readme-polish
Audit a GitHub README and rewrite it using modern 2025-2026 patterns — centered header, badges, hero image, GitHub alert callouts, emoji-prefixed features, expandable details, Mermaid diagrams, tables over dense prose. Produces a scannable README that works for a 10-second skim and a deep dive.
git clone --depth 1 https://github.com/yzhao062/anywhere-agents /tmp/readme-polish && cp -r /tmp/readme-polish/skills/readme-polish ~/.claude/skills/readme-polishSKILL.md
# Readme Polish
## Overview
Modern GitHub READMEs are **scannable first, readable second**. A skimmer should understand what the project does, who it is for, and how to install it in 10 seconds. Motivated readers get more detail from collapsibles, tables, and follow-up sections.
This skill takes an existing README (or a blank slate) and rewrites it using the visual and layout patterns that well-regarded 2025-2026 open-source projects have converged on. It does **not** invent the content; content comes from the project itself. It shapes how that content is presented.
## When to Use
- A README is a text wall with no visual anchors and no one can tell what the project does from a first glance.
- An OSS release is about to ship and the README has not been pass-edited for discoverability.
- The README mixes reference material with quickstart, so the install path is buried.
- Publishing badges, adding a hero image, or restructuring sections would measurably help adoption.
- A practitioner with social credibility wants the README to convey both "this is legit" and "here is how to use it" without competing for attention.
## When NOT to Use
- The README already follows the patterns below and the content is fine. Do not thrash for style.
- The project is internal / private / never published. No audience to optimize for.
- The project needs documentation at site-level scale (tutorials, API reference, cookbook). README polish only covers the repo root; use Sphinx / MkDocs / Docusaurus for real docs sites.
## Phase 1: Audit the current README
Before editing, classify what is there against the modern-README checklist. The goal is to identify which patterns are missing, which are misapplied, and which content should be moved, collapsed, or deleted.
Use [`references/checklist.md`](references/checklist.md) as the audit grid. For each row, mark present / absent / broken.
Key audit questions:
1. **First-paint (above-the-fold)**: is the project name, one-line tagline, badges, and install command visible before the reader has to scroll? If not, content above those is displacing them.
2. **Credentials placement**: is there a 100+ word maintainer bio blocking the first section? Move it into a `> [!NOTE]` callout or footnote.
3. **Why-section shape**: if the README has a "Why you'd use this" or scenario block, are the items narrative (cause / effect, before / after) or feature-shaped? Narrative content reads warmer as bold-lead paragraphs; feature claims read cleaner as emoji-prefixed bullets. Pick one pattern per section.
4. **Install path clarity**: can the reader find a single working install command in under 5 seconds? If not, there is probably too much pre-install framing.
5. **Reference material above-the-fold**: are limitations, related projects, detailed repo layout, maintenance policy visible before install? They should be collapsed.
6. **Visual anchors**: does the first screen have badges, a hero image, or a diagram? Without at least one, the README feels like documentation instead of a product page.
7. **Examples block (What This Looks Like)**: does the README show what the project LOOKS LIKE in practice (screenshot, file tree, themed diagram, before / after, mock)? If yes, do 3+ examples each use a different visual format, or are three monospace blocks stacked in a row? The latter reads as a text wall regardless of content quality.
8. **Themed Mermaid**: if Mermaid diagrams exist, do they use `%%{init: ...}%%` to match brand palette, or do they render with the default red/orange/blue and clash with the rest of the page?
9. **Heading case**: is Title Case applied consistently across all H2 and H3? Sentence-case headings mixed with Title Case headings read as AI-generated or unedited (agent-style RULE-G).
10. **Version-boundary honesty**: if a release ships primitives that are not yet wired, does the README name the boundary explicitly (what works today, what is queued), or does it present roadmap claims as shipped behavior?
11. **Anchor hygiene**: do dot-nav links resolve to real sections (GitHub's auto-generated anchor rules: lowercase, hyphens for spaces, strip punctuation)?
## Phase 2: Apply modern patterns
See [`references/patterns.md`](references/patterns.md) for the full catalog with copyable snippets. Summary of the highest-impact patterns:
### Hero / first-paint
- **Centered header** via `<div align="center">`. Title, one-line tagline, badge row, dot-separated nav, one-liner elevator pitch.
- **Shield.io badges** for package version (PyPI, npm), license, CI status, GitHub stars. Keep to 4-5; more than that becomes noise.
- **Dot-separated nav links** (`[Install](#install) · [Workflow](#workflow) · [Features](#features)`) below badges. Helps skimmers jump.
- **Hero image** — a PNG or SVG that makes a skimmer stop scrolling. Two paths:
- HTML mockup rendered via headless Chrome → PNG (good for feature grids, dashboards). See `ci-mockup-figure` skill for capture workflow.
- `<picture>` tag with light/dark variants for logos. Required only when the project has a logo/wordmark.
- **Maintainer credibility** in a `> [!NOTE]` callout, not a prose paragraph. Ideal length: 2–3 sentences with verifiable signals (package stars, citations, institutional affiliation).
### Content body
- **GitHub alert callouts** for emphasis: `> [!NOTE]`, `> [!TIP]`, `> [!WARNING]`, `> [!CAUTION]`, `> [!IMPORTANT]`. Each renders as a colored box with an icon. Do not overuse — one per major section at most.
- **Emoji-prefixed one-liner bullets** for "What you get" / "Features" when items are feature claims. Each bullet = 1 emoji + bolded feature name + one-line takeaway. 5–8 bullets is the sweet spot.
- **Bold-lead scenario paragraphs** for "Why you'd use this" when items are narrative (cause / effect, before / after). Pattern: `**Bold lead sentence.** Setup. Without X, problem. With X, fix.` Reads warmer than emoji-bullet lists when content is story-shaped.
- **"What This Looks Like" examples block** bReview loop for staged changes. Detects content type, prepares a review request for Codex (Terminal-relay manual default, opt-in Auto-terminal codex-exec subprocess, or IDE plugin), categorizes feedback, revises, and iterates. Works for code, papers, proposals, or any text-based output.
Create space-efficient figures for papers and proposals. HTML mockups for systems, dashboards, and timelines; TikZ or skia-canvas for abstract diagrams with arrow routing. Covers tool selection, design, capture, and LaTeX insertion. The primary goal is maximizing information per page — every figure must earn its space.
Context-aware router that detects work type and dispatches to the right skill. Ships with a minimal default routing table; extend it in your fork with your own skill entries.