๐ Weekly Skill Leaderboard โ November 6, 2026
๐ Weekly Skill Leaderboard โ November 6, 2026
The week the standard went neutral, and the podium got crowded. On October 22, Anthropic donated the Agent Skills (SKILL.md) open standard to the Linux Foundation's Agentic AI Foundation โ the format itself unchanged, the governance now community-owned. Days later, Microsoft shipped Agent Skills for .NET and added builder skills for .NET and Python (Nov 4), and Apple's macOS Golden Gate 27.3 beta quietly gave Siri AI a SKILL.md developer framework. The ecosystem's numbers keep compounding: skills.sh has crossed 700,000 community skills with 2M+ CLI installs. This week we ranked 9 categories, 27 skills โ after screening out 69 entries we covered in the past 30 days. Every pick below cleared our dedup gate.
Method: ranking weighted by community signal (stars/installs) โ recency (commits/releases in the past 30 days) โ modularity (clean SKILL.md, scripts/references structure) โ real-world utility โ innovation. Install counts are cumulative npx skills add telemetry from skills.sh; we flag manipulated-looking figures where sources call them out.
๐ Category Rankings
๐งโ๐ป Coding & Development
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | Grilling (mattpocock/skills) | Matt Pocock | 149.1K installs; repo 255K stars, +2,207/day | SKILL.md + references | The interview-you-until-it's-committed skill |
| ๐ฅ | vercel-react-best-practices | vercel-labs | 451.2K installs | SKILL.md + compiled AGENTS.md | 40+ rules, 8 categories, impact-ranked |
| ๐ฅ | typescript-expert / nodejs-best-practices (agentic-awesome-skills) | sickn33 | Repo 47,709โ / 6,491 forks, MIT, updated Jul 27 | SKILL.md + resources | The strongest independent coding bundle |
Why these ranked here. Pocock's repo stays the single largest independent skills library on GitHub (255K stars, gaining ~2,207 stars/day in the September trend data), and Grilling โ now 149.1K installs โ remains the archetype of the "workflow skill": the AI interrogates your vague plan until it becomes a committable decision. Vercel's react-best-practices holds silver on raw adoption (451.2K installs) and a rule library that compiles into a single AGENTS.md agents actually load. sickn33's agentic-awesome-skills repo (47.7Kโ , MIT, actively updated) wins bronze for shipping a coherent expert bundle โ typescript-expert, nodejs-best-practices, clean-code โ that installs cleanly across agents.
๐ฃ Harness & Orchestration
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | ECC | affaan-m | 251Kโ , +1,485/day | SKILL.md + harness config | Harness optimization: steadier, more auditable agents |
| ๐ฅ | Warden | Sentry (David Cramer) | Featured at Vercel Skills Night | SKILL.md + GitHub Actions | Skills run as linters on every pull request |
| ๐ฅ | orca | stablyai | Trending 24h (Sept data), actively maintained | SKILL.md + git worktrees | Parallel-benchmark your coding agents head-to-head |
Why these ranked here. ECC is the week's biggest story โ 251K stars and roughly +1,485 stars/day, built on the thesis that agent value now comes from stable, auditable harnesses rather than raw model IQ. Warden turns "skills" into a static-analysis layer: every PR gets skills-linted via GitHub Actions, which is the most production-ready governance pattern we've seen from a vendor. orca takes bronze for a genuinely useful ops trick โ running Claude Code, Codex, and OpenCode in isolated git worktrees on the same task and comparing outputs.
๐จ Design & Creative
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | frontend-design | Anthropic | 502K installs (top-3 all-time, skills.sh June data); 856K by Sept snapshot | SKILL.md | The taste-correction layer for agent UI |
| ๐ฅ | anti-ui-slop | uizze.com | 680K installs (Sept) | SKILL.md + curated DB (800K real screens) | Detects and fixes the purple-gradient AI-slop look |
| ๐ฅ | expo/skills | Expo (Evan Bacon) | Official vendor repo; live Expo Go demo | SKILL.md + scripts | Native iOS upgrades driven end-to-end by Claude Code |
Why these ranked here. The market's center of gravity moved from "agents write UI" to "agents write UI that isn't embarrassing." Anthropic's frontend-design leads all-time adoption (502K installs in the June skills.sh census, 856K by the September snapshot) and defines the anti-generic direction-setting approach. anti-ui-slop is the discipline's funniest and most useful shot-caller โ it names the slop patterns (purple-cyan mesh gradients, glassmorphism-everything, floating 3D geometry) and grounds judgments in 800K real-screen data points; note it's from the same uizze family whose sibling skills we reviewed Oct 9โ16, but this specific skill is fresh to our leaderboard. Expo's official skills take bronze after Evan Bacon's Skills Night demo showed Claude Code applying SwiftUI component upgrades and auto-fixing Expo Go crashes in production.
๐ฌ Media Generation
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | remotion-best-practices | Remotion (remotion-dev) | 150K installs in 8 weeks; #1 video skill; repo 25.3Kโ | SKILL.md + 28 rule files | Programmatic video: English โ correct React motion-graphics code |
| ๐ฅ | music-video-subtitle-generator | MiniMax-AI (MiniMax-H3/skills) | Official MiniMax repo, recently released | SKILL.md | Music-video prompt + typography pipeline with lyric sync discipline |
| ๐ฅ | tuzi-video-gen | tuziapi | 183 installs / 37โ , updated Jun 3 | SKILL.md + TypeScript scripts | Multi-model Veo/Sora/Kling/Seedance with ffmpeg segment chaining |
Why these ranked here. Remotion's skill is the crossover story of the season: 6M views on the launch demo, then 150K installs in eight weeks โ the #1 video skill and the top independent (non-platform-vendor) skill on skills.sh. Its 28 modular rule files load on demand, which is exactly how domain knowledge should be packaged. MiniMax's official music-video-subtitle-generator earns silver as the most rigorous take on lyric-typography sync we've seen from a model vendor shipping SKILL.md-native tooling. tuzi-video-gen takes bronze for honest engineering: one CLI, four video model families, and long-video mode that chains segments by feeding each clip's last frame as the next reference image.
๐ฌ Research & Analysis
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | rigorpilot-skills | lllllllama | 4 skills in the all-time top 100; ~450K installs niche | SKILL.md + scripts | Auditable deep-learning paper reproduction |
| ๐ฅ | deep-research (claude-deep-research-skill) | 199-biotechnologies | 802โ , 8,896 installs | SKILL.md + phased workflow | Claim-tracked reports with executive summary + bibliography |
| ๐ฅ | deep-research (agent-skills) | petekp | Featured in playbooks catalog | SKILL.md (9KB) + references | Confidence levels + source-quality tables as first-class output |
Why these ranked here. rigorpilot wins on innovation-per-user: a four-skill suite that turns deep-learning paper reproduction into auditable steps โ 450K installs for a niche that used to be a Ph.D. chore. 199-biotechnologies' deep-research takes silver on adoption (8,896 installs) with a planโgatherโverifyโreport structure that keeps every fact source-backed. petekp's deep-research earns bronze for specification discipline: explicit High/Medium/Low/Uncertain confidence levels, source-quality assessments, and conflicts reported rather than silently resolved. (We'd have ranked imbad0202's academic-research-skills โ 35Kโ โ but it was in our Oct 9 review cycle, so it sits out on dedup.)
๐ Office & Productivity
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | document-skills (docx / pdf / pptx / xlsx) | Anthropic | Powers claude.ai document features; repo metrics 64.7Kโ | SKILL.md + Python scripts | The production reference for office-format manipulation |
| ๐ฅ | office-productivity bundle | sickn33 (antigravity) | Part of 41.5Kโ repo; 868-skill library | SKILL.md workflows | 7-phase chain: docs โ spreadsheets โ decks โ conversion |
| ๐ฅ | Lark suite (lark-doc / lark-base / lark-imโฆ) | larksuite (ByteDance) | 40 of the all-time top-100 slots; ~657โ660K installs each | SKILL.md + MCP server | One business function = one skill, the ecosystem's quiet giant |
Why these ranked here. Anthropic's document quartet remains the reference implementation for office formats โ production-grade scripts Claude runs without loading them into context. sickn33's office-productivity bundle takes silver for the most complete LibreOffice+Microsoft dual-format workflow chain. Lark claims bronze โ and an asterisk: 27 domain-distributed skills (open.feishu.cn .well-known pattern) all sit within 0.3% of each other in install counts, which the September census reasonably flags as possible bulk-install inflation. We rank it on breadth and the 13-skill official CLI distribution, and note the number caveat openly. (skillsaas' microsoft-excel and the Google Workspace CLI both cleared our quality bar but fell to the 30-day dedup gate โ reviewed Oct 30 and Oct 9 respectively.)
๐ Security & Governance
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | clawsec | prompt-security | 1.1K installs, active | SKILL.md + scanner scripts | First scanner built for agent platforms themselves |
| ๐ฅ | QE Security Compliance | community (SkillMD catalog) | 736 views / 247 downloads | SKILL.md | OWASP Top 10 + SOC2/GDPR/HIPAA gate enforcement |
| ๐ฅ | Skill Security Audit | nyldn | Grade B on skillsdirectory, red-team tested | SKILL.md | Adversarial testing baked into the audit workflow |
Why these ranked here. This category mattered more than usual this week: Trail of Bits demonstrated bypasses of all three skills.sh audit badges (Gen/Socket/Snyk) back in June, Snyk's ToxicSkills audit found 36.8% of 3,984 scanned skills carried security defects with 76 confirmed malware samples, and skills.sh's own docs say they can't guarantee listed skills. Against that backdrop, clawsec is the most directly useful tool โ it scans the agent platform's own attack surface (dependency CVEs, SAST via Semgrep/Bandit, and agent-specific static hook inspection). QE Security Compliance takes silver for standards coverage across OWASP/SOC2/GDPR with severity-gated blocking rules. nyldn's red-team-tested audit skill takes bronze. (Trail of Bits' own skills โ the category's former #1 โ are deduped from our Oct 9 review.)
๐ช Token Reduction & Efficiency
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | caveman | community | Trending 24h (Sept census) | SKILL.md | 65% output-token cut: "why use many token when few token do trick" |
| ๐ฅ | context-compression | muratcankoylan (Agent-Skills-for-Context-Engineering) | tessl registry listing Oct 2, 2026 | SKILL.md + templates | Anchored summaries + probe-based validation |
| ๐ฅ | context-optimization | community (SkillMD catalog) | 249โ251 downloads | SKILL.md | Compaction / masking / KV-cache / partitioning playbook |
Why these ranked here. caveman is the category's joy โ a one-line-thesis skill that compresses agent output tokens by 65% and survives on pure utility. The context-engineering repo's compression skill takes silver for doing compression safely: anchored summary templates that preserve decisions and file state, validated by probe checks rather than vibes. context-optimization earns bronze as the most complete playbook โ its strategy table (compaction 50โ70%, masking 60โ80%, KV-cache 70%+ hits) has become the reference numbers people quote.
๐งช Testing & Eval
| Rank | Skill | Developer | Signal | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | webapp-testing | Anthropic | 113,149 installs | SKILL.md + Playwright scripts | Drive a live app: verify behavior, capture screenshots, inspect logs |
| ๐ฅ | review-loop | community | Trending 24h (Sept census) | SKILL.md | Critic sub-agent scores output; auto-revise until โฅ8/10 |
| ๐ฅ | ProofShot | community | Listed in 2026 must-have guides | SKILL.md + browser recording | Visual UI verification with recorded evidence |
Why these ranked here. webapp-testing remains the eval-adjacent workhorse โ 113K installs and the pattern every testing skill copies (agent drives the real app, not a mocked one). review-loop takes silver for institutionalizing the acceptance gate: a critic agent scores each output against a rubric and iteration continues until the default 8/10 threshold passes โ exactly the "agents stop declaring premature victory" fix the ecosystem needs. ProofShot takes bronze for evidence-first UI verification with browser recordings as proof of behavior. (Anthropic's skill-creator โ the meta-eval benchmark at 309K installs โ and darkrishabh's agent-skills-eval both cleared our October review cycles and are deduped out.)
๐ Movers & Shakers
- ๐ Biggest riser: mattpocock/skills โ 255K stars with ~2,207 stars/day at the September trend peak, and grill-me now sitting at ~1.065M installs as the all-time #2. A personal repo outmuscling every official vendor collection remains the ecosystem's best proof that workflow skills beat framework skills.
- ๐ New entry of the week: Microsoft Agent Framework skills for .NET + Python (Nov 4) and macOS Golden Gate 27.3's Siri SKILL.md framework (Nov 5 beta) โ two platform giants adopting SKILL.md-native distribution in the same week the standard landed in the Linux Foundation. Distribution is no longer a question; it's a race.
- ๐ Biggest drop / caution flag: genmedia-labs RunComfy cluster โ the September census documented the same 38-skill set published under at least 4 different GitHub organizations, synchronized install counts, and an unanswered moderation Issue on vercel-labs/skills. Trending-dominant, but treat it as a paid-API customer funnel, not an organic community skill.
๐ฎ Next Week's Watch List
- wan-3-0-prime-reference-to-video / seedance-2-5-image-to-video โ occupying the Hot 1โ3 slots as new video models drop; same paid-API wrapper caveat applies.
- KimYx0207/agent-teams-playbook โ 6-phase adaptive multi-agent orchestration for Claude Code Agent Teams; small (96โ ) but the cleanest write-up of the "parallel external brains + summarize back" pattern we've seen.
- neondatabase/agent-skills & prisma/skills โ the DB-vendor skills wave (official Postgres/ORM knowledge as SKILL.md) keeps accelerating; expect a dedicated vendor-skills category soon.
Ranked by AgentSkillReview. Data sourced from skills.sh census data (Sept 2026 snapshot via mosfet.hatenablog.com), aiagentaddons.com, claudeskills.info, cultofclaude.com, skillselion.com, playbooks.com, Vercel Skills Night coverage, TechTimes (Nov 5), Microsoft devblogs (Nov 4), claude-world.com/powerdrill.ai (AAIF donation), cnblogs/GitCode trend reporting, and OWASP Agentic Skills Top 10 + Snyk ToxicSkills security audits. Dedup enforced against 69 skills reviewed Oct 7โNov 6 in our SQL registry.
Install counts = cumulative npx skills add telemetry (per-execution, not unique users), per skills.sh's own disclosure. Read them as momentum, not census.