agentleFS
agiea-brainagiea-brain/agency-skills/skills/jared-perry/email-voice-setup/SKILL.md

email-voice-setup

Skill1 starsChanged 16 days agoMITView on GitHub

Use this skill when someone wants their AI to draft emails that sound like them, and the drafts it writes today read as AI-written or get rewritten line by line before they go out. Produces a personal email-voice skill built from the person's own sent mail and their own edits: a register map, rules backed by counts and quoted lines, three test drafts, and a routine that keeps it improving. Also use for "write emails in my voice", "make my drafts sound like me", "my AI emails sound like a robot", "build my email style guide", "teach my assistant how I write".

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 11 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈11

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask permission and set the scope.

    line 21
  2. Very specific:

    Write the skill from references/voice-skill-template.md and save it in the user's own skills folder.

    line 28
  3. Distinctive:

    Test it on real mail. Pick three emails the user owes, each in a different register.

    line 29
  4. Distinctive:

    MUST ask before reading any mailbox, and keep the sample and the finished skill in the user's own files.

    line 46
  5. Distinctive:

    MUST count only mail the user wrote themselves.

    line 47
  6. Distinctive:

    MUST back every rule with a count or a quoted line, and name the register it applies to.

    line 48
  7. Distinctive:

    MUST write into the finished skill that it creates drafts only and never sends.

    line 49
  8. Distinctive:

    MUST write into the finished skill that a reply stays in its original thread.

    line 50
  9. Distinctive:

    MUST write into the finished skill that it never inserts a link or claims an attachment it has not verified, and leaves a visible …

    line 51
  10. Distinctive:

    NEVER describe the voice from the user's self-description alone.

    line 52
  11. Distinctive:

    NEVER polish out the user's real habits.

    line 53
  12. Distinctive:

    NEVER keep other people's private details in the finished skill beyond what an example needs.

    line 54

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Applies when a person's AI-drafted emails need heavy editing before they send them. Produces a voice skill the person owns, built from evidence in their own outbox rather than adjectives about their style.

The play

  1. Ask permission and set the scope. Agree which mailbox, which date range, and that everything read stays in the user's own files. Confirm the user's address so only mail they wrote gets counted.
  2. Pull the sample: 30 to 40 substantive sent emails from the last one to two months, spread across the kinds of people they write to. What to skip and how to spread it is in references/measurement-checklist.md.
  3. Collect the rewrites: any AI draft the user edited before sending, with both versions. Diff each one with references/rewrite-diff-worksheet.md. What they cut is the strongest signal in the whole build.
  4. Count before describing. Run the measurement checklist across the sample and write every finding as a number. "Warm" is not a finding. "One to three exclamation points in 37 of 40 emails" is.
  5. Reconcile the two sources. The sent mail wins on how often something happens. A rewrite wins for the kind of email it came from. Scope every rule to the register it came from, or one client email ends up setting rules for every vendor note.
  6. Interview for what the mail cannot show. Ask about ten questions from references/interview-questions.md, three or four at a time, and play back what you heard before moving on.
  7. Build the register map: one row per kind of recipient, with median length, warmth, greeting, closer and exclamation range. Most people have five to eight. Record the real median length, not the one the user describes.
  8. Write the skill from references/voice-skill-template.md and save it in the user's own skills folder. Every pattern carries its count and one quoted line from their mail.
  9. Test it on real mail. Pick three emails the user owes, each in a different register. Draft them with the new skill loaded, have the user mark them up, and fold each fix back in. Stop when the edits are small.
  10. Set up the loop. Every future rewrite gets diffed and becomes a pattern, not just a log line. Re-sweep the sent mail once a quarter.

Branches:

  • Under 20 usable sent emails (new role, new mailbox): build from the rewrites and the interview, mark every rule provisional, and re-sweep at 30.
  • Someone else writes part of their mail (an assistant, a shared inbox): exclude it, and ask the user which threads are theirs.
  • They write in two languages: one register map per language. Habits rarely carry across.

What good looks like

  • First thing I check that most people skip: the length distribution. People describe their emails as longer and more careful than they are. A fifth are often under 50 words, so a skill that defaults to the 300-word email writes the top tenth of their mail.
  • The mediocre version: a list of adjectives ("friendly, direct, professional") written from the user's self-description, plus rules from one rewrite applied to every email. It produces drafts that sound like a nicer AI.
  • I know it is right when the user sends a test draft with only small edits, and every rule in the skill can point to a count or a line they actually wrote.
  • Red flag: a draft more polished than their real mail. Typos, run-on sentences and asides are part of the voice. Smoothing them out is how a draft starts reading as AI-written.

Rules

  • MUST ask before reading any mailbox, and keep the sample and the finished skill in the user's own files.
  • MUST count only mail the user wrote themselves.
  • MUST back every rule with a count or a quoted line, and name the register it applies to.
  • MUST write into the finished skill that it creates drafts only and never sends.
  • MUST write into the finished skill that a reply stays in its original thread.
  • MUST write into the finished skill that it never inserts a link or claims an attachment it has not verified, and leaves a visible placeholder instead.
  • NEVER describe the voice from the user's self-description alone.
  • NEVER polish out the user's real habits.
  • NEVER keep other people's private details in the finished skill beyond what an example needs. Strip names before the skill is shared with anyone.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.