⏳ This skill is pending AI review.

Scores will appear once the review pipeline completes.

v0.2.0

me

ddutchie@ddutchie⭐ 8 stars· last commit 1mo ago· 0 open issues

Writes in the user's own voice using their saved style profile. Use for anything the user will send or publish as themselves (emails, Slack, PRs, posts, docs). /me init sets it up.

Choose how to use this skill

You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.

1. Native installer

This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.

Do not guess an installer command or replace an existing version without reviewing the diff.

2. Complete package recommended

Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.

No complete ProSkills package is published for this listing yet.

3. Prompt-only

Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.

Need only the instruction file?

Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.

No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.

—/10

// RATINGS

⭐GitHub Stars
⭐ 8 on GitHubGitHub ↗

New / niche

🟢ProSkills Score
—
📍

Not yet listed on ClawHub or SkillsMP

// README

write like me

Your agent writes emails, Slack messages and PR descriptions for you. They don't sound like you.

They sound like every other AI: "I hope this finds you well", "Let's dive in", em dashes everywhere, three bullet points with bold lead-ins. You end up rewriting everything, or you send it anyway and people can tell.

me is an Agent Skill that fixes this. Run /me init once. It reads a few things you've actually written, asks about six quick questions and saves a style profile. After that, whenever your agent writes something as you, it writes in your voice.

/me init                          # 5-minute setup: samples + a short interview
/me reply to Sam: can't make thursday, suggest next week
/me learn                         # learn from how you edited the last draft
/me show                          # what the agent thinks your voice is
/me edit stop using "just"        # change the profile directly

The agent also uses the profile on its own when it drafts something you'll send, such as a commit message, a PR description or an email. You don't have to type /me every time.

See EXAMPLES.md for before-and-after samples.

How it works

  1. Samples beat adjectives. "Casual but professional" means nothing. Your real Slack messages show that you write in lowercase, never use a full stop at the end, and say "yep" instead of "yes". init builds rules from what you actually do.
  2. One file, yours. Your voice lives in ~/.me/STYLE.md, plain markdown that you can read and edit. It's never stored in the skill folder, so updates can't overwrite it.
  3. Registers. You write differently in Slack and in an email to a stranger. The profile keeps rules for each place you write.
  4. It learns from your edits. When you rewrite a draft, /me learn compares the two versions, separates voice changes from content changes, and suggests new rules. Nothing changes without your approval.
  5. No caricature. Occasional quirks like "…" or CAPS are measured per 1000 words, so a two-line reply doesn't carry your whole quirk load. Habits you have every time, like lowercase in Slack, apply everywhere.
  6. An anti-AI list. It ships with a list of common AI tells. Your profile can override any of them; if you really do love em dashes, keep them.

Where your profile lives

PathUse
.me/STYLE.md (in a project)A project voice, such as your company's docs style. Overrides the global profile section by section
~/.me/STYLE.mdYour personal voice, used everywhere
STYLE.md inside the skill folderFor claude.ai, which has no home directory (see below)

Every draft ends with a line such as voice: ~/.me/STYLE.md, so you can see which profile was used. On a teammate's machine or in CI, ~/.me/STYLE.md doesn't exist. If no profile is found, the skill says so instead of quietly falling back to the default voice. To make a voice travel with a repo, commit its .me/STYLE.md.

Install

Claude Code

/plugin marketplace add ddutchie/write-like-me
/plugin install me@write-like-me

Then run /me init.

Any agent (Cursor, Codex, Gemini CLI, Copilot, OpenCode, …)

npx skills add ddutchie/write-like-me

Manually

Copy skills/me/ into your agent's skills folder, for example ~/.claude/skills/me/.

Claude.ai

  1. Download me.zip from the latest release and upload it under Customize → Skills.
  2. Run /me init. Claude.ai can't save files, so at the end it gives you your profile as text.
  3. Save that text as STYLE.md inside the me/ folder, zip the folder again and re-upload it.

Privacy

Your profile and samples stay in plain files on your machine. The skill never sends them anywhere. .gitignore excludes .me/, but check before you commit a project-level profile to a public repo.

Contributing

The most valuable contributions are:

  • New entries for ai-tells.md, especially from languages other than English.
  • Better interview questions.
  • Before-and-after examples for EXAMPLES.md, with permission from whoever wrote them.

Keep SKILL.md under 500 lines (spec).

To release, bump version in SKILL.md, .claude-plugin/plugin.json and marketplace.json, merge to main, then push a matching tag:

git tag v0.2.0 && git push origin v0.2.0

A GitHub Action checks the tag matches all three versions, builds me.zip and publishes the release.

License

MIT

// HOW IT'S BUILT

// FILES

6 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.

  • skills/me/SKILL.md8.0 KB0eadc2a01228upstream ↗
    ---
    name: me
    description: Writes in the user's own voice using their saved style profile. Use for anything the user will send or publish as themselves (emails, Slack, PRs, posts, docs). /me init sets it up.
    license: MIT
    metadata:
      version: "0.2.0"
    ---
    
    # me
    
    Write the way the user writes: their words, rhythm and habits, not the default AI voice.
    
    The user's voice lives in a style profile, a markdown file this skill creates and keeps up to date. This file tells you how to set it up, apply it and improve it.
    
    ## When this applies
    
    Use the profile for text the user will **send or publish as themselves**, such as emails, Slack or chat messages, PR descriptions, commit messages, issue comments, social posts, blog drafts, docs and cover letters.
    
    Do **not** restyle your own replies to the user. Explanations, plans and status updates stay in your normal voice unless the user asks otherwise.
    
    ## Find the profile
    
    Check all three places:
    
    1. `.me/STYLE.md` in the current project. This is a project voice, such as a company's docs style.
    2. `~/.me/STYLE.md`, the user's global voice.
    3. `STYLE.md` next to this `SKILL.md`, for environments without a persistent home directory, such as claude.ai uploads. Use it only when neither of the others exists.
    
    When both 1 and 2 exist, layer them: the project profile overrides the global one section by section, and anything it leaves out comes from the global one. A file that is empty or has no Samples, Mechanics or Registers section counts as missing.
    
    **Always say which profile you used.** After every draft, add one line naming what you loaded:
    
    ```
    voice: ~/.me/STYLE.md
    voice: .me/STYLE.md over ~/.me/STYLE.md
    ```
    
    This line is for the user, never part of the text. Keep it out of the draft, commit messages and any file you write; if you wrote the text straight into a file or commit, put the line in your reply instead.
    
    **If no profile exists, say so loudly.** A missing profile usually means another machine, a teammate's checkout or CI. Falling back quietly produces the default voice with nothing to show for it.
    
    - If the user asked for writing directly (`/me <text>`, "write this like me"), don't draft. Say no profile was found, list the paths you checked, and offer `/me init`. Write neutral prose only if they tell you to go ahead.
    - If you're drafting on your own (a commit message, a PR description), write plain, neutral prose that follows `references/ai-tells.md`, and **start** your reply with `voice: none found (checked .me/STYLE.md, ~/.me/STYLE.md, skill folder), neutral style used`.
    
    ## Commands
    
    Read the first word of the arguments: `$ARGUMENTS`.
    
    | Argument | Do this |
    |---|---|
    | `init` | Run the setup interview (below) |
    | `learn` | Learn from the user's edits (below) |
    | `show` | Summarise the profile in 5–8 bullets, then the `voice:` line for what was loaded |
    | `edit <change>` | Apply the change to the profile, show the diff and save once the user confirms |
    | *(any other text)* | Rewrite that text in the user's voice |
    | *(empty)* | If no profile exists, run `init`. Otherwise show the profile path and these commands |
    
    ## Writing in the user's voice
    
    1. Load the profile, and read `references/ai-tells.md` once per session.
    2. Choose the **register** that fits the destination (Slack, email, PR, post and so on). If the profile has no entry for it, use the nearest one.
    3. Write the draft. The **samples** outrank the rules: when they disagree, sound like the samples.
       - **Don't turn the user into a caricature.** Tell habits apart from quirks.
         - **Habits** are what the user does nearly every time in a register, such as lowercase in Slack or no full stop at the end. Apply them always.
         - **Quirks** show up now and then: "…", "!!", CAPS for emphasis, dropped apostrophes, a pet phrase. Budget them by length, not per message. The profile's **Quirks** table gives each one's rate per 1000 words in each register. Expected uses = rate × draft words ÷ 1000. Round down. Below 1, the quirk is optional: use it once only if a spot in the draft calls for it, and leave it out when unsure.
         - Example: "…" at 20 per 1000 words gives a 300-word email about 6 and a 25-word Slack reply 0.5, so at most one and often none. A two-line reply never carries the full quirk load. Short drafts sound like the user through habits, word choice and length.
         - If the profile has no rates (profiles from before v0.2), estimate them from its Samples before drafting.
    4. Check the draft against the profile's **Never** list and against `references/ai-tells.md`, then fix every hit.
    5. Keep the facts. Change voice, not meaning. Never invent opinions, experiences or commitments the user didn't give you.
    6. Deliver the draft with no preamble like "Here's a version in your voice:", then the `voice:` line. When you had to guess at something important, add at most one more line.
    
    For length, match the register. If the user's Slack messages are 12 words long, yours should be too.
    
    ## `init`: build the profile
    
    Aim for about five minutes. Ask **one question at a time** and wait for each answer. Take the questions from `references/interview.md`.
    
    1. **Samples first.** Ask for 2–5 things the user actually wrote, such as Slack messages, emails, a PR description or a post. Real samples tell you far more than the answers do. If the user has none, go on with the interview only and mark the profile `confidence: low`.
    2. **Interview.** Ask the core questions in `references/interview.md`. Skip any question the samples already answered.
    3. **Analyse.** Pull out the concrete patterns: typical sentence length, capitalisation, punctuation (em dashes? Oxford comma? `...`?), contractions, emoji, spelling variant, how the user opens and closes, recurring phrases, how they hedge or push back, and what they never do. Report patterns you observed, not impressions. "Lowercase in Slack, sentence case in email" beats "casual tone".
       - Sort each pattern into a **habit** (in nearly every sample of that register) or a **quirk** (only some of them).
       - For each quirk, count the words in that register's samples and the quirk's uses, and record the rate per 1000 words. List a quirk only if it appears at least twice. If a register has under about 300 words of samples, mark its rates `rough`.
    4. **Draft.** Fill in `references/profile-template.md`. Quote short real excerpts under **Samples**.
    5. **Confirm.** Show the draft and ask what's wrong. Then write two short test pieces, such as a Slack reply and an email, and ask whether they sound right. Revise until the user is happy.
    6. **Save.** Ask whether this is the global voice (`~/.me/STYLE.md`) or a project voice (`.me/STYLE.md`). Create the folder if needed and write the file. If you can't write files (claude.ai), print the full profile in a code block and tell the user to save it as `STYLE.md` inside their copy of this skill folder, then re-upload it.
    
    ## `learn`: improve from edits
    
    The best signal is what the user changes in your drafts.
    
    1. Get the pair: your draft, and the version the user actually sent. Look for it in the conversation, or ask the user to paste it.
    2. Diff the two. Sort each change into **voice** (wording, rhythm, formality, punctuation, length) or **content** (facts, decisions, scope). Learn only from voice changes.
    3. Suggest at most 3 new or changed rules, each quoting the edit it came from. Leave out one-off changes; a rule needs a pattern, or a change the user says is always true.
       - If the user keeps deleting a quirk you added, suggest lowering its rate rather than dropping it.
    4. Once the user approves, update the profile and add a dated line to its **Learned from edits** log.
    
    When you notice the user heavily rewrote one of your drafts during normal work, offer `/me learn` in one line. Don't run it without asking.
    
    ## Rules for the profile file
    
    - Never change the profile without the user's approval. Show the diff first.
    - Keep it under about 300 lines. Merge rules that overlap, and trim old samples before adding new ones.
    - The profile is personal. Don't copy it into other projects, commits or shared files unless the user asks.
    
  • skills/me/references/ai-tells.md1.9 KBc94dcc999853upstream ↗
    # AI tells
    
    These patterns make writing read as machine-generated. Remove them unless the user's profile or samples show they actually write this way. The profile always wins.
    
    ## Openers and closers
    - Restating the request: "Great question!", "Certainly!", "I hope this email finds you well."
    - Announcing the answer: "Here's a breakdown", "Let's dive in", "In this post, we'll explore…"
    - Closing offers: "Let me know if you have any questions!", "Feel free to reach out", "Happy to help!"
    - Summary sign-offs: "In summary", "Overall", "Ultimately", "At the end of the day"
    
    ## Vocabulary
    - Inflated verbs: delve, leverage, utilize, harness, foster, bolster, streamline, elevate, empower, unlock, navigate (used figuratively)
    - Inflated nouns and adjectives: tapestry, landscape, realm, testament, journey, robust, seamless, pivotal, crucial, comprehensive, cutting-edge, game-changer
    - Filler intensifiers: truly, incredibly, absolutely, genuinely, deeply
    - Corporate hedging: "it's worth noting that", "it's important to remember", "arguably"
    
    ## Structure
    - Rule of three everywhere ("fast, reliable, and scalable")
    - "It's not X, it's Y" and "Not only X, but also Y" constructions
    - Every paragraph the same length, and every list exactly 3–5 bullets
    - Bold lead-ins on every bullet when the user writes plain lists
    - Headers on a message that's three paragraphs long
    - Ending on a neat one-line moral or rhetorical question
    
    ## Punctuation and formatting
    - Em dashes (—) when the user doesn't use them
    - Emoji as bullet points
    - Title Case In Headings when the user uses sentence case
    - Colons introducing every list
    
    ## Tone
    - Relentless positivity and enthusiasm the user didn't express
    - Over-apologising, or over-thanking ("Thank you so much for your patience!")
    - Hedging every claim, or no hedging at all when the user normally hedges
    - Explaining things the recipient obviously knows
    
  • skills/me/references/interview.md1.9 KBe813144641c4upstream ↗
    # `/me init` interview
    
    Ask one question at a time and keep each question short. Skip a question when the samples already answer it. Paraphrase each answer back only when it is ambiguous.
    
    ## Samples (always first)
    
    > Paste 2–5 things you've actually written, the more varied the better: a Slack message, an email, a PR description, a post. Messy is fine. Real writing tells me more than any answer will.
    
    If they paste only one kind of thing, ask once for a different kind.
    
    ## Core questions
    
    1. **Where you write.** Where do you write most: Slack or Teams, email, GitHub, social, docs? Which of those do you want help with?
    2. **Three words.** Pick three words for how you want to come across, and one word for how you *never* want to sound.
    3. **Mechanics.** Quick ones. Which spelling do you use (US, UK or another)? Do you capitalise in chat? Use emoji? Em dashes? Exclamation marks?
    4. **Openers and sign-offs.** How do you usually start and end an email, and a Slack message? ("hey", no greeting, "Cheers", nothing?)
    5. **Pet peeves.** Which words or phrases make you cringe when AI writes them for you?
    6. **Pushing back.** How do you say no, or disagree with someone senior? Give an example if you can.
    
    ## Optional (ask only if the user wants to go deeper)
    
    7. **Humour.** Dry, silly, none? Where is it OK?
    8. **Structure.** Bullets or paragraphs? Headers in docs? How long is too long?
    9. **Hedging.** When you're unsure, what do you say? ("I think", "probably", "not 100% sure but")
    10. **People.** Is there anyone whose writing you'd like to sound more like, or less like?
    11. **Languages.** Do you mix in another language, slang, or local expressions? Where?
    
    ## Closing
    
    After drafting, write two short test pieces in different registers, for example:
    
    - A Slack reply declining a meeting.
    - A short email asking a stranger for something.
    
    Ask: "Does this sound like you? What would you change?" Put their corrections straight into the profile.
    
  • skills/me/references/profile-template.md2.3 KB7ff43ddc66d8upstream ↗
    # Style profile template
    
    Copy this structure into `STYLE.md`. Replace every placeholder with a concrete pattern observed in the user's writing. Delete any section that has nothing real in it; an empty section is better than a made-up one.
    
    ```markdown
    ---
    name: <user's name or handle>
    updated: <YYYY-MM-DD>
    confidence: <low | medium | high>   # low = no samples, high = 5+ varied samples
    ---
    
    # How <name> writes
    
    ## Snapshot
    <2–3 sentences someone could read in 10 seconds and then sound roughly right.>
    
    ## Principles
    - <e.g. Lead with the ask. Context comes after, if at all.>
    - <e.g. Short sentences. Rarely more than 15 words.>
    - <e.g. Direct but warm. Says "no" plainly, then offers an alternative.>
    
    ## Mechanics
    - Spelling: <US | UK | ...>
    - Capitalisation: <e.g. lowercase in Slack, sentence case elsewhere>
    - Punctuation: <e.g. no em dashes; Oxford comma; "..." when trailing off>
    - Contractions: <always | sometimes | rarely>
    - Emoji: <none | only in Slack: 👍 🙏 | ...>
    - Exclamation marks: <e.g. max one per message, only for thanks>
    
    ## Quirks
    <Marks that show up only some of the time. Rate = uses per 1000 words of that register's samples; add "rough" when there were under ~300 words. Anything in nearly every message is a habit and belongs in Mechanics or Registers.>
    
    | Quirk | Register | Per 1000 words | Example |
    |---|---|---|---|
    | <e.g. "…" to trail off> | <e.g. Slack> | <e.g. 18> | <e.g. "my only feedback is…"> |
    | <e.g. CAPS for emphasis> | <e.g. Slack> | <e.g. 6 (rough)> | <e.g. "WAY simpler"> |
    
    ## Vocabulary
    - Uses: <recurring words and phrases, quoted exactly>
    - Hedges with: <e.g. "I think", "pretty sure">
    - Never uses: <words they'd never say>
    
    ## Registers
    | Where | Length | Opening | Closing | Notes |
    |---|---|---|---|---|
    | Slack / chat | <e.g. 1–2 lines> | <e.g. none, or "hey"> | <none> | <e.g. lowercase, no full stop at the end> |
    | Email | <e.g. 3–6 lines> | <e.g. "Hi Sam,"> | <e.g. "Thanks, G"> | |
    | PR / commit | | | | <e.g. imperative, bullets for changes> |
    | Social / blog | | | | |
    
    ## Never
    - <Things the user explicitly hates, on top of references/ai-tells.md>
    
    ## Samples
    > <Short verbatim excerpt, and where it's from (Slack, email, ...)>
    
    > <Another>
    
    ## Learned from edits
    - <YYYY-MM-DD>: <rule>, from "<before>" → "<after>"
    ```
    
  • skills/me/LICENSE1.0 KB6441e54748b0upstream ↗
    MIT License
    
    Copyright (c) 2026 Gerard Slee
    
    Permission is hereby granted, free of charge, to any person obtaining a copy
    of this software and associated documentation files (the "Software"), to deal
    in the Software without restriction, including without limitation the rights
    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
    copies of the Software, and to permit persons to whom the Software is
    furnished to do so, subject to the following conditions:
    
    The above copyright notice and this permission notice shall be included in all
    copies or substantial portions of the Software.
    
    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
    SOFTWARE.
    
  • README.mdupstream ↗

    Contents are not inlined in the catalog. Open the upstream file.

Plain-text copy, no login: /skill-bundles/ddutchie-write-like-me-skills-me.md

// REPO STATS

8 stars
0 open issues
Last commit: 1mo ago