Strips AI writing patterns from prose drafts like emails, docs, reports, blog posts, commit messages using rules against negation-contrast, three-item lists, and repetition.
From bopen-toolsnpx claudepluginhub b-open-io/claude-plugins --plugin bopen-toolsThis skill uses the workspace's default tool permissions.
evals/benchmark.jsonevals/evals.jsonreferences/examples.mdreferences/phrases.mdreferences/structures.mdreferences/words.mdSearches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Searches prompts.chat for AI prompt templates by keyword or category, retrieves by ID with variable handling, and improves prompts via AI. Use for discovering or enhancing prompts.
Executes pre-written implementation plans: critically reviews, follows bite-sized steps exactly, runs verifications, tracks progress with checkpoints, uses git worktrees, stops on blockers.
Strip AI writing patterns from prose. Follow these three rules exactly.
Before writing any sentence, check: does it say what something ISN'T before saying what it IS? Or does it append "not just X" to trash the alternative and prop up the claim? If either, delete the negative part. Keep only the positive claim. Let it stand on its own.
Examples:
When you write a list, count the items. If there are exactly 3 items in parallel structure, either remove one (making it 2) or add one (making it 4).
This applies everywhere: adjective lists, verb lists, noun lists, parallel sentences.
Never restate the same point in different words for emphasis. Say it once clearly and move on. If the first sentence already conveys the meaning, the restating sentence is dead weight.
After writing, check the last sentence of every paragraph. Count its words. At most ONE paragraph may end with a sentence under 15 words. All other paragraphs must end with a sentence of 20+ words that includes a specific detail, number, or example.
After writing any prose, you must do a concrete revision pass before delivering. This is not optional — the first draft will contain AI patterns no matter how carefully you write it. Do these checks mechanically:
Do this revision pass silently — don't mention it in your output. Just deliver the cleaned text.
Before delivering revised prose:
After editing, rate the text 1–10 on each dimension:
| Dimension | Question to ask |
|---|---|
| Directness | Does it state facts, or announce them? |
| Rhythm | Are sentences varied in length and structure? |
| Trust | Does it respect reader intelligence? |
| Voice | Does it sound like a person wrote it? |
| Density | Is anything cuttable without losing meaning? |
Below 35/50: revise again.
AI version (bad):
Speed. We don't just build fast infrastructure — we build infrastructure that's fast, reliable, and scalable. Your transactions process in milliseconds, not minutes. That's the difference.
Humanized version (good):
Speed. Transactions process in under 200ms. We cache settlement data locally so round-trips to the clearinghouse don't block your checkout flow. Most integrations go live in a day or two.
Notice what changed: the binary contrast ("don't just X — we Y") is gone, the tricolon ("fast, reliable, and scalable") became specific claims, the punchline ending ("That's the difference") became a practical detail, and the paragraph ends with a long informational sentence rather than a mic-drop.
Good humanized prose: