What problem does it solve? Text written by AI assistants carries recognizable patterns: filler phrases, em dash overuse, vague attributions, sycophantic openers, and conversation residue that only makes sense to the session that produced it. This Skill gives an agent a concrete, citable rule set to detect and remove those patterns from docs, READMEs, docstrings, comments, UI copy, commit messages, and chat replies. ## Core Features & Use Cases - Rule catalogue with stable ids: CATALOGUE.md holds dozens of named tells (AI vocabulary, rule of three, throat-clearing, nominalizations, false agency), each tagged by scope (artifact, chat, or both) so a finding can cite a rule id, quote the hit, and name the fix. - Cold-reader and single-source principles: SKILL.md adds structural rules beyond the catalogue, covering conversation residue (change narration, reviewer reassurance), one authoritative home per fact, and referencing tickets by title rather than bare ids. - Automated chat review hook: chat_review.py is a Claude Code Stop hook that runs a small reviewer model over each final reply against the chat-scoped rules, feeds flagged passages back for revision, fails open on errors, and logs every decision as JSONL. - Use Case: Run the skill on a drafted README to strip phrases like "seamlessly", "it's important to note", and "In today's fast-paced landscape", replacing them with plain statements of fact, or install the hook so every chat reply is screened before the user sees it. ## Quick Start Ask the agent to review a file with the writing-for-humans skill, for example: de-slop docs/guide.md against the AI prose catalogue and rewrite every flagged sentence.
