Source-respecting rules for drafting, rewriting, copyediting, and auditing durable text.
betterwords.md is the product. The skill, plugin manifests, and platform notes are convenience methods for loading that file in different hosts.
The cross-agent Skills CLI installs betterwords for the current project by default:
npx skills add aritusama/betterwordsInstall it for the current user instead:
npx skills add aritusama/betterwords -gThe installer discovers the single skill at skills/betterwords.
Add this repository as a Codex marketplace:
codex plugin marketplace add aritusama/betterwords
codex plugin add betterwords@betterwordsYou can also install betterwords from Plugins in the ChatGPT desktop app after adding the marketplace. This repository marketplace is separate from the public Plugins Directory; betterwords does not claim a public directory listing.
- Claude supports the same skill and a namespaced Claude Code plugin. See Claude setup.
- Gemini CLI supports extension and standalone-skill installation; Antigravity supports the plugin folder. See Gemini and Antigravity setup.
- GitHub Copilot can use repository instructions. See GitHub Copilot setup.
- Hosts without skill support can use
betterwords.mdwith the short instruction adapter. See ChatGPT setup for custom GPT and Project use.
Use betterwords to copyedit this release note without changing supported facts.
Use betterwords to audit this draft for quality.
Betterwords is a writing-quality system. It cannot determine whether a person or model wrote a text. Do not use it to label authorship, evade required disclosure, or fabricate sources, quotations, experience, or evidence.
betterwords is independently written. These sources shaped named parts of the ruleset:
- William Strunk Jr.'s The Elements of Style, later expanded by E. B. White, established the baseline for active voice, positive form, concrete language, paragraph unity, parallel structure, related words, emphasis, and removing needless words.
- George Orwell's Politics and the English Language informed the meaning-first rule, plain wording, stale-metaphor checks, active voice, and the instruction to cut words that do no work.
- ASD-STE100 Simplified Technical English, Issue 9 informed rule 6.12 on procedural steps, simultaneous actions, conditions, and notes. Betterwords applies that narrow writing guidance; it does not replace the full standard or its controlled dictionary and does not claim STE compliance.
- The Human Detectors paper and its annotator data informed checks for manufactured sourcing, quotation voice and integration, comma-tailed sentences, speech-tag cycling, generic headings, pre-summarizing introductions, sanitized framing, and nominalization.
- StoryScope informed structural checks for over-tidy composition, prompt echo, and loss of useful complications. These checks are applied as writing-quality diagnostics, not as proof of authorship.
- The Rise of Verbal Tics in Large Language Models informed the aggregate treatment of repetitive formulaic patterns. EQ-Bench Slop Score and Hamed Paydarfar's editorial field report, 5 Dead Giveaways You Are Reading AI-Generated Text, supplied comparison examples for contrastive and sentence-shell patterns. These sources do not validate authorship classification; the rules use them only as writing-quality prompts.
- Wikipedia's editorial field guide, Signs of AI writing, supplied additional observed phrase, structure, and formatting patterns.
- The Tropes directory inspired contextual checks for false-suspense setups, compressed significance adverbs, coined concept labels, repeated substance, and historical analogies used as accumulated proof. Betterwords adapted these observations as writing-quality rules, not phrase bans or authorship signals.
- blader/humanizer and Pangram served as independent comparison points for coverage and failure modes. betterwords is not a fork of either project and does not optimize text for detector scores.
MIT.