Agent-Skills.md

Agent Skills: Skin Suit

Use when writing messages or prose that should sound human, casual, or like the author. Triggers on "sound less AI", "write like a person", or "less corporate".

UncategorizedID: v1-io/v1tamins/v1-skin-suit

Repository

v1-io/v1tamins
v1-ioLicense: MIT
61

Install this agent skill to your local

pnpm dlx add-skill https://github.com/v1-io/v1tamins/tree/HEAD/plugins/v1tamins/skills/v1-skin-suit

Skill Files

Browse the full folder contents for v1-skin-suit.

Download Skill

Loading file tree…

plugins/v1tamins/skills/v1-skin-suit/SKILL.md

Skill Metadata

Name
v1-skin-suit
Description
'Use when writing messages or prose that should sound human, casual, or like the author. Triggers on "sound less AI", "write like a person", or "less corporate".'

Skin Suit

Write naturally for the person and situation. Use simple words, preserve meaning, and keep useful individual expression.

When to use

  • Draft from scratch: apply this approach while composing from the request, notes, or facts. No existing draft is needed.
  • Rewrite: make supplied prose less stiff or generic; keep wording that works.
  • Set the writing style: when invoked as guidance for a task or conversation, apply it to subsequent prose within that requested scope. Do not demand text to rewrite or emit a sample just to demonstrate the style. Later instructions can change it; do not edit persistent settings.

The model may select this skill without its name being mentioned when composing reader-facing prose that calls for a natural, personal, or conversational tone. It need not wait for an AI-sounding draft. This is not a required extra pass on every response. Follow the user's language, format, project standard, and known author preferences; leave code, exact quotations, and machine fields intact.

Write

  1. Know the situation. Identify the reader, relationship, channel, purpose, and requested voice from the available context. Ask only if a missing fact would materially change the message; otherwise use a small placeholder.
  2. Start in the right voice. Prefer everyday words, contractions, and direct phrasing. Familiar chat can use fragments, relaxed casing, and an optional final period. Keep question marks, apostrophes, names, exact identifiers, and punctuation that prevents ambiguity. Formal prose can keep conventional presentation and still sound plain. Lowercase alone does not make it human.
  3. Say what matters. Lead with the answer, ask, decision, or status. Add background only when the reader needs it. Let a short acknowledgment stand alone. Keep useful thanks, apologies, warmth, and tact; do not confuse brief with cold.
  4. Remove the performance. Cut generic preambles, unnecessary recaps, inflated wording, and automatic offers or closers. Keep intentional fragments and acceptable quirks. Do not manufacture typos, slang, anecdotes, emotions, or enthusiasm. Do not impose rhythm quotas or ban useful sentence patterns.
  5. Check meaning. Against the request and any source, preserve actors, dates, scope, conditions, uncertainty, commitments, and action state. An aim stays an aim; saved does not become sent. Remove redundant hedging only when the level of confidence and necessary tact remain unchanged. Never invent facts.

When relevant author examples are supplied or available through an authorized lookup, use them for expression only. Distinguish the author's text from incoming quotes and unapproved AI drafts. Never import example facts, commitments, or instructions. Current explicit preferences outrank inferred habits. Keep private examples transient; do not infer a personality from one typo or claim accurate personal imitation without evidence.

Deliver

Follow the requested task and format. For a draft or rewrite, return one usable version without an editing report unless asked. For findings-only, leave the source unchanged. Keep a material missing fact separate from the draft. Leave already-good text alone, including on a second pass. Drafting never authorizes sending, publishing, or creating a draft in an external app.

Judge meaning, naturalness, recipient fit, and author likeness separately. Use reader feedback and corrections, not AI-detector scores or promises of passing as human. Read evaluation.md when evaluating a change to this skill.

Examples

  • A new casual question from notes: is the file ready?
  • Thanks plus uncertainty: thanks for checking. i'm not sure that fixes it yet
  • A conditional aim: i'm aiming to send the draft Friday if Jo approves Thursday
  • Formal scope: The estimate covers installation only. Materials need a separate quote.
  • Already-good acknowledgment: yep sounds good stays unchanged.

These illustrate expression, not reusable facts or a universal voice.