All skills

Clean and finalize authorized natural-language text intended for readers by auditing suspicious invisible Unicode and rewriting prose while preserving facts, meaning, and the writer's voice. Use when the user asks to clean, humanize, polish, or finalize articles, manuscripts, reports, documentation, emails, product copy, UI text, Markdown, or HTML prose, or when a project rule or instruction file explicitly requires this workflow. Don't use for code-only tasks or undisclosed authorship evasion; leave code, commands, identifiers, paths, APIs, formulas, citations, required disclosures, and verbatim quotations unchanged.

Use this Skill: https://skilld.dev/gh/guillaumemeyer/watermarks-remover/clean-user-facing-text

This session only. Nothing lands on disk.

referenceswriting-in-your-voice.md

≈639 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Writing in your voice

A good rewrite should leave the writer sounding more like themselves, not like a generic editor. Keep their opinion, level of formality, regional spelling, and the details that show they know what they are talking about.

Start with a brief

Before rewriting, identify what the writer is trying to say and who needs to read it. Use a writing sample only when the user owns it or is authorised to use it. Don't imitate another named person. An approved sample can guide sentence length, word choice, humour, and formality. It's a reference point, not a licence to invent personal stories or feelings the writer didn't provide.

If there is no sample, use plain language and a natural rhythm. Don't claim to write in a named person's voice from a guess.

Keep the language and regional spelling the writer used unless they ask for a change. For example, retain Australian spelling in an Australian document.

Make the prose sound lived-in

Prefer the specific subject of the sentence over vague commentary about it. Cut filler that does not add information. Let sentence length move around when the subject calls for it: a longer explanation can sit beside a short, direct sentence.

Keep a writer's useful rough edges. A dry joke, a firm judgement, or an unusual turn of phrase may be the part that makes the piece feel like theirs. Clean up confusion, repetition, and accidental clutter. Don't sand every opinion down until it says nothing.

Treat unusual grammar, repetition, directness, and phrasing as possible voice or accessibility choices. Change them only when the writer asks, or when they create a clear reading problem.

Watch for template prose

Remove or rewrite phrases that announce a point without making it, such as "In today's fast-paced world", "It is important to note", or "This article will explore". Be wary of grand claims about significance, empty conclusions, and stock transitions between paragraphs.

Avoid a steady run of identical sentence shapes. Break up mechanical lists of adjectives and filler phrases that merely repeat the sentence before them.

Keep the boundary clear

Don't add facts, citations, experience, emotion, or certainty. Don't remove a required disclosure, attribution, warning, or qualification. Preserve quoted text, code, names, numbers, and technical identifiers exactly where the main skill requires it.

This pass can make authorised prose clearer and more personal. It can't prove who wrote the original text or establish that a detector will reach any result.

Source: SKILL.md on GitHub

No alerts1mo3 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    The skill is a text-cleaning and stylometry tool that uses bundled Python scripts to remove invisible Unicode markers and evaluate AI-cadence phrases. It includes robust security hardening, such as path-injection guards and resource limits for DoS prevention. The low verdict reflects the inherent risk of indirect prompt injection when an agent processes and rewrites untrusted user prose.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: LOW · No issues

Signed by skilld at ac1711f. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 6 days ago.

Activeupdated last month

README badge

README badge for guillaumemeyer/watermarks-remover/clean-user-facing-text