Draft, revise, or review a substantive writing deliverable when editorial quality is itself part of the requested outcome. Use for authored documentation, reports, public copy, or explicit editing and style critique. Do not trigger for ordinary conversation, engineering reasoning, explanations of current work, or requests for longer or more expressive answers alone.
Scanned 9/20/2026
Install to Claude Code
npx -y skills add macintog/codex-spine --skill prose-quality --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Prose Quality?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/macintog-prose-quality)More formats (shields.io, HTML) on the badges page.
---
name: prose-quality
description: Draft, revise, or review a substantive writing deliverable when editorial quality is itself part of the requested outcome. Use for authored documentation, reports, public copy, or explicit editing and style critique. Do not trigger for ordinary conversation, engineering reasoning, explanations of current work, or requests for longer or more expressive answers alone.
---
# Prose quality
Improve the writing without changing what is true, authorized, or meant.
## Choose the register
Choose depth, voice, and form for the audience and requested deliverable. Use
expressive prose when useful or requested, without a prescribed persona or an
elaborate-writing checklist. In an expansive register, local brevity and
house-style defaults do not constrain expression; source protection still does.
Follow requests to shorten or change tone. Ordinary conversation does not need
this workflow merely to become thoughtful or expressive.
## Protect the source
- Bind the audience, purpose, deliverable, and requested degree of editing.
- Preserve exact commands, paths, symbols, measurements, citations, legal text,
quotations, and deliberately literal transcript excerpts.
- Treat a request to review prose as read-only unless the user also asks for an
edit. Do not turn prose polish into code, product, or policy changes.
- Do not replace causal analysis, code review, log interpretation, or other
substantive work merely because its answer will contain prose. Apply this
skill only when the wording is itself part of the requested outcome.
- Treat old drafts, checkpoints, queues, and generated next steps as source
material only. They cannot select a new writing task. Finishing one revision
does not authorize related rewrites or another cleanup pass.
- Do not fabricate facts, certainty, attributed opinions, first-person
experience, or a persona. Evidence-backed recommendations in your own voice
remain appropriate when relevant to the task.
## Choose the relevant guidance
In the expansive register, the references below supply source protection and
factual checks only. Their stylistic prescriptions do not govern the response
unless the user specifically requests them.
- Read [references/prose-patterns.md](references/prose-patterns.md) when
drafting or revising covered prose, or when the user asks to remove AI tells,
slop, filler, jargon, stiffness, or generic wording.
- Also read [references/technical-prose.md](references/technical-prose.md) for
documentation, READMEs, RFCs, procedures, technical explanations, reports,
PR descriptions, or commit messages. Do not force documentation structure
onto short conversational answers or product UI copy.
- Read [references/source-and-license.md](references/source-and-license.md)
only when auditing provenance, redistributing this packet, or reusing its
source wording.
## Revise in this order
Use this sequence for direct prose. For expansive prose, preserve claims and
protected source material, and choose your own editorial approach.
1. Preserve the intended claim and evidence boundary. Correct unsupported or
ambiguous claims before polishing sentences.
2. Put the answer, decision, or required action where the reader needs it.
Remove duplicated setup and structure the rest around the reader's job.
3. Replace vague abstractions with the actual actor, mechanism, symbol,
command, number, or consequence. Keep one name for each thing.
4. Cut filler and synthetic patterns without flattening the writer's voice.
Vary rhythm naturally and state considered judgment only when the genre and
evidence permit it.
5. Read the result once for meaning and once for line-level clarity. Verify
that protected literal material, links, facts, units, and scope survived.
Return the revised prose, or ranked findings when the task is review-only.
Mention a material factual ambiguity instead of silently writing around it.
Report newly noticed follow-up work as a finding and stop.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!