conorbronsdon/avoid-ai-writing

avoid‑ai‑writing – a portable skill for AI agents

What it does

• Audits a piece of prose and flags AI‑writing patterns (the “AI‑isms” that give away that a text was generated by a language model).
• In rewrite mode it not only points out the problems but also produces a cleaned‑up version, runs a second pass on that rewrite, and returns a concise change summary.
• In detect mode it only reports the flags, letting you decide whether they are genuine issues.
• In edit mode it can modify a file in‑place (via the agent’s edit tool) while preserving any already‑human sections.
• An optional voice profile (casual, professional, technical, warm, blunt) controls the tone of the rewritten output.

Why it’s more than a simple prompt

• Structured audit – the skill returns four clearly separated sections (issues, rewritten text, change summary, second‑pass audit) so you can see exactly what was altered and why.
• Two‑pass detection – after the first rewrite a second scan catches any remaining tells that survived the initial edit.
• Rich pattern catalog – 74 pattern categories covering vocabulary, phrasing, structure, and tool‑specific fingerprints, backed by a deterministic detector engine (54 internal type categories). The catalog lives in references/patterns.md and is enforced by CI.
• Tiered word‑replacement table – 112 entries across three tiers (frequency markers, clarity edits, high‑density phrases) with concrete alternatives (e.g., leverage → use).

How to install The skill is distributed as a directory‑based skill that can be dropped into the skill folder of several popular AI‑agent platforms:

Platform	Installation method
Claude Code	Clone the repo into ~/.claude/skills/avoid-ai-writing or copy the self‑contained dist/avoid-ai-writing.md and reference it in CLAUDE.md.
Claude Cowork	Install from the Claude marketplace: /plugin marketplace add conorbronsdon/avoid-ai-writing then /plugin install ….
OpenClaw	clawhub install avoid-ai-writing or clone into ~/.openclaw/skills/.
Cursor	Pull the .mdc rule into .cursor/rules/ via a curl command.
Hermes, OpenAI Codex, ChatGPT plugins, GitHub Copilot, Windsurf, Cline, etc.	Drop the generated dist/avoid-ai-writing.md (or the appropriate rule file) into the tool’s rules/ or skills/ directory.

Running the skill Once installed, simply ask the agent to clean or audit a passage, e.g.:

• “Remove AI‑isms from this post.”
• “Audit this draft for AI tells.”
• “Detect AI patterns in the following paragraph.” The agent will invoke the skill in the requested mode and return the structured output described above.

Key files

• SKILL.md – the entry point that loads the pattern catalog.
• references/patterns.md – the full, human‑readable list of 74 patterns.
• detector/ – the deterministic detection engine (54 internal categories, stylometric signals, etc.).
• dist/avoid-ai-writing.md – a single‑file, self‑contained version for platforms that only accept a flat rule file.
• scripts/ – helper scripts for packaging the OpenAI plugin and validating the zip.

Use cases

• Content teams that need to ensure marketing copy, blog posts, or documentation sound human‑written rather than AI‑generated.
• AI‑agent developers who want their assistants to automatically “humanize” their own outputs before presenting them to users.
• Auditors checking for undisclosed AI‑generated text in academic, legal, or compliance contexts.

License – MIT (see LICENSE).

---

All details are taken directly from the repository’s README; no additional features are inferred.