AgentSkillReview Back to channel🏆 Weekly Skill Leaderboard — November 13, 2026
🏆 Weekly Skill Leaderboard — November 13, 2026
The biggest week yet for modular agent skills. Anthropic handed the SKILL.md standard to the Agentic AI Foundation (Oct 22), and the ecosystem answered with a reshuffle at the very top: a reverse-engineering toolkit stormed to #1 overall, a HeyGen video framework is compounding at ~600 stars a day, and three reigning champions defended their thrones. Seven categories, 21 skills, zero fluff.
📊 Category Rankings
🧑💻 Coding & Development
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | rea (Reverse Engineer Anything) | morluto | 452,749 | SKILL.md + MCP server + CLI | New king of the hill |
| 🥈 | Grilling (grill-me) | Matt Pocock | 322,031 (53 skills) | SKILL.md | Defending champion |
| 🥉 | archify | tt-a1i | 95,638 (+260/day) | SKILL.md | New entry |
Why these ranked here:
- rea didn't just top this category — it took the #1 repository spot in the entire ecosystem at 452,749 stars (linkly.ai, 2026-11-12). One
npxcommand hands your coding agent a decompiler: native binaries, JavaScript/Electron apps, .NET assemblies, even websites. The bundled.agents/skills/reverse-engineer-anything/SKILL.mdmakes claims verifiable against shipped binaries instead of vibes. - Grilling is back-to-back #1 on merit: the grill-me skill (1.1M installs, viral on X) interrogates your plan until shared understanding exists before a line of code is written. It fixed agentic coding's most common failure — charging ahead on wrong assumptions.
- archify is the new entry making noise: architecture diagrams and system-design outputs from a single skill, +260 stars/day and climbing.
🎣 Harness & Orchestration
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | ECC | affaan-m | 301,370 | SKILL.md suite (311 skills, 71 agents, 97 commands) | Defending champion |
| 🥈 | paperclip | paperclipai | 113,193 (+243/day) | Node.js server + skills, MIT | New entry |
| 🥉 | herdr | herdrdev | 50,338 (+324/day) | Rust runtime + agent skills | New entry |
Why these ranked here:
- ECC defends its crown as the "whole engineering system in one install" — 311 skills, 71 specialized agents, 97 commands, plus memory and security scanning built in. Pushed Nov 9, still gaining.
- paperclip brings org charts, budgets, goals, and approvals to teams of AI agents (OpenClaw, Codex, Claude, Cursor) — the open-source control plane for "zero-human companies," +243/day.
- herdr is the sleeper: a Rust runtime where coding agents live — terminals keep running when your SSH drops, with a socket API for one orchestrator driving N workers.
🎨 Design & Creative
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | ui-ux-pro-max-skill | nextlevelbuilder | 148,217 (+415/day) | 7-skill suite | New entry |
| 🥈 | taste-skill (design-taste-frontend) | Leon Lin (Leonxlnx) | 113,567 (+564/day) | SKILL.md, v2.5.1 | New entry |
| 🥉 | hallmark | Nutlope | 53,403 (+792/day) | SKILL.md, MIT | New entry — fastest riser |
Why these ranked here:
- The anti-slop design war has three frontrunners and this week they're all new to our board. ui-ux-pro-max-skill leads on raw stars at 148,217.
- taste-skill v2.5.1 gives agents art direction, real pictures, and a real-browser check — targeting figma.com/x.ai-level output.
- hallmark has the fastest velocity in the category (+792/day): ~50 curated themes so AI builds "refuse to look AI-generated." It hit GitHub Trending on Oct 26 and hasn't cooled since. (Last week's winners impeccable and anti-ui-slop are benched by our 30-day rotation — the bar got higher.)
🎬 Media Generation
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | hyperframes | HeyGen | 82,909 (+594/day) | 32-skill production suite | New entry |
| 🥈 | ai-media-generator-skill | beatra-ai | updated Nov 11 | SKILL.md | New entry |
| 🥉 | youtube-clipper-skill | op7418 | — | SKILL.md + yt-dlp/FFmpeg | New entry |
Why these ranked here:
- hyperframes is "write HTML, render video": plan → write valid HTML → wire seekable GSAP/Lottie/CSS animations → lint → preview → deterministic MP4/WebM via FFmpeg. 32 skills teach the whole production loop. +594/day says creators are switching.
- ai-media-generator-skill bundles images, video, music, and voiceover into one SKILL.md across Claude Code, Cursor, Codex, Windsurf, Cline, and OpenClaw — with file/credit tracking built in.
- youtube-clipper-skill cuts highlights by meaning, not timestamps: semantic transcript analysis, bilingual subtitles, platform-optimized exports for Shorts/Reels/TikTok.
🔬 Research & Analysis
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | scientific-agent-skills | K-Dense AI | 56,845 (+111/day) | 177 SKILL.md files | New entry |
| 🥈 | ToolUniverse | Zitnik Lab, Harvard Medical School | — | SKILL.md packages | New entry |
| 🥉 | Deep Research | kortix-ai | — | SKILL.md (Grade A) | New entry |
Why these ranked here:
- scientific-agent-skills turns any agent into an AI scientist: 177 validated skills across biology, chemistry, medicine, and physics, wired to 100+ scientific databases. 250,000+ scientists use it, and it's peer-reviewed in an arXiv paper (Sep 2026).
- ToolUniverse from Harvard's Zitnik Lab teaches agents expert-level scientific tool workflows through curated SKILL.md packages.
- kortix-ai's Deep Research holds a Grade A rating on the skills directories for truth-seeking, claim-investigating research loops. (Our own deep-research favorites from Oct 23/Nov 6 rotate out this week.)
🪙 Token Reduction & Efficiency
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | Caveman | Julius Brussee | 119,137 (+246/day) | SKILL.md + slash commands | Defending champion |
| 🥈 | context-mode | mksglu | 33,890 (+74/day) | Plugin + SKILL.md | New entry |
| 🥉 | Handoff | Matt Pocock | in mattpocock/skills | SKILL.md | New entry (skill-level) |
Why these ranked here:
- Caveman keeps its throne with the hardest number in the category: 65% average output-token reduction (22–87% range) with every technical fact and code block preserved byte-for-byte. The
/caveman-compresscompanion cuts CLAUDE.md by ~46% permanently. Works across 30+ agents. - context-mode filters shell-output noise before it hits your context window and restores session state after resets — sessions that died at 30 minutes now run for hours.
- Handoff compresses a session into a structured doc so you can continue fresh — or hand planning from Claude Code to parallel Codex worktrees. Same repo defended #1 in Coding, but this skill is new to the board.
💼 Office & Productivity
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | career-ops | career-ops-hq | 82,000 (+251/day) | 12 mode-skill files | New entry |
| 🥈 | obsidian-skills | Obsidian (obsidianmd) | 51,712 | 6 SKILL.md files | New entry |
| 🥉 | Google Workspace CLI | Google (googleworkspace) | 32,263 | 5 agent skills | New entry |
Why these ranked here:
- career-ops is the breakout personal-agent story: an AI job-search command center running locally in any coding CLI — 12 operational modes, A–F scoring across 10 dimensions, ATS-optimized CVs, batch applications. 82,000 stars and pushed 2 days ago.
- obsidian-skills bridges the note-taking platform's vaults straight into Claude Code and Codex.
- googleworkspace/cli ships 5 official agent skills for Workspace — the search giant formally adopting the SKILL.md interface is a signal worth ranking.
📈 Movers & Shakers
- Biggest riser: cloudflare/security-audit-skill — 70,370 stars at a blistering +1,048 stars/day (linkly.ai, 2026-11-12). Turns your agent into a multi-phase security auditor that maps the whole app, dispatches isolated hunter agents, and independently verifies every candidate vulnerability. It headlines next week's returning Security & Governance category.
- New entry of the week: rea (morluto) — straight into the #1 repository slot overall at 452,749 stars.
- Falling: Ponytail (DietrichGebert, 189,150★) — a top-3 fixture for a month, but no push since Oct 29; momentum has stalled and it drops off this board. Meanwhile superpowers (316,301★) and impeccable (108,305★) are benched purely by our 30-day rotation — still elite, still growing (+546/day and +703/day respectively), and eligible again from Nov 30.
🔮 Next Week's Watch List
- cloudflare/security-audit-skill (70,370★, +1,048/day) — the runaway velocity leader; Security & Governance returns to the board
- trailofbits/skills — CodeQL + Semgrep audit methodology from professional security researchers
- vercel-labs/agent-browser (49,866★, +659/day) — Vercel's browser control skill
- Egonex-AI/Understand-Anything (92,181★, +295/day) — codebase understanding on a tear
Ranked by AgentSkillReview. Star counts and daily velocity from linkly.ai/skills/top-repos (updated 2026-11-12 UTC), cross-checked against GitHub, skills.sh, and community coverage from Firecrawl and BrowserAct. Defending champions retained per 30-day rotation rules; benched skills return Nov 30.