conorbronsdon/avoid-ai-writing

avoid‑ai‑writing – A skill for AI agents that cleans up “AI‑isms”

What it does

• Provides a portable skill (a text‑based instruction file) that can be installed in a variety of AI‑agent environments (Claude Code, OpenClaw, Hermes, Cursor, Codex, etc.).
• In rewrite mode it scans a piece of prose, flags a catalog of 62 AI‑writing patterns, rewrites the text to remove them, then runs a second pass to catch any leftovers. The output is four sections: identified issues, the cleaned rewrite, a summary of changes, and a second‑pass audit.
• In detect mode it only reports the issues, grouped by severity, so you can decide whether a change is needed.
• An edit mode lets the agent make minimal, in‑place edits to a file, returning a concise edit‑report rather than the whole document.
• Optional “voice profiles” (casual, professional, technical, warm, blunt) let the rewrite adopt a desired tone.

How it works

• The core is a SKILL.md file that lists 112 word‑replacement entries (three tiers) and 62 pattern categories ranging from vocabulary‑level tells (e.g., “leverage → use”) to structural fingerprints (e.g., hashtag stuffing, placeholder tokens, uniform sentence rhythm).
• A companion detector engine (in detector/) implements 48 low‑level type categories that compute stylometric signals such as function‑word entropy, punctuation distribution, and repeated boiler‑plate phrases.
• The skill runs a two‑pass audit: the first pass flags and rewrites, the second pass re‑examines the result to catch any patterns that survived.

Installation

• The recommended way is via the community skills CLI (npx skills@1.5.23 add conorbronsdon/avoid-ai-writing). The CLI detects which agents you have installed and copies only the SKILL.md file (or the full repo if needed).
• Manual install steps are provided for each supported platform (Claude Code, Claude Cowork, OpenClaw, Cursor, Hermes, OpenAI Codex, and several others). Typically you just place SKILL.md (or a small ported rule file) in the agent’s skills directory.

Using the skill

• Once installed, you invoke it with natural language, e.g.:
  • “Remove AI‑isms from this post.”
  • “Audit this draft for AI tells.”
  • “Detect only, don’t rewrite.”
• The agent reads the SKILL.md instructions, runs the detector, and returns the structured report described above.

Why it’s useful

• Simple “make this sound human” prompts only catch the most obvious phrasing. This skill gives a structured audit with exact quotations of each flagged segment and a clear before/after view.
• The two‑pass system catches subtle leftovers that a single pass would miss (e.g., recycled transitions, lingering hedges).
• The extensive, deterministic word‑replacement table avoids vague “vibes‑based” rewrites; every flagged term has a concrete, plainer alternative.
• Because the skill is just a markdown file, it works across any agent that supports the Agent Skills format, making it a reusable building block for many LLM‑powered workflows.

Who might use it

• Content creators who want to ensure their writing doesn’t look like it was generated by an LLM.
• Teams building AI assistants that need to sanitize their own output before publishing.
• Researchers studying AI‑generated text detection, who can use the catalog of patterns as a benchmark.

---

TL;DR: avoid‑ai‑writing is a cross‑agent skill that detects and rewrites AI‑generated writing patterns, offering detailed reports and optional tone profiles. Install it via the skills CLI or manually drop the SKILL.md file into your agent’s skill directory, then trigger it with natural‑language commands like “clean up AI‑isms.”