← SkillSafe / House Style

The voice your team already writes in — found, written down, and held to

Paste two or three things you have already published. Before anything is sent anywhere, this page reads them: how long your sentences run, whether you say “you”, how often the passive hides who did the thing, which words you reach for and which spellings you cannot decide between. Then pick a lane.

Each example ships with a saved model run, so you can see a whole lane — profile, guide or review — without signing in and without spending a credit.

nothing pasted yet

Separate pieces with a line of three dashes, or a ## Heading, or a blank line between blocks.

Drop files in, or Everything is read in your browser. Nothing is uploaded until you run a lane.
Paste some copy to price the run.

Runs use the gpt-terra model and cost credits. The prescan above never does.

How this works

  1. One work object. Every lane reads the same box: prose your organisation wrote. The enforce lane adds the guide, and nothing else.
  2. The prescan is free and it is real. It splits your paste into samples, tokenises sentences properly, and measures rhythm, person, passive rate, reading grade, punctuation, heading case, serial commas, spelling splits and terminology variants. Nothing leaves your browser to do it.
  3. Credential-shaped literals are masked here, not there. Addresses, phone numbers, key-shaped strings and signed links are replaced with placeholders before the text is sent, and the model is told they were masked.
  4. Every flag gets an answer. The prompt requires the model to confirm, clear or set aside every prescan flag by id, and the reconciliation table shows any it missed.
  5. The guide is an artifact, not a paragraph. The codify lane emits style_rules JSON that the enforce lane reads back — parsed and validated in this page, with the parse error shown if it fails.

Derived from three open-source Agent Skills in anthropics/knowledge-work-plugins: @anthropics/discover-brand, @anthropics/guideline-generation and @anthropics/brand-voice-enforcement. This app is a derived work and is not affiliated with those authors.