agentleFS
fire17fire17/awesome-readme/SKILL.md

awesome-readme

Skill0 starsChanged 4 months agoMITView on GitHub

Transform any repo's README into an instant-star page — the exact playbook that produced github.com/fire17/fable-masterclass's README (the canonical exemplar and the minimum bar). Ground-truths the repo first, then builds the full battery — hand-crafted theme-safe SVG banner, live shields.io badges (CI/release/license/social + custom value badges), a "the part that should stop you" killer hook section, mermaid diagrams (styled, native-rendering), a ≤30-second quickstart, linked feature/law tables, collapsible depth sections, an honest step-by-step making-of/provenance section, a safety/undo table, a star CTA framed in the project's own voice + star-history chart, GitHub alerts, centered HTML layout — and then VERIFIES everything live: every badge URL resolves, the banner serves as image/svg+xml, every anchor is checked against GitHub's ACTUAL rendered ids (never guessed — emoji anchors keep invisible variation selectors), every relative link resolves, every number in the README is observed not invented, CI green after push. Use when the user types /awesome-readme or /awr, asks to "make the README beautiful/rich/appealing", "make this repo an instant star", "polish the repo page", "add banners and badges", "make the README at least as good as fable-masterclass", or when shipping a repo whose README is still plain. Takes an optional repo path (defaults to cwd's repo).

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. fralapo/awesome-agent-skills★ 1same name
    2. shbernal/awesome-readme★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Honesty is the aesthetic: Every number observed, every claim receipted, defects-caught-by-process listed proudly.

    line 11
  2. Very specific:

    Read the current README, the main artifact(s), CHANGELOG, CI workflow + latest run state, release/tags, install path, LICENSE.

    line 16
  3. Distinctive:

    If this is someone's repo you don't own or the working tree is dirty: stop and confirm before touching.

    line 19
  4. Very specific:

    ONLY system font stacks (Georgia, 'Times New Roman', serif / Verdana, Geneva, sans-serif) — GitHub's CSP blocks webfonts in SVG.

    line 45
  5. Distinctive:

    No scripts, no external refs inside the SVG (CSP strips them).

    line 47
  6. Distinctive:

    Run ALL of these after push.

    line 74
  7. Distinctive:

    Update-run discipline: git fetch first.

    line 94
  8. Very specific:

    If the repo lives in a registered workspace (e.g.

    line 95

≈ “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

🌟 /awesome-readme — the instant-star README playbook

Make the target repo's README at least as rich as the canonical exemplar — ~/Creations/Fable-Masterclass/README.md (live: https://github.com/fire17/fable-masterclass) — and verify every element against the real rendered page. This skill encodes the exact process that built it, including the defects that process caught. Two iron rules above everything:

  1. Honesty is the aesthetic. Every number observed, every claim receipted, defects-caught-by-process listed proudly. A beautiful README that oversells earns a takedown comment with receipts; one that shows its verification earns the star.
  2. Never guess what you can fetch. Anchors, badge URLs, rendered output — GitHub's reality is one curl away (the founding war story: an emoji heading's anchor id keeps the invisible U+FE0F variation selector — id="user-content-️-the-16-laws" — a guessed link 404s; the fetched one works).

Phase 0 — Ground truth the repo (before writing a word)

  • Read the current README, the main artifact(s), CHANGELOG, CI workflow + latest run state, release/tags, install path, LICENSE. gh repo view <repo> --json description,visibility,stargazerCount.
  • Identify: the audience (who lands here?), the 30-second win (what can they run immediately?), and — most important — the "part that should stop you": the one genuinely remarkable, TRUE fact about this project. Every great README has exactly one; find it in the repo's reality, never invent it. (Exemplar's: "a frontier model distilled itself into a markdown file smaller models can load.")
  • Inventory what exists to show: real badges available (CI? release? coverage?), real numbers (tests passing, refs verified, benchmark results — only ones you can observe), real receipts (gates passed, defects caught).
  • If this is someone's repo you don't own or the working tree is dirty: stop and confirm before touching.

Phase 1 — The completeness checklist (miss nothing)

A README leaving this skill has ALL of these, top to bottom (each maps to an exemplar section — copy the structure, never the words):

  1. Hero banner — custom SVG in assets/banner.svg, <img width="100%"> inside <div align="center">.
  2. Badge row — ≥6 live badges (see Phase 2 catalog), social-stars badge last.
  3. Epigraph + nav row — one italic line that captures the soul; then **[📖 X](#…)** · **[⚡ Y](#…)** quick-nav (anchors verified in Phase 4).
  4. The killer section — "the part that should stop you": the remarkable-true-fact, unpacked in 3–5 bullets that each carry a receipt, closed with a > [!IMPORTANT] one-line pitch.
  5. ≥1 mermaid diagram — the system/flow/distillation story, styled nodes (style X fill:#1a1030,stroke:#e8b84a,color:#f5d67b reads well in both themes).
  6. Quickstart — the ≤30-second path: one copy-paste block, then "then type X / open Y". Nothing before it that a hurried reader must scroll past.
  7. The core table — the project's enumerable heart (laws/features/commands/API) as a table whose rows LINK to exact doc anchors.
  8. Collapsible depth — <details><summary><b>…</b></summary> for the full curriculum/config/advanced matter; the page stays scannable.
  9. The making-of / provenance section — how this was actually built, step by step, with a mermaid pipeline and an honest tool/skill inventory table. Include "defects caught by the process" — it reads as trust, because it is.
  10. Safety/undo table — what install touches, what it never clobbers, the one-line uninstall, escape hatches.
  11. Trust section — how claims are enforced (CI proving references/tests/installs on every push), not just asserted.
  12. Star CTA — framed in the project's own voice/values (never "please star"), + star-history chart: [![Star History Chart](https://api.star-history.com/svg?repos=<owner>/<repo>&type=Date)](https://star-history.com/#<owner>/<repo>&Date).
  13. Cross-links to sibling projects; License line; a closing <sub><i>signature line</i></sub> centered.

Project-type extras (add when they fit): terminal-tool → asciinema/gif demo above the fold; app/UI → screenshots in a 2-col table (light+dark); library → API table + benchmark table (observed numbers only); dataset/model → schema + eval table.

Phase 2 — Assets that always render

SVG banner rules (hand-craft; never external images):

  • ~1280×340 viewBox; dark gradient bg (#0b0f1a→#101a2e→#1a1030 family) reads on both GitHub themes; gold/blue accent gradients; role="img" + aria-label.
  • ONLY system font stacks (Georgia, 'Times New Roman', serif / Verdana, Geneva, sans-serif) — GitHub's CSP blocks webfonts in SVG.
  • A geometric motif drawn from the project's soul (the exemplar: dashed concentric circles = "the circle IS the craft"). Title, italic subtitle, a stats line (16 CHAPTERS · 16 LAWS · …), a thin accent rule.
  • No scripts, no external refs inside the SVG (CSP strips them).

Badge catalog (shields.io + native; use every one that's REAL, skip any that isn't):

CI:       https://github.com/<o>/<r>/actions/workflows/ci.yml/badge.svg   (link to the workflow page)
Release:  https://img.shields.io/github/v/release/<o>/<r>?color=e8b84a
Custom:   https://img.shields.io/badge/<label>-<value>-<color>            (URL-encode; %2F for /, %20 or -- rules)
License:  https://img.shields.io/badge/license-<spec>-blue
Stars:    https://img.shields.io/github/stars/<o>/<r>?style=social

Custom badges carry the project's real receipts: references-21%2F21 machine--verified, author-Claude%20Fable%205, loads on-Haiku · Sonnet · Opus. Each badge links somewhere meaningful (the verifier script, the announcement, the section).

Mermaid: flowchart LR for architecture/flows, flowchart TD for timelines/process; <br/> + <i> inside labels; quote labels containing parentheses (A["text (like this)"]) or mermaid fails to parse — a broken mermaid renders as an error block, so keep syntax minimal and check the pushed page.

GitHub-flavored tricks: > [!IMPORTANT]/[!NOTE]/[!WARNING] alerts; <div align="center"> for hero/CTA; <details> for depth; <sub> for the closing line; emoji as section wayfinding (one per heading, consistent register).

Phase 3 — Writing principles (the voice)

  • Lead with the outcome everywhere: section titles state the payoff, first sentences answer "so what".
  • Receipts over adjectives: "21/21 references machine-verified on every push" beats "high quality". If a stat can't be observed right now, it doesn't go in.
  • The insane-fact section is an argument, not hype: state the fact plainly, then unpack WHY it matters in bullets a skeptic can check. Write for the HN commenter with receipts.
  • The making-of is a feature: people star processes they wish they had. Name the tools honestly — including "none, by design" where true — and list the defects the process caught.
  • The CTA borrows the project's own philosophy (exemplar: stars framed as the book's own arena where claims fight). Generic begging repels; thematic invitation converts.
  • Preserve the project's existing soul: this skill restructures and enriches — it never rewrites the project's identity or clobbers hand-written sections without carrying their content forward.

Phase 4 — The verification battery (what "doesn't miss a thing" means)

Run ALL of these after push; each is observed, never assumed:

# 1. banner serves as SVG
curl -fsSL -o /dev/null -w "%{http_code} %{content_type}\n" https://raw.githubusercontent.com/<o>/<r>/main/assets/banner.svg   # want: 200 image/svg+xml
# 2. every badge URL resolves (loop the README's shields/actions URLs)
grep -o 'https://[^)"]*\(shields\.io\|badge\.svg\|star-history\)[^)"]*' README.md | while read u; do curl -fsSL -o /dev/null -w "%{http_code} $u\n" "$u"; done   # want: all 200
# 3. anchors verified against GitHub's RENDERED ids — never guessed
curl -fsSL https://github.com/<o>/<r> | grep -o 'id="user-content-[^"]*"' | sort -u > /tmp/ids
grep -o '](#[^)]*)' README.md | sed 's/](#//;s/)//' | while read a; do grep -q "user-content-$(printf %s "$a" | python3 -c 'import sys,urllib.parse;print(urllib.parse.unquote(sys.stdin.read()))')\"" /tmp/ids && echo "OK #$a" || echo "MISS #$a"; done
# 4. relative links resolve to real files
grep -o '](\([^)#h][^)]*\))' README.md | sed 's/](//;s/)//' | while read f; do [ -e "$f" ] && echo "OK $f" || echo "MISSING $f"; done
# 5. CI green on the README commit
gh run list --limit 1   # want: completed success

Plus two human-eye checks: open the pushed page once for mermaid render errors (a syntax slip renders as an error box) and dark/light banner legibility. And the honesty sweep: reread the final README hunting for any number or claim you did not personally observe this session — remove or verify each.

Phase 5 — Ship & bookkeeping

  • Update-run discipline: git fetch first; commit README + assets with a message that lists what was added; push; run the battery; fix and re-push until all green.
  • If the repo lives in a registered workspace (e.g. ~/Creations): add a dated changelog line to its registry entry and re-run the registry validator; if a skill was touched, obey the vault law (sync_skill.py).
  • Report with verification status attached: what was added, what the battery observed, anything NOT verified stated as exactly that.

Anti-patterns (each burned someone, most of them the exemplar's author)

  • Guessing GitHub anchors — emoji headings keep invisible variation selectors in their ids; fetch user-content-* ids and match exactly (use %EF%B8%8F-style encoding for invisible chars).
  • Badges that 404, point nowhere meaningful, or advertise stats that aren't real (a coverage badge with no coverage job).
  • Webfonts/external images inside the banner SVG — GitHub's CSP silently breaks them; system font stacks only.
  • Inventing numbers ("blazing fast", "99% tested") — every figure must be observed output.
  • Hype without receipts in the killer section — the HN failure mode; each bullet carries something checkable.
  • Unquoted mermaid labels with parentheses — renders as an error block on the live page.
  • Light-theme-only art — the banner must be readable on both GitHub themes (dark bg + high-contrast text wins on both).
  • Clobbering hand-written README content instead of carrying it forward (sacred-words rule applies to repos too).
  • Declaring done without the battery — "it looked fine in preview" is a claim, not a release.

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.