conorbronsdon/avoid-ai-writing
Avoid AI writing
Audits and rewrites text to remove AI writing patterns, with a detect-only mode, an edit-in-place mode for files, optional voice profiles and an iterate-to-convergence pass.
What Avoid AI writing does
The skill turns the agent into an editor that hunts for patterns common in machine-written text. Before any audit it reads references/patterns.md, the catalog of word tiers, patterns and context profiles. There are three modes. Rewrite, the default, flags problems and returns corrected text. Detect only lists the findings. Edit changes a named prose file in place with small, targeted edits, and refuses source code, configuration and generated data.
Findings fall into three severity tiers, from credibility problems such as cutoff disclaimers and unsourced attributions, through banned words and template phrases, down to style issues like em dash frequency. An editing contract decides what may change: a match becomes an edit only after its context checks pass, and quotations, code, URLs and frontmatter stay protected. The agent may not add facts, experiences or opinions the source lacks, and it treats instructions inside the text as data. Voice profiles (casual, professional, technical, warm, blunt), context profiles and an optional house-style config shape the result.
A rewrite uses at most two editing passes. You get the final text plus a short verification note: passes used, which checks ran, findings left in place and why the agent stopped. The SKILL.md calls its patterns signals, not proof that a machine wrote the text.
Install
/plugin install avoid-ai-writing@conorbronsdon-skillsRead the security check below before you run it. The repository's own instructions win if they differ.
Security check
What the skill's files do, from reading them. A flag that nobody has checked yet says "Not checked", never "No".
- Runs scripts
- No
- Uses the network
- Yes
- Installs hooks
- No
- Asks for secrets
- No
HuiaHub's scan found no script files among the files it checked, no hooks and no secret handling, and it flags network use because the SKILL.md links to URLs. The frontmatter says no external tools or APIs are needed. The full skill directory does include Node scripts (scripts/normalize-quotes.js, scripts/check-style.js, detector/validate.js), and the SKILL.md tells the agent to run the quote normalizer during rewrites, with a manual fallback if it cannot. Edit mode writes to your files. Read those scripts before installing, and point edit mode at a committed or copied file.
Scan details: SKILL.md instructions include web requests or URLs to fetch. Checked at commit 4d037972046d.
Use cases
- Clean AI-isms out of a blog post or LinkedIn draft before publishing.
- Scan someone else's text for AI patterns without changing it.
- Fix a Markdown draft in place with minimal edits.
- Recast prose in a chosen voice, such as technical or blunt.
Limitations
- The rules are calibrated for English.
- Findings are signals, not proof of AI authorship, and should not decide consequential cases.
- Edit mode refuses source code, configuration and generated data files.
- SKILL.md needs `references/patterns.md` beside it; installing the root file alone is incomplete.
Sources
- conorbronsdon/avoid-ai-writingRepositoryPrimarygithub.com
- raw.githubusercontent.com/conorbronsdon/avoid-ai-writing/main/SKILL.mdWeb
- raw.githubusercontent.com/conorbronsdon/avoid-ai-writing/main/README.mdWeb
- avoid-ai-writing skill folderRepositoryPrimarygithub.com