# avoid-ai-writing

> An open-source agent skill that audits and rewrites content to remove AI writing patterns, supporting both rewrite and detection-only modes.

avoid-ai-writing is a portable writing skill for AI agents that audits and rewrites content to eliminate AI writing patterns ("AI-isms"). It works with Claude Code, OpenClaw, Hermes, and any agentskills.io-compatible agent. The skill uses a structured two-pass detection system with 36 pattern categories and a 109-entry word replacement table to catch and fix AI tells with precision.

- **Rewrite mode** — *the default mode; flags AI patterns and rewrites the text, then runs a second pass to catch patterns that survived the first edit.*
- **Detect mode** — *flags AI patterns without rewriting, grouping flags by severity (P0/P1/P2) and distinguishing real problems from judgment calls.*
- **109-entry word replacement table (3 tiers)** — *Tier 1 words are always flagged, Tier 2 words flag when they cluster, and Tier 3 words flag only at high density, reducing false positives.*
- **36 pattern categories** — *covers content patterns (significance inflation, promotional language), language patterns (copula avoidance, synonym cycling), structure patterns (em dash overuse, inline-header lists), communication patterns (chatbot artifacts, sycophantic tone), and meta patterns (rhythm uniformity, over-polishing).*
- **Two-pass detection** — *the second pass re-reads the rewrite and catches recycled transitions, lingering inflation, and copula swaps that snuck through the first edit.*
- **Claude Code installation** — *clone the repo into `~/.claude/skills/avoid-ai-writing` or copy `SKILL.md` directly; optionally set up as a slash command.*
- **OpenClaw installation** — *install via `clawhub install avoid-ai-writing` or clone into `~/.openclaw/skills/avoid-ai-writing`.*
- **Web app** — *a community-built web app at avoid-ai-writing-app.vercel.app lets users paste text and run a full audit and rewrite by burning 1,000 $avoid Solana tokens.*
- **Structured output** — *rewrite mode returns four discrete sections: issues found, rewritten version, change summary, and second-pass audit.*

## Features
- Rewrite mode with two-pass AI pattern detection
- Detect-only mode with severity grouping (P0/P1/P2)
- 109-entry word replacement table across 3 tiers
- 36 pattern categories covering content, language, structure, communication, and meta patterns
- Compatible with Claude Code, OpenClaw, and Hermes agents
- Slash command support for Claude Code
- ClawHub installation support for OpenClaw
- Community web app with $avoid token burn mechanism
- Structured four-section output in rewrite mode
- Rewrite-vs-patch threshold detection

## Integrations
Claude Code, OpenClaw, Hermes, agentskills.io, ClawHub

## Platforms
CLI, API, WEB

## Pricing
Open Source

## Links
- Website: https://newsletter.chainofthought.show/
- Documentation: https://github.com/conorbronsdon/avoid-ai-writing
- Repository: https://github.com/conorbronsdon/avoid-ai-writing
- EveryDev.ai: https://www.everydev.ai/tools/avoid-ai-writing