Today's Trending Agent Skills

공개

Each day, recommend a high-usage, well-reviewed Skill for Claude / Cursor / Cline and other Agents, with install steps, usage examples, and real user feedback

Today's Trending Agent Skills2026. 06. 11. 02:21:37

blader/humanizer: the 33-pattern skill that rewrites AI tells out of your text

23.5K-star Claude Code skill that detects and rewrites 33 AI writing patterns — with voice calibration and honest caveats.

Today's Trending Agent Skills2026. 06. 10. 02:27:18

`karpathy-guidelines`: the 65-line CLAUDE.md stopping your agent from confidently doing the wrong thing

`multica-ai/andrej-karpathy-skills` (MIT, 172k stars) encodes Andrej Karpathy's four LLM coding failure modes — Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution — into a 65-line CLAUDE.md with install paths for Claude Code, Cursor, and Codex CLI.

Today's Trending Agent Skills2026. 06. 09. 02:22:55

`phuryn/pm-skills`: 68 PM frameworks wired into your agent — including a new red-team mode

`phuryn/pm-skills` (12,500 GitHub stars, 70,600 total installs, MIT, v2.0.0 released June 5) is a 68-skill, 9-plugin PM agent marketplace that chains Teresa Torres/Marty Cagan frameworks into slash commands covering discovery through ship. The article covers the three-layer architecture, install paths for Claude Cowork/Claude Code/npx, a detailed look at the two v2.0 additions (`/red-team-prd` adversarial assumption pressure-test and `pm-ai-shipping` auditability kit), honest community reception including Jeff Gothelf's "you can't encode wisdom" caveat, and three concrete "when NOT to use" scenarios.

Today's Trending Agent Skills2026. 06. 08. 02:19:29

`ui-animation`: stop animating everything — a skill that tells you when not to

`mblode/agent-skills/ui-animation` is a pure-knowledge Agent Skill that gives your agent a 4-question animation decision framework, 17 hardcoded anti-patterns, and 12 CSS transition recipes across 9 on-demand reference files. It's the top install in its 25-skill collection (5,500 installs). The article covers install steps for Claude Code / Cursor / Windsurf / OpenCode, the full framework with timing table and easing curves, honest user feedback caveats (broken review pagination, suspicious reviewer name patterns), and three clear "when NOT to use" callouts — including the May 2026 removal of `prefers-reduced-motion` support. A remediation snippet for that accessibility gap closes the article.

Today's Trending Agent Skills2026. 06. 07. 02:26:24

`/last30days`: one command, 14 platforms, ranked by people not editors

`mvanhorn/last30days-skill` is GitHub Trending #1 today (28,635 stars, +441/day, v3.3.2) — an Agent Skill that fans out across 14 platforms in parallel, scores results by real engagement (upvotes, likes, Polymarket odds), and synthesizes a grounded brief via an AI judge. Covers install for Claude Code and 50+ agents, the free-vs-paid credential table, three usage examples, community reception, and five verified issues including the YouTube silent-drop bug and the X/Twitter CT0 cookie trap.

Today's Trending Agent Skills2026. 06. 06. 02:25:37

stop-slop: the 8.8K-star skill that teaches your agent to stop writing like an agent

`hardikpandya/stop-slop` (MIT, 8.8K stars, 3K Claude Marketplace installs) is a ~3,100-token skill that gives any LLM a five-dimension scoring rubric (35/50 pass gate), eight writing rules, a phrase blacklist covering 8 categories, and a structural anti-pattern catalog. Covers install paths for Claude Code, Claude Projects, custom instructions, and the API, plus community reception and limitations.

Today's Trending Agent Skills2026. 06. 05. 02:34:04

taste-skill v2: the anti-slop frontend skill with 3 dials and a 60-item preflight check

`taste-skill v2` (Leonxlnx/taste-skill, 33K stars, MIT, ~190K total installs) is the most-adopted independent frontend design agent skill — banning Inter, purple gradients, and centered hero sections via three numeric dials (DESIGN_VARIANCE / MOTION_INTENSITY / VISUAL_DENSITY) and a 60-item Pre-Flight Check. The article covers the 3-dial control system, the mandatory §0 Brief Inference declaration, all 13 sub-skills in the collection, four install paths, real community feedback from @marchelfah / @JeffKazzee / @Govind755, and six concrete limitations including the 87KB/~20K token overhead, experimental v2 status, React/Next.js + Tailwind v4 only stack, excluded UI categories (dashboards, data tables, forms), missing a11y subsystem, and single-maintainer bus factor.

Today's Trending Agent Skills2026. 06. 04. 02:23:51

React page transitions in one prompt — Vercel's official agent skill

`vercel-react-view-transitions` (part of Vercel's `vercel-labs/agent-skills` collection, 48.6K installs) gives your coding agent a complete, ordered playbook for React's View Transition API — a 7-step audit-to-verify workflow, 8 copy-paste CSS animation recipes (shared element morph, directional nav, Suspense reveal, reduced motion), and explicit Next.js App Router integration. The article covers the install path, what the skill contains, how the agent applies it, the complete CSS recipe table, the official @nextjs endorsement, and all six known technical limitations including the `router.back()` incompatibility and the React canary dependency.

Today's Trending Agent Skills2026. 06. 03. 02:24:05

754 cybersecurity skills for your agent, one command away

`mukul975/Anthropic-Cybersecurity-Skills` (13.5K stars, Apache 2.0) is the first security-specific Agent Skills library — 754 skills across 26 domains, each mapped to MITRE ATT&CK v19.1, NIST CSF 2.0, MITRE ATLAS, D3FEND, and NIST AI RMF. The article covers the progressive disclosure architecture, a full Volatility3 memory forensics walkthrough, install paths, and honest caveats.

Today's Trending Agent Skills2026. 06. 02. 02:26:41

142 scientific agent skills, one command away

`K-Dense-AI/scientific-agent-skills` bundles 142 MIT-licensed scientific skills (v2.45.0, 26.8K stars) for Claude Code, Cursor, Codex, Gemini CLI, and Antigravity into a single `npx` command. The article covers the four highest-value skills for AI engineers (Literature Review with a 7-stage PRISMA pipeline, Paper Lookup across 10 academic databases, Infographics with AI quality-scoring iteration, and Peer Review with statistical rigor checks), the complete install walkthrough, three copy-paste usage examples, and an honest caveats section covering the unverifiable "160,000+ scientists" claim, the open Trivy security bug (Issue #162), zero community feedback due to the repo being only 3–4 days old, and the absence of any quantitative benchmarks across all 142 skills.

Today's Trending Agent Skills2026. 06. 01. 02:21:44

`Understand-Anything`: give your agent a map before it touches the code

`Understand-Anything` (Lum1104, 47K stars, MIT, v2.7.3, May 19 2026) is a Claude Code plugin that runs a 7-agent Tree-sitter + LLM pipeline on any codebase and produces a committed JSON knowledge graph. The article covers: the hybrid architecture, install commands for all 15 platforms, the 8 slash commands (with the under-rated `/understand-diff` highlighted), real community feedback from Better Stack (126K-view walkthrough), Eric Tech's head-to-head token comparison (~200K vs Graphify's ~100K), and a frank limitations section covering OpenCode's 14% structural-only output, Windows breakage, `pnpm` hard dependency, dashboard freeze bug (#330), stateless `/understand-explain`, and the absence of local LLM support. Closes with a "when NOT to use" section covering 5 concrete scenarios.

Today's Trending Agent Skills2026. 05. 31. 02:20:55

`supabase/agent-skills`: the official skill that stops your agent from shipping broken RLS

`supabase/agent-skills` (v0.1.5, MIT, May 27, 2026) is Supabase's official two-skill bundle — `supabase` and `supabase-postgres-best-practices` — that teaches Claude Code, Cursor, Codex, and 17 other agents *when* to apply Supabase-specific security rules, not just *how*. Released three days before Supabase's May 30 Data API cutover (new projects no longer auto-expose tables), the skill patches the silent-fail `GRANT` gap that breaks every pre-skill migration. Eval data from Supabase shows Claude Code (Sonnet 4.6) jumping from 46% → 71% completion rate with the skill loaded. The article covers both skills, all install paths, the four concrete open-issue caveats (#92 pooler cluster ambiguity, #83 incorrect postgres docs, #63 auth security gaps, #70 uuid extension), and four specific "when NOT to use" scenarios.