35 Claude Code skills on GitHub: 9arm-skills, PaperSpine, book-to-skill, native-feel-skill, hallmark

Added: 2026-05-28

[00:00:00]Welcome back to GitHub Awesome. This is GitHub Trending Skills episode 3. 35 open-source skills trending on GitHub right now. Let's get into it.

[00:00:11]Nine arm skills is one engineer's personal Claude code skill library.

[00:00:16]A debugging discipline that forces you through reproduce, trace, falsify, cross-reference before touching any code. A postmortem writer that refuses to draft until you have a real repro and a validated fix.

[00:00:29]A scrutinized skill for outsider perspective PR review. And a management talk rewriter for when engineers need to brief leadership.

[00:00:37]Paper spine is a Claude code and Codex skill for rewriting academic papers properly. Not a prose polisher. It first locks in the paper's controlling motivation, studies the target journal or conference, learns rhetorical structure from strong exemplars, then builds section blueprints and a rewrite matrix before touching a single paragraph. Three Python stdlib scripts handle style metrics, revision auditing.

[00:01:04]You buy a 600-page technical book, read three chapters, then never open it again.

[00:01:10]Book to skill flips that. It's a Claude code skill that turns any technical book PDF into a Claude code skill of its own.

[00:01:16]Drop the PDF in, get back a structured queryable skill your agent can study, reference, and apply while you actually work.

[00:01:24]Your bookshelf becomes your AI's training material.

[00:01:28]Every electron app feels like an electron app. Raycast doesn't.

[00:01:33]Native feel skill distills exactly how Raycast pulls that off into an installable agent skill.

[00:01:39]Eight architectural tenets, a four-layer architecture pattern, a WebKit and WebView 2 survival guide, and a 75-item ship audit, all reverse-engineered from Raycast 2.0 and beta.app.

[00:01:51]Drop it into Claude code and build cross-platform desktop apps that don't betray themselves the moment someone touches them.

[00:01:59]Service Graph is a set of agent skills that give your AI a structured directory of over 100,000 US professional services firms: law, marketing, SEO, IT, consulting, accounting, and more.

[00:02:12]Filter by industry, location, size, rating, and third-party listing presence without scraping a single page.

[00:02:19]The four-tier funnel keeps discovery free and anonymous. Contact info costs quota only when you've actually shortlisted someone.

[00:02:27]PPT design prompt is a small Python CLI that converts web-first brand design.md files into presentation image-oriented design guides for PowerPoint, Keynote, and PDF decks. Point it at your local brand source files, run design md2ppt convert, and it writes reformatted outputs to PPT image/ with a conversion manifest. Not a deck generator, not a screenshot scraper, just a clean bridge between brand docs and image model inputs.

[00:02:55]Power Design is a Claude skill for generating HTML slide decks that don't look like AI made them. You give it a brand URL and it pulls the design DNA via Firecrawl, or you pick from 72 pre-built brand systems including Stripe, Linear, Apple, and Vercel. Every slide is then checked against 20 codified design principles from Tufte, Duarte, Refactoring UI, and WCAG. One idea per slide, 40% white space, 8-point grid, data-ink ratio above 80%.

[00:03:24]Louis Rossmann's team published the writing rules they used to keep Claude from producing AI slop. It's a portable Claude.md plus two skills: 24 hard anti-slop rules covering band words, filler phrases, dramatic headings, hollow it's not X it's Y constructions, and the M-dash problem, plus a voice profile built from corpus analysis of 513,000 words of Rossmann's own writing.

[00:03:50]Charlie Hill open-sourced the Claude skill system behind his 350k follower 100 million view per year content operation.

[00:03:59]17 skills total. Voice builder interviews you and writes your voice profile. Then every other skill reads it before drafting a word. Post writer, hook generator, post scorer against your actual history via Appify, real script reverse engineering, YouTube thumbnail generator, analytics dashboard, content matrix.

[00:04:20]Got a startup idea? Before you spend 3 months building it, run it through this.

[00:04:25]The Codex startup pressure test skill finds your core assumption, exposes fatal flaws, checks whether the problem is real, maps current behavior and real competitors, plans your first 10 customer moves, and defines a 2-week MVP test. One NPX install and it's in Codex.

[00:04:42]Run it brutal mode if you want the full hit.

[00:04:45]Google AI overviews and AI mode are changing what SEO actually means. And this agent skill packages Google's official guidance into something your AI agent can act on. Audit pages for AI search readiness, turn advice into a prioritized implementation plan, check crawlability, structured data, media SEO, e-commerce details, and get myth-busting on AEO hacks and LLMs.text.

[00:05:10]If you're building agent skills and want to actually measure whether they work, agent skills eval is the tool for that.

[00:05:16]It's a TypeScript SDK and CLI that loads agent skills.io style skill folders, runs them with and without the skill for a baseline comparison, grades outputs with a judge model, and generates a static HTML report with pass rates, token usage, timing, and assertion evidence.

[00:05:35]Any search is a unified real-time search skill for AI agents. General web search, 23 vertical domain searches, parallel batch search, and full page content extraction in one package.

[00:05:47]Works with Claude Code, Cursor, and WinSurf. The CLI runs on Python, Node.js, PowerShell, or Bash, and auto-detects your runtime on first run.

[00:05:56]Anonymous access works without an API key, though rate limits are tighter.

[00:06:01]AI-generated UIs all look the same. Hero section, three features, CTA, footer, repeat forever.

[00:06:07]Hallmark is an anti-AI slop design skill for Claude Code and Codex. Pick a macro structure, dress it in one of 22 themes, run 65 slop test gates plus a pre-emit self-critique before anything ships. Two pages built by Hallmark for two different briefs feel like different sites, not color swaps of the same template.

[00:06:28]Open Mobius is a full ICT and smart money concepts trading knowledge skill for Claude Code, Codex, Open Claw, and Hermes.

[00:06:36]Your agent now speaks the language of institutional trading, from market structure and displacement to precision entry models. If you're building quant tools on top of ICT or SMC methodology, this is the missing domain knowledge layer.

[00:06:51]Robotics Skill Suite is a massive curated collection of 76 audit-ready Claude skills for industrial robot, cobot, AMR, and ROS 2 life cycles. 38 specialized builders autonomously generate ISO 12100 risk assessments, ROS 2 architectures, and IEC 62443 cybersecurity segmentation into formatted Excel workbooks. 38 matching reviewers cross-check the math, enforce biomechanical limits for cobots, verify CL PLR ratings.

[00:07:21]You recorded 2 hours of podcast or stream footage, and now someone has to find the clips. Clipify makes that Claude Code's problem instead of yours.

[00:07:30]It's a Claude Code skill that automatically finds the funny or interesting moments in a long video, cuts them, reframes the footage to 9:16 with face tracking, and burns Opus-style captions.

[00:07:43]Drop in a video, get back social ready short clips. No manual scrubbing, no separate editing app.

[00:07:50]HTML and Canvas is the official Remotion repo demoing the new HTML and Canvas API using Chromium's experimental draw element image to capture DOM elements and composite them through custom WebGL shaders. Claude code drives the whole thing in the demo. If you build programmatic video with Remotion and want GPU powered visual effects on your HTML content, start here.

[00:08:14]Agent figure gallery is a drop-in scientific plotting skill for Claude code, Codex, cursor, and other agents that teaches them to produce publication quality figures by default. Correct axis labels, clean layouts, proper color choices, the kind of output you'd actually put in a paper or presentation.

[00:08:32]Drop it in once, every chart your agent generates gets better.

[00:08:37]Most anti-hallucination prompting is just telling the model to be careful.

[00:08:41]Buddhist method is a Claude code skill that encodes Buddhist epistemic principles as an engineering discipline.

[00:08:48]Direct observation over assumption, tracing errors to their actual root cause rather than treating symptoms, and staying steady under the pressure of a deadline or a panicking user.

[00:09:01]Android testing skills is a set of skills covering compose UI tests, Android instrumented tests, JVM unit tests, and ADB from Skydoves, the author of several well-known Android libraries.

[00:09:14]Drop them into Claude code or Codex and your agent writes tests that actually understand the Android testing stack. If you build Android apps with AI, this is the missing layer.

[00:09:25]Every academic journal has its own voice, structure, and unwritten conventions that reviewers notice immediately when violated. Journal adapt writing skill is a Claude skill that reads published papers from your target journal, extracts the writing convention section by section, then revises your manuscript to match.

[00:09:45]Actual journal specific adaptation, the kind that signals to reviewers that you understand the venue.

[00:09:52]You've got years of video footage scattered across multiple drives and no way to find anything in it. Frame Dex turns that archive into a queryable knowledge base. It's a Claude code skill that generates a sidecar markdown file next to each clip with GPS location, a speaker diarized multilingual transcript, English translation, face detection, and an AI scene description with a keep, review, or call rating.

[00:10:18]Working in automotive, autonomous vehicles, or embedded EV software?

[00:10:23]Automotive skill suite is the 150-plus skill counterpart to that robotic suite we covered. Getting vehicle software legally certified for the road is a nightmare of compliance and safety standards.

[00:10:35]This automates the entire life cycle.

[00:10:37]Installable Claude skills covering ISO 26262 functional safety, automotive spice.

[00:10:45]Science skills is a collection of highly optimized skills that radically speed up agentic scientific workflows. Plug it into your agent and it instantly gets heavily grounded programmatic access to over 30 massive scientific databases, Alpha genome, the AlphaFold database, UniProt. Your agent autonomously queries complex protein structures, or cross-references rare genetic mutations.

[00:11:10]Code X complexity optimizer is a skill focused on safe read-only code base complexity analysis.

[00:11:17]Install it and your AI assistant transforms into a ruthless senior staff level performance engineer. Hunts down cyclomatic complexity, identifies memory bottlenecks, flags security vulnerabilities, generates a structured optimization report telling you exactly where and why to refactor.

[00:11:34]Waits for your explicit approval before editing a single line.

[00:11:39]Tired of re-explaining your workflow and past bugs to AI agents every morning?

[00:11:44]Yantsu's skill is the cure for AI amnesia. Yantsu is a background AI that silently observes how you work on your desktop, conversations, tool choices, hesitations, decisions, crystallizes them into structured reusable tribal knowledge.

[00:11:58]This skill.md drop-in bridges Yantsu's memory back-end directly into Claude code or cursor.

[00:12:05]AI coding agents write clean code with zero compliance awareness. They'll log passwords and expose PII without hesitation. Compliance skills by Sprinto open sources their internal SOC 2, ISO 27001, and HIPAA knowledge as a Claude code skill.

[00:12:21]The PII detector fires autonomously in the background.

[00:12:25]Agent writes a GDPR violation, it intercepts and forces a rewrite before it ships.

[00:12:31]Mercury skills is an open library of skill.md playbooks for AI coding agents.

[00:12:36]Drop a skill into your agent's skills directory and it gains structured expertise on demand. 12 plus categories covering development, front, back-end, DevOps, AI and ML, security, product, marketing, design, and business. Over 100 skills total. Copy a single skill or clone the whole library.

[00:12:56]Two agent skills for testing distributed systems properly, not just running a few integration tests and shipping. One skill reads the code base, extracts the product's actual claims, generates generates hypotheses tied to those claims, and writes a structured test plan with a coverage adequacy argument.

[00:13:13]The second executes it. Fault injection, linearizability checking, 10 state verdicts, and blame classification down to SUT or environment.

[00:13:24]If you're tired of AI generated charts drowning in grid lines, rainbow color scales, and 3D bars that lie about the data, this is for you.

[00:13:33]Tufte Claude skill distills all three of Edward Tufte's books into 10 rules and auto applies them every time you ask Claude code to make a chart. Data ink ratio, lie factor, kill list, chart selection, a 12-item checklist. Nine before and after pairs in the repo show exactly what changes.

[00:13:52]This one's a research paper turned open-source framework. MM skills gives visual desktop agents reusable multimodal procedural knowledge. Each skill bundles a skill.md, runtime state cards, and visual keyframes. At inference time, the main agent keeps only lightweight skill hints in context, then opens a temporary branch when task state suggests a skill applies.

[00:14:15]Someone packaged Boris Cherney's Claude code pro tips talk into an agent skill.

[00:14:20]Tell it what you want Claude cursor or chat GPT to do, and it writes the prompt using five principles straight from that talk. Keep it short, lead with make a plan first, anchor the model in existing code rather than re-describing it, give it a feedback loop, and put persistent context in files not the prompt.

[00:14:41]PM brain is a Claude code skill that gives product managers a local second brain. No vector DB, no cloud, just plain markdown in a folder. Feed it interview transcripts, meeting notes, or Notion exports. It writes to knowledge, hypotheses, decisions, and stakeholders, and tags every claim with a provenance marker so you can see whether something is documented, verbal, or a hunch.

[00:15:06]This one's a bit different. A pastry chef with 18 years in professional kitchens built a Claude code sub agent for baking.

[00:15:13]Search a bilingual French and Chinese recipe library by name or ingredient, calculate baker's percentages, scale any formula to a target batch size and ask technique questions about ganache ratios, mousse stability, hydrocolloid selection, and more.

[00:15:32]>> [music]