Llm Writing
Polish human-facing copy—docs, posts, specs—by resisting default LLM tone, structure, and conversational bleed before you ship or publish.
Overview
llm-writing is a journey-wide agent skill that steers agents away from default LLM writing habits—usable whenever a solo builder drafts human-facing prose before publishing or shipping.
Install
npx skills add https://github.com/haowjy/creative-writing-skills --skill llm-writingWhat is this skill?
- Surfaces default LLM pulls: label summaries, conclusion-without-evidence, negation definitions
- Flags conversational bleed (“let’s break this down”) in standalone documents
- Explicitly loads intent-modeling when not already in context
- Behavioral checklist—not a blanket ban on smooth transitions when purposeful
- Journey-wide: any human-readable artifact, not code-only review
Adoption & trust: 1 installs on skills.sh; 241 GitHub stars; 3/3 security scanners passed (skills.sh audits); trending (+100% hot-view momentum).
What problem does it solve?
Your agent output reads like a chat transcript or generic AI summary, so readers lose trust even when the underlying ideas are sound.
Who is it for?
Founders shipping READMEs, changelogs, landing copy, and support articles who want quick pattern recognition without a full editorial team.
Skip if: Pure code generation, data pipelines, or cases where you already locked voice in an approved style guide and only need mechanical formatting.
When should I use this skill?
Use when producing written artifacts for humans and you need to catch default LLM writing patterns and conversational bleed.
What do I get? / Deliverables
You get prose written for the document’s reader—with purpose anchored, evidence where needed, and chatty corrections removed—after pairing with intent-modeling when required.
- Reader-oriented revised prose
- Notes on resisted default patterns
- Alignment with stated intent after intent-modeling
Recommended Skills
Journey fit
Useful at every journey phase - explore requirements and options before committing to a direction.
Where it fits
Rewrite waitlist copy that currently corrects questions no visitor asked.
Turn API reference sections from label lists into mechanism explanations.
Edit programmatic SEO pages so intros are not redundant restatements.
Polish lifecycle emails so they address subscribers, not the prompting session.
How it compares
Use as a writing-quality lens on artifacts; pair with intent-modeling for goals, not as a substitute for factual research skills like academic paper search.
Common Questions / FAQ
Who is llm-writing for?
Solo and indie builders who delegate drafts to agents but need landing pages, docs, and emails to read like professional documents—not threaded replies.
When should I use llm-writing?
During Build docs, Validate landing copy, Launch SEO or distribution posts, Grow lifecycle content, and Operate customer-facing updates whenever the output is meant for an offline reader.
Is llm-writing safe to install?
It is prose guidance without shell or network calls in SKILL.md; still review the Security Audits panel on this Prism page before adding skills from unfamiliar repos.
Workflow Chain
Requires first: intent modeling
SKILL.md
READMESKILL.md - Llm Writing
# LLM Writing Load `/intent-modeling` if it isn't already loaded. Recognizing default LLM writing behaviors is most of the battle — once you notice the pull, you can judge whether to resist it for this particular piece. ## Behavioral Pulls Filling structure without anchoring to purpose. Summarizing with labels instead of explaining how things work. Stating conclusions without evidence. Smoothing over uncertainty. Encoding corrections as prohibitions. Defining by negation. Restating what was just said as a transition. Not always wrong — the failure is when they happen by default. ## Conversational Bleed Writing as if responding to a user when producing a document for a reader who wasn't in the conversation. "It's not X — it's Y" corrects a misconception nobody has. "Let's break this down" addresses a question nobody asked. Write for the reader. They have no context from the conversation that prompted the document.