How to use the draft-in-voice agent skill for ghostwriting

What the draft-in-voice skill makes the agent do

An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a particular kind of task. This one handles ghostwriting. The agent loads a voice profile stored at people/<slug>-voice in your brain, along with background context about that person, and then drafts two or three options that match the profile's rules.

Before it shows you any draft, the agent checks each option against the profile. The profile lists things to avoid, a target length, the register to use, signature phrasing and banned words. The skill is written so that drafts which trip a named tell or break the length band are fixed before you see them.

If no validated profile exists, the skill does not guess a voice. It stops drafting and walks you through building a profile from the person's writing samples, starting with a consent step. The skill never posts or sends anything on its own.

• Drafts tweets, replies, short posts, launch copy, recruiting blurbs and emails.
• Loads the profile from people/<slug>-voice and first-party context from the brain.
• Writes two or three options and runs a voice self-check on each.
• Refuses to draft without a validated profile and offers to build one.
• Leaves posting or sending to you, through your normal approval process.

What you need before installing

The skill depends on the gbrain knowledge base tool. It uses the gbrain get, gbrain search, gbrain timeline and gbrain backlinks commands to read pages. You also need a validated voice profile page at people/<slug>-voice. If you do not have one, the skill can help build it, but building requires a corpus of real writing samples.

The skill does not run scripts and does not use the network, according to its listing. Its only file is SKILL.md.

• The gbrain knowledge base tool, with the commands listed above available to the agent.
• A validated voice profile at people/<slug>-voice in the brain.
• Real first-party writing samples if you need to build a profile.

Install the draft-in-voice skill

Download the skill as a ZIP file from the page where you found this guide. Unzip it so that the draft-in-voice folder, containing SKILL.md, sits in one of the skills locations below. Use ~/.claude/skills/ to make the skill available to you across projects, or a project's .claude/skills/ folder to limit it to that project.

If you use the Claude apps, you can instead upload the ZIP under Settings, then Capabilities, then Skills.

• Personal install: unzip into ~/.claude/skills/.
• Project install: unzip into the project's .claude/skills/ folder.
• Claude apps: upload the ZIP under Settings → Capabilities → Skills.

How to trigger the skill with natural requests

You do not need a special command. Ask in plain language and include the person's name or slug, along with the content you want. The skill responds to these trigger phrases from its listing: draft in voice, write this as, make this sound like, ghostwrite, draft a tweet as, write a post as, in their voice, in my voice and build a voice profile.

Say who the voice belongs to and what you want. If you leave the register unclear, the agent will ask one short question: casual for chat-style replies, or statement for launch and technical posts.

• Draft a tweet as alice-example, announcing the new docs page.
• Write this as the founder of the team: a reply to the thread about pricing.
• Make this sound like our CTO: a two-line recruiting blurb for the platform role.
• Ghostwrite a short email to a partner in their voice, using the attached notes.

What a good session looks like

The agent first loads the profile and the subject's main page, timeline and backlinks. It uses these pages for facts, so claims in the draft come from the brain rather than from memory. If your draft is about a specific company or idea, the agent may also search for that topic first.

You get two or three options that vary the angle, not the voice. Each one comes with the self-check result. Review the options, edit as needed, and send the text through your own channels. The agent does not post it for you.

• Check that each factual claim traces to a page in the brain.
• Confirm the register matches the platform you are posting to.
• Compare the draft with a few real posts by the person if something feels off.

Build a voice profile when none exists

If you ask for a draft and no validated profile exists, the skill stops and offers to build one. The build starts with a consent step: confirm you are authorized to write for this person and record the scope, such as tweets and launch copy but not email. Without recorded consent, the build stops.

The corpus must be first-party writing, such as published posts, essays, emails they wrote and talks they gave. Third-party descriptions and press coverage do not count. The stated threshold is 20 or more samples spanning at least six months. A profile saved below that bar is a draft and does not unlock drafting. If you plan to write tweets, at least five samples must be short posts.

• Step 0: record consent and scope.
• Step 1: gather at least 20 first-party samples spanning 6+ months.
• Check the brain first with gbrain search, gbrain backlinks and any media/ archives.
• Ingest new samples through the normal ingest skills before building.

Safety notes and limits

The skill can only enforce consent as far as the agent follows its instructions. Record consent accurately and do not use the skill to impersonate someone who has not agreed to it. The skill also checks voice fit, not factual accuracy beyond the brain pages it reads, so it is not a substitute for legal or PR review of high-stakes announcements.

Private details stay private unless the person has said them publicly. The skill is written to drafting only, so you should still review every option before it goes anywhere.

• Do not use it to impersonate someone without their consent.
• Send high-stakes or factual announcements through legal or PR review.
• Review every draft yourself. Posting and sending stay with you.