samgalanakis/pilcrow

Pilcrow is a writing linter and editor skill suite that plugs into AI coding tools (like Claude Code, Cursor, or GitHub Copilot). A linter normally catches code style problems automatically, pilcrow does the same thing for written prose, applying rules from well-known style guides by Strunk, Williams, Zinsser, Pinker, Orwell, and King. It works in two modes. The command-line tool (CLI) analyzes markdown, plain text, or HTML files and reports writing problems with their exact location, a snippet of the offending text, and an optional suggestion. These checks include both deterministic rules (things a pattern-matcher can catch reliably) and LLM-judged rules (things that require language understanding, like spotting "AI tells", phrases that sound like they were written by an AI). It never edits your text, it only reports findings. The skill layer runs inside your AI assistant. Running /pilcrow with no arguments finds the prose in scope, identifies what genre it is (essay, tutorial, readme, etc.), lints it, and proposes a sequence of editor commands for you to confirm before anything changes. Named commands like "polish," "tighten," "humanize," "clarify," and "argue" each apply a specific methodology drawn from one of the style guides. A separate "document" command scans your existing writing to build a VOICE.md file capturing your personal style, which then influences future suggestions. You would use pilcrow if you write a lot of documentation, blog posts, or any prose in a code project and want systematic feedback beyond spell-checking. The full README is longer than what was provided. It is built in TypeScript and distributed as an npm package.