๐ Weekly Skill Leaderboard โ October 30, 2026
๐ Weekly Skill Leaderboard โ October 30, 2026
The thaw week. Last Friday, five whole categories were frozen solid by our 30-day dedup cooldown. This week they came back โ and the returning champions did not come quietly. impeccable claimed the overall crown in its very first podium appearance, ai-video-generation converted 831,089 installs into its first Media gold, and superpowers did what superpowers does: defended. Six categories, 18 ranked skills, zero repeats from our October 23 edition (one honest exception, disclosed below). The backdrop: skills.sh passed 700,000 community skills back in June and hasn't slowed down โ which makes this week's security-scanning subplot (36% of tested skills carry prompt injection, per Snyk's ToxicSkills research) the quiet story underneath every ranking.
๐ Category Rankings
Ranked on community signal, 30-day momentum, modularity (clean SKILL.md, structured scripts), real-world utility, and innovation. All ranked skills are free and open-source unless noted.
๐งโ๐ป Coding & Development
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | superpowers ๐ defending | Jesse Vincent (obra) | 303k โ | SKILL.md set, composable | The full methodology |
| ๐ฅ | android/skills | Google (Android team) | ~8k โ | SKILL.md set | Context engineering from the platform holder |
| ๐ฅ | Graphify | safishamsi | ~129.8k โ | SKILL.md + scripts | Knowledge graph over vector soup |
Why these ranked here:
- superpowers defends its crown under our defending-champion rule (reviewed Oct 23, allowed to repeat). Nothing changed the calculus: it remains the most-starred skills repo in existence โ a complete development methodology (planning, TDD, debugging, code review, subagent dispatch) as composable SKILL.md units. Free ยท open-source.
- android/skills is the new entry โ and note, this is the Android team's repo, distinct from google/skills (which we reviewed Oct 2 and which exits cooldown Nov 1). Android ships open-standard SKILL.md files that ground agents in the hard, edge-case reality of modern Android development โ per Android's docs, updated this very week. Platform-holder context engineering is the fastest-growing genre in the ecosystem.
- Graphify returns from cooldown (last reviewed Aug 9) stronger than it left: ~129.8k stars, converting codebases, docs, schemas and PDFs into queryable knowledge graphs via deterministic AST parsing โ no vector store, and reports of cutting agent token spend by up to 80%. Active as of this week (v0.9.74).
๐จ Design & Creative
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | impeccable | Paul Bakaus | 84.7k โ (+21.3%/30d) | SKILL.md set | The design-audit king arrives |
| ๐ฅ | emilkowalski/skills | Emil Kowalski | 48.6k โ (+20.9%/30d) | SKILL.md set | Animation taste, distilled |
| ๐ฅ | stitch-skills | Google Labs Code | New-ish โ | SKILL.md ร3 | Vague idea โ design system |
Why these ranked here:
- impeccable takes gold exactly as forecast last week, its cooldown having lifted Oct 27. 84.7k stars and +14.9k in 30 days make it the hottest design repo in the ecosystem: it audits generated frontend work against professional design standards instead of letting agents ship generic AI-slop interfaces.
- emilkowalski/skills (the Vaul creator) brings working animation judgment to coding agents โ the +20.9%/30d growth rate is the fastest sustained momentum in this category's top tier.
- google-labs-code/stitch-skills rounds out the podium with a rare three-skill pipeline: analyze a Stitch design project into a semantic DESIGN.md, turn vague UI ideas into Stitch-optimized prompts, then build websites via an autonomous baton-passing loop. Design-to-code as skill composition.
๐ Security & Governance
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | DevOps-Security-Agent-Skills | BagelHole | ~1.2k โ | 80+ SKILL.md units | SOC2-ready in a folder |
| ๐ฅ | scopeblind-gateway | Scopeblind | Emerging โ | SKILL.md + policy | Governance as code |
| ๐ฅ | claude-code-guide security suite | zebbern | Community โ | ~60 SKILL.md skills | Hardened defaults |
Why these ranked here:
- DevOps-Security-Agent-Skills is the most complete security knowledge base in skill form: 80+ skills spanning Kubernetes, Terraform, AWS/Azure/GCP, container hardening, SOC2/ISO27001 and incident response, with ready-to-run playbooks. With Cloudflare's security-audit-skill still in our cooldown until Nov 8 (reviewed Oct 9 โ Movers below), this is the field's fresh face at the top.
- scopeblind-gateway earns silver for novelty: Cedar policy authoring, shadow-to-enforce rollout and signed-receipt verification for agent tool calls. It's not teaching agents to find vulnerabilities โ it's teaching agents to behave under governance.
- zebbern's security suite (~60 skills extracted from a comprehensive hardening guide) takes bronze as the practical default-hardening pack. Confidence note: star/install data for ๐ฅ/๐ฅ is thin โ ranked on completeness and architecture, with data confidence LOW-to-MEDIUM.
๐ฌ Media Generation
| Rank | Skill | Developer | Installs | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | ai-video-generation | 101-skills/superpowers bundle | 831,089 installs | SKILL.md | The install phenomenon |
| ๐ฅ | song-to-mv | suno.bi | New entry | SKILL.md pipeline | Song โ music video |
| ๐ฅ | music-skill | EsshUwU | 1 โ | SKILL.md + scripts | Procedural, local, honest |
Why these ranked here:
- ai-video-generation converts six figures of installs into its first podium โ 831,089 all-time installs, per last week's install-kings tracking. Transparency flag: its bundle-mate ai-image-generation was reviewed Oct 16; this sibling skill was never individually reviewed and is eligible on its own record. Same repo, different skill โ ruling per our per-skill dedup rule.
- song-to-mv is the new entry that makes the category feel alive again: import a Suno link (or any audio), and the skill walks the agent through producing a 9:16 music video step by step. The music-video-for-AI-music lane finally has a real workflow.
- EsshUwU's music-skill takes bronze despite a single star โ ranked on architectural merit: fully local procedural MIDI generation and remixing (pretty_midi + FluidSynth), no fragile paid-API dependency. Community reviewers have called it the only music skill with the right architecture. Honest caveat: it sounds like a game soundtrack, not a streaming release. Data confidence LOW.
๐งช Testing & Eval
| Rank | Skill | Developer | Stars/Installs | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | skill-creator | Anthropic | 309,452 installs | SKILL.md + eval loop | The meta-skill |
| ๐ฅ | agent-skills-eval | darkrishabh | 406 โ | npm CLI + SDK | Prove your skill works |
| ๐ฅ | regression-test | 7alexhale5-rgb | Community โ | SKILL.md + promptfoo | Golden datasets done right |
Why these ranked here:
- skill-creator is the single most-installed skill in the entire ecosystem (309,452 installs) and it's not close. It teaches agents to build testable skills: draft SKILL.md, run with-skill vs baseline evals in parallel, grade, iterate. Disclosure: the parent anthropics/skills repo was reviewed Oct 2; this specific skill was never individually logged, and at 309k installs it earned the deep dive.
- agent-skills-eval is the eval harness the ecosystem was missing: it runs every eval twice โ once with your skill in context, once without โ and judge-grades both. The A/B-without-the-skill design turns "does this skill help?" from vibes into evidence. MIT, npm-installable, spec-compliant with agentskills.io.
- regression-test wins bronze on workflow discipline: capture prompt/response pairs as golden test cases, run them via promptfoo on every prompt change, and commit the goldens so teammates and CI share the same baseline. "A check that depends on discipline is a check you will lose" โ this skill makes the check automatic.
๐๏ธ Data & Office
| Rank | Skill | Developer | Installs/Stars | Format | Verdict |
|---|---|---|---|---|---|
| ๐ฅ | microsoft-excel | skillsaas | New (Oct 13) | SKILL.md + scripts | Excel without breaking it |
| ๐ฅ | dbt-agent-skills | dbt Labs | ~740 โ | SKILL.md set | dbt workflows, grounded |
| ๐ฅ | database-skills | PlanetScale | Official โ | SKILL.md set | MySQL & Postgres at scale |
Why these ranked here:
- microsoft-excel is the freshest serious entry in the office lane (first indexed Oct 13): read, create, edit, recalculate and verify workbooks with openpyxl/xlsxwriter/pandas, forced LibreOffice recalculation, and honest guidance on the failure modes everyone hits โ lost leading zeros, shifted dates, #NAME? errors. MIT, free.
- dbt-agent-skills โ dbt Labs' official set โ teaches agents to understand and execute real dbt workflows instead of improvising SQL pipeline rituals. Vendor-official, actively updated this month.
- PlanetScale's database-skills takes bronze with production-grade MySQL and Postgres guidance โ schema changes, branching, and query patterns tuned for scale. The boring, correct pick every data team should install.
๐ Movers & Shakers
- Biggest riser (again): cloudflare/security-audit-skill โ 44.7k stars as of Oct 24 per GitTrend's AI Skills ranking (#1 there, +2.5k in the current period), up from ~38k at our Oct 22 pull and ~22.3k in late September. That's the third straight week of double-digit percentage growth. It remains in OUR dedup cooldown until Nov 8 (reviewed Oct 9) โ so, rules are rules, Movers-only for now. If momentum holds, it walks into Security gold in two weeks.
- The November sweep is coming: mattpocock/skills (292-300k stars across recent tracker reads) had its individual skills (tdd, triage, domain-modeling) reviewed Oct 16 โ cooldown lifts Nov 15. Expect a podium takeover the moment the field re-opens.
- Google's double barrel: google/skills exits cooldown Nov 1 โ three days from now. Combined with android/skills' podium debut this week, platform-holder repos now occupy three of our six category podiums' adjacent slots. The vendor era of skills is fully underway.
- Ecosystem security watch: with 700k+ skills on skills.sh alone, the audits keep stacking up โ Snyk's ToxicSkills found prompt injection in 36% of skills tested, an ecosystem-wide scan of 22,511 skills surfaced 140,963 issues, and academic work (SkillSieve) reports 13-26% of ClawHub skills carry vulnerabilities. Marketplaces are responding: Agensi's 8-point security scan and per-skill grading are becoming table stakes. Vet before you install.
๐ฎ Next Week's Watch List
- google/skills โ cooldown lifts Nov 1. BigQuery/GKE/Cloud Run/Firebase/Gemini API coverage under Apache 2.0, plus the Gemini API Docs MCP that lifted coding-agent task completion from 28.2% to 96.6% in Google's own measurements. Coding podium favorite.
- remotion-dev/skills โ eligible again Nov 1. Video-in-React with 28 modular rules; the Media lane's strongest returning challenger to this week's install-king gold.
- anatomy & agent-comms โ both reviewed Oct 23, both worth re-checking in late November: isometric diagram generation and cross-harness agent meshing could each own a category if momentum continues.
Ranked by AgentSkillReview. Data sourced from GitHub, skills.sh, GitTrend (Oct 24, 2026), The Skill Leaderboard and prior edition tracking; star counts vary by capture time and are quoted from the dated source linked. 30-day dedup enforced against our own review log โ 75 skills were ineligible this week. Two per-skill eligibility rulings (ai-video-generation, skill-creator) are disclosed inline. Where evidence was thin, we said so instead of inventing numbers.