agentsclimarketplace

Clear writing

Skill gigio1023/clear-writing/clear-writing

Use when creating, restructuring, rewriting, humanizing, or reviewing prose documents in English or Korean: README, guides, specs, API references, ADRs, memos, wiki pages, blog drafts. Covers de-AI-ifying tone ("humanize", "deslop", "AI 티 빼줘"), turning notes into standalone docs ("이 메모를 문서로"), surgical Korean polishing ("윤문해줘"), terminology review ("용어 이상한지 봐줘", "term check"), and repository-grounded doc authoring ("README 작성", "문서 정리해줘", "스펙 작성"). NOT for Python docstrings, prompt coaching, diagrams, PR/commit/issue copy, translation, or implementation changes.From its SKILL.md

Install
npx -y skills add gigio1023/clear-writing --skill clear-writing

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.9 KB, ~1.3k tokens by cl100k_base, as published. Nobody here has run it

Clear Writing

One skill for prose-document work: author repository-grounded docs, revise existing text, and run focused passes (terminology, Korean patterns). Chat answers do not need this skill; the always-on core rules govern those.

Intake

Establish before touching text:

  1. Deliverable and audience — what document, for whom, doing what job.
  2. Language — Korean work additionally loads korean-tells.md.
  3. Grounding — repository-based (claims must match code/config) vs standalone text (claims must match the source text).
  4. Authority — "review/진단/check/피드백" returns findings without edits (quote the line, name the pattern, state the fix); "다듬어줘/고쳐줘/써줘/적용해줘" authorizes edits within the given scope. Multi-file or repo-wide requests: confirm the target list first, then process sequentially under the same discipline. Publishing, commits, PRs, or posting need a separate explicit request.
  5. Profile — default, or strict per profiles.md. Repo style rules outrank profiles.
  6. Voice sample — if the user's own writing is available, it outranks every style rule (voice-preservation.md).

Job selection

JobWhenPrimary reference
Authoringnew or materially updated repo-grounded docsauthoring.md
Revisionexisting text: humanize, restructure, compose notes into a docrevision.md
Passsingle-concern check: terminology, or Korean 윤문 onlyterminology.md / korean-tells.md

Mixed requests are normal: "README 정리해줘" with stale facts = authoring grounding + revision prose work. Passes also run as sub-steps inside the other two jobs; fold their findings into the main report.

Invariants (every job)

  • Preserve facts, numbers, commands, paths, quotations, conditions, requirement levels, and visible uncertainty. Style cleanup never resolves a factual conflict silently.
  • Removal only. Never insert stock transitions, invented facts, examples, citations, or replacement clichés.
  • Voice wins over rules: when a pattern rule fires on the author's deliberate, repeated choice, report it instead of applying it.
  • No terminology replacement without domain grounds; never alter identifiers, schema fields, commands, or quoted text without explicit request.
  • Never claim text is AI-authored; report the wording problem itself.
  • core-rules.md is the generation baseline for everything this skill writes, including its own reports.

Delivery

Every deliverable passes gates.md: reader/purpose, fact preservation, language quality, change-rate guard (30% warn / 50% stop), editor-slop self-check on your own rewrite, and verification actions. Lead the response with the deliverable; keep the report to files touched, major decisions, checks run, and unresolved spans. No grades, no change-rate percentages as scores.

Reference files

FileLoad when
core-rules.mdalways-on baseline; D4 install decision pending
authoring.mdauthoring job
templates.mda page shape would save time (menu, not mandate)
doc-patterns.mdtechnical-doc content problems need worked repairs
style-zoom-rules.mdchoosing a structural repair level
revision.mdrevision job
anti-slop-core.mdany humanize/deslop work, all languages
structure-anti-patterns.mdstructural verbosity, doc-shape cleanup
voice-preservation.mdany rewrite of someone's text
korean-tells.mdKorean documents
terminology.mdterminology pass
known-slop-terms.mdbroad terminology scan
replacement-patterns.mddrafting term alternatives
verification-procedure.mda term decision needs external evidence
profiles.mdstrict house style requested or implied by repo policy
gates.mdbefore every delivery

Gotchas

  • Phrase swapping is not humanizing; genericness is the real defect. Add concrete, checkable detail or the text stays hollow.
  • Compression is not clarity; keep the connective reasoning readers need.
  • Do not reformat chat/DM/email into formal documents.
  • "Professional" is not the goal; natural, precise, and credible is.
  • Do not grade documents by pattern counts or change ratio, in either direction.
  • Stop and ask instead of guessing when: the target text or scope is missing; repository rules conflict; a style fix cannot resolve a factual error or ambiguity; you cannot tell duplicated filler from essential context; edits would collide with a user's uncommitted changes.

Keep looking

Skills are one crate of 326,422. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.