Agent Skills
› mohitagw15856/pm-claude-skills
› plain-language-rewrite
plain-language-rewrite
GitHub将专业术语密集文本改写为通俗易懂语言，同时严格保留事实、数据和限定词。通过忠度检查、术语清单和受众校准，确保内容精准且易于非专家理解。
Trigger Scenarios
使文本更易读
为非专家翻译内容
消除公告中的行话
让父母或新员工能理解的改写
Install
npx skills add mohitagw15856/pm-claude-skills --skill plain-language-rewrite -g -y
SKILL.md
Frontmatter
{
"name": "plain-language-rewrite",
"homepage": "https:\/\/mohitagw15856.github.io\/pm-claude-skills\/skill\/plain-language-rewrite.html",
"metadata": {
"openclaw": {
"emoji": "🧠"
}
},
"description": "Rewrite jargon-dense text into plain language without losing precision — the translation pass that keeps every fact and qualifier, the jargon triage (terms to replace, terms to keep-and-define), and the reading-level honesty for the actual audience. Use when asked make this readable, translate this for non-experts, de-jargon this announcement, or rewrite this so my parents\/customers\/new hires understand it. Produces the rewrite with a fidelity check, the jargon ledger, and the kept-terms glossary line."
}
Plain Language Rewrite Skill
Plain language has a bad reputation because it's done badly — "simplifying" that quietly drops the qualifiers, rounds the numbers, and turns precise claims into vibes. Done right, it's a translation with a fidelity guarantee: every fact, number, condition, and hedge in the original survives; what changes is the packaging — sentence length, word choice, structure, and the jargon triage (most terms translate; some are load-bearing and get kept-and-defined instead, because renaming a term of art creates confusion downstream). The test is a reader who knows nothing and a checker who knows everything, both satisfied.
What This Skill Produces
- The rewrite — the same content at the target reading level, structured for the new reader
- The fidelity check — facts/numbers/qualifiers in original vs. rewrite, diffed to zero loss
- The jargon ledger — each term: translated (to what) / kept-and-defined (why it's load-bearing)
- The audience calibration — what the target reader knows, and the one-notch rule applied
Required Inputs
Ask for these if not provided:
- The text — verbatim; translation works on actual sentences
- The audience, concretely — "customers who don't know insurance" beats "general public"; the rewrite calibrates to what they already know, and one text can't serve experts and novices simultaneously (say so when asked to)
- The load-bearing terms — which jargon is contractual, regulatory, or searchable-by-the-reader (those get kept-and-defined; renaming "deductible" to "your share" helps nobody who then reads their policy)
- The stakes — legal/medical/financial texts keep their hedges verbatim-in-meaning; marketing texts have more freedom — the fidelity bar scales up with stakes
Framework: The Translation Rules
- Fidelity is the contract: before rewriting, extract the original's claims list (facts, numbers, conditions, hedges); after, diff the rewrite against it. "Usually," "up to," and "unless" are content, not decoration — a plain rewrite that drops them is a different document with better fonts.
- The jargon triage, not the jargon purge: each term gets a verdict — translate (internal shorthand, latinate padding: "utilize"→"use", "prior to"→"before") or keep-and-define (terms of art the reader will meet again: "your deductible — the amount you pay before insurance starts paying"). The ledger records both, because reviewers will ask.
- Sentence surgery before word surgery: the biggest readability gains are structural — one idea per sentence, front-loaded subjects, lists for parallel items, 20-ish words average. Swapping words inside 60-word sentences polishes the unreadable.
- Calibrate one notch below the audience's floor: write to slightly simpler than the audience needs — comprehension costs nothing extra for strong readers, and the weakest reader in the audience is the one the text exists to reach. But never condescend in tone: plain is respectful; simplified-with-a-pat-on-the-head is not.
- The two-reader test: a target-audience reader retells the content correctly (comprehension) AND a subject-matter expert confirms nothing became false (fidelity). Passing one test is easy; the skill is passing both, and the output states both results.
Output Format
Plain Rewrite: [text] — for [audience]
The Rewrite
[The translated text, structured for the reader]
Fidelity Check
| Original claim/number/qualifier | Where it survives in the rewrite |
|---|---|
| [Zero-loss demonstrated, or the flagged intentional cuts with reasons] |
Jargon Ledger
| Term | Verdict | Rendered as |
|---|
The Two-Reader Result
[Comprehension: retellable? · Fidelity: expert-clean? · the stakes note for legal/medical texts: hedges preserved verbatim-in-meaning]
Quality Checks
- The claims list was extracted before rewriting and diffed after
- Every qualifier survives — no "usually" became "always"
- Load-bearing terms were kept-and-defined, not renamed
- Sentence structure did the heavy lifting before word swaps
- Both readers pass: the novice retells it; the expert signs it
Anti-Patterns
- Do not simplify by deletion — dropped hedges are the classic plain-language malpractice
- Do not rename terms of art — the reader meets "deductible" again five minutes later
- Do not write down at the audience — plain and respectful are the same register
- Do not serve two audiences with one text — name the split when asked to
- Do not skip the expert pass on high-stakes text — readable-but-wrong is the worst outcome on the board
Version History
- 471c606 Current 2026-07-22 13:35
