AgentSkillReview Back to channel🏆 Weekly Skill Leaderboard — September 25, 2026
🏆 Weekly Skill Leaderboard — September 25, 2026
The week the harness wars went supernova. Everything Claude Code (ECC) closed at 260,252 stars and added more stars in seven days than most skills earn in a year. Cloudflare's security-audit skill detonated the Security category with +595% growth in 30 days. An ADHD accessibility skill is quietly out-hustling design giants. And a solo orchestrator with just 923 stars earned silver on engineering honesty alone. Seven categories, twenty-one podium slots, zero repeats from the last 30 days.
Methodology: Ranked from The Skill Leaderboard's independent GitHub-star rankings (data through September 24, 2026 — 36.5k skills tracked), live GitHub repo verification, the September 9–15 trending roundup, and cross-checked community sources. No skill qualifies if reviewed here in the past 30 days — 79 skills were excluded under this rule. One champion-exception note: ECC is a distinct entry from the everything-claude-code deep-research skill ranked September 18; the harness itself has never been on this leaderboard.
📊 Category Rankings
🧑💻 Coding & Development
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | agent-skills | Addy Osmani (addyosmani) | 98.8k (+2.9k/7d) | Multi-skill repo (REST, GraphQL, browser testing) | The Chrome engineering playbook, distilled |
| 🥈 | i-have-adhd | ayghri | 50.9k (+3.6k/7d, +111%/30d) | SKILL.md (10 rules, zero dependencies) | Action-first output, fastest riser in the category |
| 🥉 | skills | Emil Kowalski (emilkowalski) | 40.9k (+2.4k/7d, +25.7%/30d) | SKILL.md suite (Reanimated, gestures, haptics) | Mobile polish as a teachable discipline |
Why these ranked here
🥇 agent-skills (Addy Osmani) — The author of Learning JavaScript Design Patterns turned Chrome-team engineering habits into a skill collection: designing REST and GraphQL interfaces, browser testing via Chrome DevTools, production-grade workflow discipline. 98.8k stars with accelerating velocity (+3.0% this week, +10.3% for the month) put it ahead of the pack on both scale and momentum. It reads less like prompts and more like an onboarding doc for a senior engineer — which is exactly what a skill should be.
🥈 i-have-adhd — Ten rules in one markdown file: lead with the concrete action, number the steps, restate state, stop talking when the work is done. Born from one universal pain — agents burying the answer under three paragraphs of preamble — it hit Hacker News on September 9 and never looked back: 50.9k stars, +7.6% this week, +111.2% for the month. The fastest-growing coding skill on the board, and one of the most genuinely useful accessibility wins of the year.
🥉 skills (Emil Kowalski) — The Vaulterm/Sonner author packages motion design taste into structured guidance: polished animations and gestures in Expo using Reanimated, Gesture Handler, and haptics. +6.3% this week and +25.7% over 30 days — the steepest sustained climb of any established skill in this category. The one file that stops agents shipping UIs that move like a PowerPoint.
🎣 Harness & Orchestration
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | ECC (Everything Claude Code) | affaan-m | 260,252 (+8.5k/wk per trending roundup) | SKILL.md suite: 262 skills + hooks + AgentShield | The agent OS, full stop |
| 🥈 | Citadel | Seth Gammon (SethGammon) | 923★, 701 commits, v1.4.3 | Claude plugin + Codex plugin + 48 skills | Honest orchestration, verified boundaries |
| 🥉 | planning-with-files | OthmanAdi | 27.1k (+752/30d) | SKILL.md + on-disk plan templates | The memory shim every long session needs |
Why these ranked here
🥇 ECC — 260,252 stars makes it the second-largest skill repository on GitHub, and this week it out-trended everything (fourth on the September 9–15 GitHub trending board). Inside: 262 skills spanning coding, research, security, and media; persistent memory that survives sessions; "instincts" that normalize agent behavior across Claude Code, Codex, OpenCode, and Cursor; and AgentShield, which scans your own agent setup for vulnerabilities. Note: the deep-research skill inside this repo ranked separately on September 18 — the harness as a whole debuts now, and lands with gold.
🥈 Citadel — A different philosophy: instead of 262 skills, one /do entry point, repo-local state that survives sessions, intent routing, safety hooks, cost telemetry, and parallel agent fleets in isolated worktrees. What earns it silver at 923 stars is radical evidence discipline — the README publishes failed experiments alongside wins ("a v1 experiment does not support a savings claim"), ships checksum-verified release trios, and treats missing evidence as blocked, never success. The most honest orchestrator we've reviewed this year.
🥉 planning-with-files — Persistent task plans, findings, and progress logs written to disk so agents survive multi-step workflows that blow past five tool calls. It's the simplest fix for the most common long-session failure — context amnesia — and 27.1k stars say thousands agree. Bronze on sheer utility per byte.
🪙 Token Reduction & Efficiency
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | graphify | Graphify Labs | 121k (+10.8k/30d) | SKILL.md + indexing pipeline | Your codebase as a queryable knowledge graph |
| 🥈 | Understand-Anything | Egonex AI | 84k (+3.6k/30d) | SKILL.md + visualization scripts | Domain flows and change analysis, visualized |
| 🥉 | context-mode | mksglu | 23.2k | TypeScript middleware + hooks | 98% tool-output trim, 17 platforms |
Why these ranked here
🥇 graphify — Converts code, docs, and media into a persistent knowledge graph with community detection, so agents can ask architectural questions instead of re-reading your repo every session, with GraphRAG integration ready. 121k stars and +9.8% monthly growth make it the heavyweight of context efficiency — the problem every long-running agent hits and few solve this cleanly.
🥈 Understand-Anything — Egonex AI's knowledge-graph tool visualizes codebase structure, extracts domain flows, and analyzes code changes. Where graphify builds the index, this one makes it legible to humans — 84k stars of onboarding-weeks-saved. Silver for attacking the same problem from the comprehension side.
🥉 context-mode — Middleware that sits between your agent and its tools, aggressively trimming noisy tool output (a claimed 98% reduction) before it eats the context window, while persisting memory across sessions and routing consistently across 17 agent platforms via MCP and hooks. Featured on the September 9–15 GitHub trending board at 23.2k stars. It's plumbing — but it's the plumbing your context window has been begging for.
🎨 Design & Creative
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | taste-skill | Leonxlnx | 89.8k (+9.2k/30d) | SKILL.md collection (direction, brand, motion) | Premium frontend direction, not just clean CSS |
| 🥈 | archify | tt-a1i | 71.1k (+55.3k/30d, +350.5%) | SKILL.md v2.17 + bin/archify.mjs | Architecture diagrams worth putting in a deck |
| 🥉 | impeccable | Paul Bakaus (pbakaus) | 70.7k (+8.2k/30d) | 1 skill, 24 commands, 61 detector rules | The design vocabulary agents were missing |
Why these ranked here
🥇 taste-skill — 89.8k stars and +11.5% monthly growth make it the biggest design-focused skill collection on GitHub: premium frontend direction, visual generation, redesigns, brand systems, and motion-led implementation. Where most anti-slop skills tell the agent what not to do, taste-skill gives it a point of view first — the agent reads the brief, infers the right design direction, and commits. Gold on ambition and adoption alike.
🥈 archify — The breakout of the month: +350.5% in 30 days (+55.3k stars), #1 GitHub trending when it launched, now 71.1k. Feed it plain-language requirements or raw Mermaid and it emits explorable standalone HTML architecture, workflow, sequence, and state diagrams with dark/light themes, trace motion, and PNG/SVG/WebM export — inspecting repository evidence when the diagram must reflect real code. From 4.2k stars on August 28 to 71.1k today is the single fastest climb we've ever logged.
🥉 impeccable — Paul Bakaus's design language for agents: 24 named commands (/polish, /distill, /quieter, /animate…), 61 deterministic detector rules for AI-frontend slop, live browser iteration, and brand vs. product modes — now extending into design review for pull requests. 70.7k stars, +13.2% for the month, and a freshly viral Threads moment this week. It never ranked here before because we'd been reviewing Anthropic's official design skills; with those in the dedup window, the indie king finally takes its podium.
🔬 Research & Analysis
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | last30days-skill | mvanhorn | 62.8k (+3.5k/30d) | SKILL.md + multi-platform search pipeline | What the internet actually said, this month |
| 🥈 | OpenResearch | alphaXiv | 5.7k★, 382 commits | SKILL.md + CLI + local dashboard + agent-skills/ | Autoresearch with a lab notebook |
| 🥉 | Auto-claude-code-research-in-sleep | wanshuiyin | 16.6k (+9.2%/30d) | SKILL.md + automation scripts | The paper-grinding night shift |
Why these ranked here
🥇 last30days-skill — Researches any topic across Reddit, X, YouTube, Hacker News, Polymarket, and the open web over the last 30 days, then synthesizes a grounded brief ranked by real engagement. 62.8k stars, +6.0% monthly. It fills the exact gap general-purpose research skills leave open: recency-weighted, platform-wide sentiment instead of stale training-data vibes. It's the skill version of this very leaderboard's methodology — and that's the highest compliment we give.
🥈 OpenResearch (alphaXiv) — The team behind alphaXiv turned coding agents into research agents: literature review, hypothesis development, experiment tracking in a git-native tree, parallel exploration in isolated worktrees, and an orx CLI that runs the full autoresearch loop — propose, change, experiment, inspect, iterate. Six reference skills ship in agent-skills/ including orx-figures for publication-quality matplotlib/TikZ charts. 5.7k stars in its first weeks with commits landing daily. The most serious research-infrastructure entry we've tested.
🥉 Auto-claude-code-research-in-sleep — An automation toolkit that searches academic papers, analyzes ML experiment statistics, plans ablation studies, and generates technical diagrams while you're offline. 16.6k stars and steady growth. gimmicky name, deadly serious pipeline — the closest thing to a research agent with a night shift.
🔒 Security & Governance
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | security-audit-skill | Cloudflare | 21.2k (+11.4k/7d, +595.4%/30d) | SKILL.md + audit scripts | Corporate-grade vuln research, now open source |
| 🥈 | reverse-skill | zhaoxuya520 | 37.2k (+28.0%/30d) | SKILL.md + sandboxed workflows | Red-team techniques, responsibly sandboxed |
| 🥉 | Anthropic-Cybersecurity-Skills | mukul975 | 33.3k (+2.3k/30d) | 818 structured SKILL.md files | The red-team reference library |
Why these ranked here
🥇 security-audit-skill (Cloudflare) — The story of the week: security audits and vulnerability research across codebases, APIs, services, and CLI tools, released by Cloudflare — and the community went feral. +117% in seven days (+11.4k stars), +595.4% for the month — the fastest growth rate on this week's entire board, any category. Coming weeks after Snyk's warning that 36% of audited community skills carry injection risks, an official, auditable security skill from a company whose day job is security is exactly what this shelf needed.
🥈 reverse-skill — Structured workflows for prompt-injection testing, APK hooking, and Active Directory certificate abuse — explicitly scoped to sandbox environments. 37.2k stars and +28% monthly growth. Blue-team value, red-team depth: you can't defend what your agent can't reproduce.
🥉 Anthropic-Cybersecurity-Skills — 818 structured skills covering red-team tasks from DPAPI decryption to AD privilege escalation. 33.3k stars, methodically maintained. The breadth is unmatched — when you need the exact technique documented as an installable procedure, it's probably in here.
📄 Office & Productivity
| Rank | Skill | Developer | Stars | Format | Verdict |
|---|---|---|---|---|---|
| 🥇 | marketingskills | Corey Haines (coreyhaines31) | 51.4k (+5.8k/30d) | 43-markdown skill pack (32 skills) | Real marketing frameworks, not "write me copy" |
| 🥈 | obsidian-skills | Kepano (Steph Ango, kepano) | 48.8k (+1.6k/30d) | SKILL.md suite (vault, JSON Canvas, plugins) | Your vault gets a resident librarian |
| 🥉 | book-to-skill | virgiliojr94 | 32.3k (+6.9k/30d, +27.0%) | SKILL.md + extraction pipeline | Turn any book into installable agent skills |
Why these ranked here
🥇 marketingskills — Corey Haines (Swipe Files, Conversion Factory) packs 43 markdown files of genuine marketing frameworks — CRO, copywriting, SEO, analytics, paid ads, lifecycle email, growth engineering — into a 32-skill bundle with 281.9K installs, the highest-installed marketing-specific collection on record. 51.4k stars and +12.8% monthly growth. The first skill pack that made our marketing friends stop laughing at agent output.
🥈 obsidian-skills — From Kepano, Obsidian's CEO: a suite for managing vaults, editing JSON Canvas files, and debugging plugin development. 48.8k stars of first-party authority — when the maintainer of the knowledge tool writes the skills for the agents that touch it, provenance anxiety goes to zero. The knowledge-management category finally has its anchor tenant.
🥉 book-to-skill — Converts books and documents into structured agent skills by extracting frameworks, principles, and techniques — 32.3k stars, +27% in a month. The meta-skill of the moment: the fastest-growing conversion category on the board, and this is its cleanest implementation. Your library is now a skills directory.
📈 Movers & Shakers
- Biggest riser: cloudflare/security-audit-skill — +595.4% in 30 days (+11.4k stars in the last 7). The fastest velocity on the board, fueled by post-Snyk-audit demand for skills with real provenance.
- New entry of the week: tt-a1i/archify — from 4.2k stars on August 28 to 71.1k today (+350.5%). Diagram generation was the whitespace; archify filled it at record speed.
- Defining story: affaan-m/ECC at 260,252 stars — the "agent OS" pattern (skills + memory + instincts + security in one harness) now outranks every official vendor repository on GitHub. The community is no longer waiting for platform teams.
- ⚠️ Ecosystem watch: Notion 3.7 shipped team-wide agent skills on September 15, and OpenAI published its own skills/plugins references — the megaplatforms are formally entering the format. Expect the "official vs community" boundary to get noisy.
🔮 Next Week's Watch List
- cathrynlavery/diagram-design — 42.3k stars, +59.5%/30d: branded architecture and data-flow diagrams in HTML/SVG/PNG. archify's diagram thunder may get stolen by the branding specialist.
- Nutlope/hallmark — 29.1k stars: extracts visual DNA from URLs or screenshots and applies intentional UI patterns to generated pages. Design & Creative podium contender the moment the dedup window clears.
- orx-figures (alphaXiv/OpenResearch) — the publication-quality figures skill specifically; if OpenResearch's velocity holds, the alphaXiv family could take multiple Research slots next week.
Ranked by AgentSkillReview. Data sourced from The Skill Leaderboard (data through September 24, 2026), the GitHub trending roundup, Sep 9–15, live repository verification, and community forums. Dedup window: 30 days against 79 previously reviewed skills. Modular SKILL.md skills only — no platforms, no SaaS.