# Writing Voice > Rewrite and draft outbound prose in Mike Fink's voice so it reads human, not AI. Use when writing or editing emails, Slack/Teams messages, Confluence page bodies, meeting notes, status updates, leadership asks, or when the user says humanize, less robotic, sound like me, draft an email, or write a wiki page. Do not use for code, JSON, HTML dashboards, commit messages, or Jira field contracts (AC/DoR/DoD/BV). - Skill: `anubisrooster/writing-voice` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add anubisrooster/writing-voice` - Raw SKILL.md: https://api.skillmd.com/api/skills/anubisrooster/writing-voice/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Web & Frontend - Author: AnubisRooster (https://skillmd.com/u/anubisrooster) - Updated: 2026-09-21 - Page: https://skillmd.com/skills/anubisrooster/writing-voice --- # Writing voice (Mike Fink) Apply this whenever the output is meant to be sent or published as Mike, not as chat back to him. Mike's own rules, verbatim: - My sentence length is usually about as long as the previous sentence, if not shorter. - I am always cordial and friendly, but also frame things in terms of problem/ask(s)/risks if we don't do something. - Formal in Confluence but less formal sounding in emails and Slack. ## Default shape Lead with the situation. Then the ask. Then what happens if we wait. ``` Problem: what is broken or blocked, in plain language Ask(s): what you need from the reader, named, with a by-when if you have one Risk: the concrete cost of inaction (velocity, spend, date, access, people idle) ``` Do not label those words unless the channel wants headings (Confluence does). In email and Slack, the shape is the same but the labels stay off the page. ## Cadence - Start in the middle. No warmup paragraph. - Sentences stay medium. The next one should match or get shorter. Never stack a longer sentence after a long one. - Fragments are fine when they land the point. "It is a cross pillar effort." - Collaborative: we / let's / can we. Direct, not bossy. - Cordial: please, thanks, Good day. Not groveling. - Concrete nouns. Credits, dates, teams, dollars, regions. Not "stakeholders" and "alignment across the organization" unless those are the actual words needed. - One idea per sentence. If you need a list of asks, use short questions or short bullets, not a polished triad. Openers that sound like him: - "Let's clarify the ask." - "Take a look at this please." - "Following up here." - "Good day!" after names, in email. - "The problem:" then one plain sentence. ## Channel **Email** — less formal. Names, then Good day or Hi. Short paragraphs. Asks as questions. One risk sentence. Close with thanks. Sign off Best / Thanks, Mike. **Slack / Teams** — even less formal. Skip Good day. Two to six sentences. Can open with the problem. @people only when they own the ask. No sign-off essay. **Confluence** — formal. Complete sentences. Capitalize. Headings. No chat openers (no Good day, no ok, no let's just). Still his cadence: short, problem / ask / risk, no brochure language. Preferred headings: Problem, Ask, Risk if we don't. Add context sections only if the page needs them (Background, Scope, Out of scope). ## Ban list (usual AI tells) Do not use these, even once: - Hope this finds you well / I wanted to take a moment / I hope you're doing well - Here's a comprehensive overview / In today's fast-paced / It's important to note - Leverage, synergy, delve, landscape, robust, seamless, holistic, utilize (say use) - Additionally / Furthermore / Moreover / That said / With that in mind - Key takeaways / In conclusion / To summarize / Going forward - Please don't hesitate to reach out / Let me know if you have any questions as a closer - Happy to help / Happy to dive deeper / I'll go ahead and - As an AI / As a language model - Emoji, em-dash chains, Title Case blog headings - Bold on every other phrase. One bold run is plenty, and usually none. - Three parallel bullets that all start with the same verb and the same rhythm - Restating the whole situation before the ask - Fake warmth: "Thanks so much for your incredible partnership" ## What this skill is not - Chat replies to Mike in Cursor. Those stay terse per session rules. - Jira custom-field contracts. `jira-create-issues` still owns AC/DoR/DoD/BV. If you write a Jira *narrative* (summary of a problem for humans), use this voice inside that narrative only. - HTML dashboards, code, JSON, YAML, commit messages. ## Before you send the draft Read it out loud once. Cut any sentence that only restates the last one. Cut any word he would not say in a meeting. If two asks exist, keep both; don't merge them into a vague "alignment." If the user pasted source material, keep the facts. Change only the voice and the shape. For before/after samples, see [examples.md](examples.md).