Voice Creator

Why this skill

Content generation only sounds like the author when a voice file tells the agent how that author actually writes. Writing one by hand means reverse-engineering tone, argument structure, and the small habits that make a voice recognizable, which is slow and easy to get wrong. This skill collects real samples, reads them closely, and turns what it finds into a voice file the rest of the project can apply.

What it does

The skill works in two modes, and the output format differs between them:

• Personal: a voice file for one team member or contributor, written from their published posts, articles, and any writing they share.
• Company: a voice file for the brand account, built from the website, brand guidelines, and marketing copy.

In both modes it gathers source material, studies how the author opens an argument, builds it, and closes it, then captures tone, language mechanics, and the patterns the author avoids. It also pulls together on-brand and off-brand examples so the distinction is concrete rather than abstract.

What it produces:

• A voice file written to context/voices/, named voice-[firstname].md for a person or voice-company.md for the brand.
• Sections covering content scope, tone, language rules, reasoning pattern, and the patterns that should never appear in the author's writing.

For a personal voice it needs enough real writing to work from, so it asks for more samples when too few are available rather than inventing a profile. For a company voice it reads across the site and any guidelines you provide, and notes whether it is updating an existing file or creating a fresh one.

How to use it

Call /act-voice-creator. It first asks whether the voice is for a specific person or the company brand account, then asks for the source material that fits: a name, role, and links to published writing for a person, or a website and brand guidelines for the company. You can also paste text directly into the conversation.

Once it has enough to work from, it analyzes the samples and drafts the file. When it finishes it tells you where the file landed and asks you to review it, calling out the reasoning pattern and tone as the sections most worth a second look, since those are the hardest to get right from samples alone. The result feeds straight into content work, where the relevant voice file is applied to every LinkedIn post, email, blog post, and page.