agentleFS
TrimVisTrimVis/skills/refactor-prose/SKILL.md

refactor-prose

Skill2 starsChanged 3 months agoMITView on GitHub

Iteratively refactor existing prose — privacy policies, marketing copy, web content, docs — based on user requirements or feedback. Reads surrounding pages for voice, asks clarifying questions before drafting, edits in place, and expects multi-turn revision. Use when the user says "update the privacy policy", "reword this page", "extend this copy with keywords X", "rewrite this section", or invokes /refactor-prose.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 9 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the target file in full.

    line 12
  2. Distinctive:

    Read 1–2 sibling pages (other policy pages, other marketing sections) to anchor on the existing voice, register, and structural conventions.

    line 13
  3. Very specific:

    Use AskUserQuestion with 1–3 questions max.

    line 26
  4. Very specific:

    Don't invent facts. If the rewrite needs a claim you can't verify from the file or the user's brief (a regulation citation, a stat, a …

    line 32
  5. Distinctive:

    If your change has knock-on effects elsewhere on the page (broken cross-reference, dangling pronoun), flag it briefly — don't silently fix …

    line 42
  6. Distinctive:

    Don't write a new file when the task is editing an existing one.

    line 47
  7. Distinctive:

    Don't invent legal, regulatory, or factual claims to make the prose sound complete.

    line 48
  8. Distinctive:

    Don't reformat or restructure unrelated sections "while you're in there".

    line 49
  9. Distinctive:

    Don't ask more than 3 clarifying questions in a single turn.

    line 50
  10. Distinctive:

    Don't add emojis unless the existing copy uses them.

    line 51
  11. Distinctive:

    Don't strip the user's voice and replace it with generic marketing register.

    line 52

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Refactor Prose

You're editing prose someone already wrote, not generating a blank page. The original exists for a reason — preserve the parts that are working.

Step 1 — Read before drafting

  • Read the target file in full.
  • Read 1–2 sibling pages (other policy pages, other marketing sections) to anchor on the existing voice, register, and structural conventions. Don't invent a new tone.
  • Note constraints visible in the page: required legal phrases, brand names, links, structural markers (h2/h3 hierarchy), CTAs.

Step 2 — Ask before writing

Most prose refactors fail because the AI guesses the user's intent. Before drafting, ask the small number of questions whose answers actually change the output. Examples that come up repeatedly:

  • Who is the audience now vs. before? (Sometimes the whole point of the rewrite.)
  • Is this a replacement of existing sections, an extension, or a reframing?
  • Are there sections to delete entirely, or keep verbatim?
  • For legal/policy text: which claims are facts to preserve, which are flexible?
  • For SEO/keyword work: must-include terms, target keyword density, banned terms.

Use AskUserQuestion with 1–3 questions max. Don't ask things you can infer from the file. If the user gave a complete brief upfront, skip this step — don't ask just to look thorough.

Step 3 — Draft

  • Edit in place with Edit. Prose lives in a real file with a real surrounding context; don't rewrite the whole document if only two sections change.
  • Match the surrounding voice. If the existing copy uses contractions, you use contractions. If it's formal, stay formal.
  • Don't invent facts. If the rewrite needs a claim you can't verify from the file or the user's brief (a regulation citation, a stat, a product capability) flag it inline with a clear marker like [VERIFY: ...] rather than fabricating.
  • For legal/policy text: don't add compliance language the user didn't ask for. "GDPR-compliant" is a claim, not a sentence opener.
  • Preserve structural markers (headings, anchors, frontmatter) unless explicitly changing them.

Step 4 — Iterate

Expect 2–4 revision rounds. The first draft is a proposal, not a delivery. When the user says "remove all newsletter references" or "extend with keywords X, Y, Z", treat it as a focused diff — don't take the opportunity to rewrite unrelated paragraphs.

When iterating:

  • Apply only what was asked. No drive-by polish of sections the user didn't mention.
  • If your change has knock-on effects elsewhere on the page (broken cross-reference, dangling pronoun), flag it briefly — don't silently fix it in a way the user has to hunt for.
  • Surface anything still flagged [VERIFY: ...] at the end so it doesn't ship by accident.

Don'ts

  • Don't write a new file when the task is editing an existing one.
  • Don't invent legal, regulatory, or factual claims to make the prose sound complete.
  • Don't reformat or restructure unrelated sections "while you're in there".
  • Don't ask more than 3 clarifying questions in a single turn.
  • Don't add emojis unless the existing copy uses them.
  • Don't strip the user's voice and replace it with generic marketing register.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.