good-prose（好散文）

Good Prose: Write and Edit Professional, Clear, Human-Sounding Text

Use this skill to draft or revise prose that can stand in professional correspondence, public blogs, newspaper opinion pieces, or longform emails with friends or family. Aim for clarity, specificity, and a measured tone without slang or hype.

Invocation Notice

• Inform the user when this skill is being invoked by name: good-prose.

Directives

Use these as the core craft rules. Apply them before any optional style tweaks.

Composition

• Make the point early: Lead with the main claim or request.
• Organize by paragraph: One paragraph = one idea; open with a clear topic sentence.
• Prefer the active voice: Use passive only when the actor is unknown or irrelevant.
• Use concrete verbs: Replace nominalizations with verbs.
• Keep sentences lean: Remove dead weight, redundant phrasing, and filler.
• Emphasize by position: Put the most important words at the end of the sentence.
• Maintain parallel structure: Keep lists and paired clauses grammatically aligned.
• Avoid dangling modifiers: Introductory phrases must attach to the subject.
• Keep related words together: Avoid long interruptions between subject and verb.

Style

• Prefer plain words: Use familiar words over ornate synonyms.
• Avoid threadbare openings: Skip "One of the most..." and similar cliches.
• Be specific: Replace vague adjectives with facts, numbers, or examples.
• Use "is/are/has" when accurate: Avoid "serves as," "stands as," "boasts."
• Reduce intensifiers: Cut "very," "highly," "extremely," unless essential.
• Avoid euphemism and padding: Say what happened, not what it "represents."

Mechanics

• Comma use: Use a comma before a coordinating conjunction joining two independent clauses.
• Don't splice clauses: Use a semicolon or period instead of a comma.
• No sentence fragments: Fragment only for deliberate emphasis.
• Place commas in series: Use the Oxford comma when listing three or more items.
• Apostrophes: Form possessive singular with 's, including names ending in s.

AI Guardrails

• Remove significance inflation, promotional tone, and vague attribution.
• Replace "highlighting/underscoring/showcasing" with direct statements.
• Avoid negative parallelism and forced rule-of-three lists.
• Remove chatbot fillers ("Hope this helps," "Great question," etc.).
• End with concrete outcomes or next steps, not generic optimism.

LLM's have characteristic writing style that should be avoided. Refer to references/humanize-guardrails.md for LLM failure patterns and rewrite guardrails.

Quick Intake

Ask only if missing:

• Purpose (inform, persuade, explain, request, reflect)
• Audience and relationship
• Medium and length (email, blog post, op-ed, memo, etc.)
• Tone constraints (formal, friendly, firm, neutral)
• Any must-keep facts or phrasing

If given a draft, do not ask questions unless the intent is unclear.

Workflow

1. Find the spine: Identify the central claim, request, or takeaway.
2. Shape the structure: Lead with context and stakes, then evidence, then implications or next steps.
3. Tighten sentence craft: Prefer concrete verbs, specific nouns, and short phrases.
4. Adjust voice: Keep it professional and human; remove hype, filler, and vague claims. If AI artifacts appear, consult references/humanize-guardrails.md.
5. Polish mechanics: Fix grammar, punctuation, and consistency.

Core Principles

• Clarity first: Put the point early. Move background below it.
• Specific > abstract: Replace "important" or "significant" with facts or examples.
• Use strong verbs: Reduce nominalizations and empty phrasing.
• Prefer simple syntax: Avoid stacked clauses and overlong sentences.
• Earn emphasis: Show importance with evidence, not adjectives.
• Keep rhythm: Mix sentence lengths; let short sentences land.
• Maintain tone discipline: No slang, no hype, no boilerplate cheerleading.

Editing Checks (Apply in Order)

1. Meaning: Are claims precise, supported, and non-contradictory?
2. Structure: Does each paragraph have a point and support it?
3. Brevity: Cut filler, hedging, and throat-clearing.
4. Style: Replace puffed-up language with plain English.
5. Mechanics: Fix grammar, punctuation, and formatting.

Common Fixes

• Vague openings: Replace generic framing with the specific purpose.
• Abstract nouns: Swap "implementation of improvements" for "we improved X."
• Weasel words: Replace "experts say" with named sources or remove.
• Empty intensifiers: Remove "very," "extremely," "highly," unless essential.
• Over-polite filler: Remove "I hope this finds you well" unless expected.
• AI-ish phrasing: Avoid "pivotal," "tapestry," "underscoring," "delve."

Output Format

• Provide the revised text.
• If changes are substantial, add a brief, 3-5 bullet change summary.

References

Use these references for deeper guidance when needed:

• references/humanize-guardrails.md - LLM failure patterns and rewrite guardrails.
• references/tropes.md - local copy of tropes.fyi's catalog of common AI writing tropes to avoid.