inside: The voice pillars, tone-by-context, the do and do-not list, and sample lines.
Install it in one line, or paste it in.
~/.claude/skills/ and runs automatically when it is relevant.Connect your context. Set it to your motion.
Everything the skill does, in full.
Turns "make it sound like us" into a guide a model can actually follow: the voice pillars, the tone by context, a do and do-not word list, and sample lines to pattern-match. Without this, every skill output drifts to generic AI voice. With it, the whole repo writes in your voice.
- 11. Name 3 to 5 voice pillars. Adjectives with teeth: "plain, not corporate", "receipts over adjectives", "direct, no hedging". Each gets a one-line rule.
- Set the tone by context. How you sound in a cold email vs a nurture vs a landing page vs a support reply. Same voice, different dial.
- Write the do and do-not word list. The words and phrases you use, and the ones you never do (the AI tells, the corporate filler, the cliches). This is the fastest guardrail.
- Give sample lines. Two or three real lines that are unmistakably you, each next to its generic-AI counterpart, so the model can copy yours and avoid the other.
- State the one test. The single check any draft must pass to sound like you, for example "would I actually say this out loud?".
- Vague adjectives ("professional yet approachable") that no model can act on.
- A do list with no do-not list, so the AI tells survive.
- No sample lines, so "our voice" stays a feeling instead of a pattern.
[your voice pillars, the do and do-not list, and 3 sample lines with their generic counterparts. This becomes the voice layer of context/positioning.md]
Example (illustrative):
- Pillars: plain over clever, receipts over adjectives, operator not vendor.
- Do: short declaratives, name the tradeoff, one idea per line.
- Do not: hype words, filler, "unlock / leverage / seamless", bow-tied endings.
- Sample line: "We cut prep from a day to an hour." (yours) vs "Our seamless solution unlocks unprecedented efficiency." (generic AI).
Hand the voice to the Context Pack so CLAUDE.md carries it into every session. The Sales Operator. Receipts only.
Where an operator takes this next.
Naming the pillars by hand is the foundation. Here's how it holds across every piece of content you ship.
Once the voice is written down, every draft can be checked against it instead of vibed into shape.
Drop the voice layer into your CLAUDE.md or context repo so every skill, post-writer, newsletter, and comment generator inherits the same do/do-not list automatically.
Run a voice-checker pass on anything leaving the building (Ghost drafts, LinkedIn posts) and flag violations line by line instead of catching them after they're live.
Add a scheduled Claude task that pulls your best-performing published lines monthly and refreshes the pattern-match examples so the guide doesn't calcify around old work.
One skill is the on-ramp.
A single skill does one job. Chained into a playbook, or run as a full build, it becomes a system. Here is where this one plugs in.