Agent Skills & Plugins Directory
Every coding agent now loads reusable "skills" — but they are scattered across thousands of repos. This is a hand-picked shortlist of the ones actually worth installing, for Claude Code, Cursor, Codex and Gemini CLI.
Last updated October 22, 2026 · star counts are a snapshot from that date
1. Official Catalogs — start here
Maintained by the platform vendors themselves. The safest first installs.
Anthropic's official catalog of high-quality Claude Code plugins.
The canonical starting point — curated, versioned, and growing weekly.
Open-source plugins for knowledge workers (docs, research, planning) in Claude Cowork.
Best pick if your agent work is writing and analysis, not just code.
cursor/plugins★ 13.5K
Cursor's plugin specification plus the official plugin set.
Defines where the Cursor ecosystem is heading — worth watching even if you don't use Cursor.
android/skills★ 9.2K
Google's official Android engineering skills for coding agents.
Signals that platform teams now ship agent skills as first-class developer tooling.
2. Engineering & Workflow
Skills that change how your agent writes, reviews, and ships code.
mattpocock/skills★ 279.7K
"Skills for Real Engineers" — composable engineering skills straight from a working expert's agent setup.
The most-starred skills collection on GitHub. Battle-tested and pragmatic.
addyosmani/agent-skills★ 102.9K
Production-grade engineering skills and workflows for AI coding agents.
Strong on quality gates and review discipline; pairs well with any coding agent.
DietrichGebert/ponytail★ 172.5K
"Makes your AI agent think like the laziest senior dev in the room" — biases the agent toward less code.
One idea, executed well. Genuinely reduces over-engineering in agent output.
Multi-phase security audits producing independently verified, machine-readable findings.
Vendor-backed and rigorous; the serious option for pre-release security passes.
cathrynlavery/diagram-design★ 50.5K
Editorial-quality diagram generation: 44 diagram types as self-contained HTML + SVG.
Fixes the "Mermaid slop" problem. Great for architecture docs and blog figures.
ayghri/i-have-adhd★ 55.2K
Formats agent output to be action-first and readable — stops agents burying the answer.
Silly name, real payoff. One of the highest signal-to-noise improvements you can install.
3. Knowledge & Research
Give your agent structured understanding of large codebases and domains.
Graphify-Labs/graphify★ 132.2K
Turns codebases, docs, SQL schemas and PDFs into a queryable knowledge graph — local AST parsing, no vector store.
A different bet on codebase context: deterministic graph instead of embeddings. Works with Claude Code, Cursor, Codex and Gemini CLI.
177 validated skills plus 100+ scientific databases — "turn any agent into an AI scientist".
Proof that deep vertical skill packs work. The template for other domains to copy.
HKUDS/CLI-Anything★ 56.1K
"Make all software agent-native" — exposes arbitrary CLI tools through an agent hub.
Broad reach; ideal once your agent routine goes beyond coding tasks.
4. Memory & Context
The current infrastructure bottleneck: making agents remember.
thedotmack/claude-mem★ 97.8K
Persistent context across sessions — captures, compresses and re-injects what your agent did.
The most compatible memory layer: Claude Code, Codex, Gemini, Copilot, OpenCode and more.
akitaonrails/ai-memory★ 12.6K
Long-term memory for coding agent CLIs, with handoff between different vendors' agents.
The cross-vendor handoff angle is unique — useful if you switch agents mid-project.
mksglu/context-mode★ 31.7K
Context-window optimization: sandboxes tool outputs, cutting context consumption by up to 98%.
Directly attacks the token bill. High impact on long agent sessions.
5. Learn the Ecosystem
Guides that teach the mental model instead of shipping a skill.
luongnv89/claude-howto★ 42.6K
Visual, example-driven guide to Claude Code — from basics to advanced agents, with copy-paste templates.
Best structured on-ramp if the skills ecosystem feels overwhelming.
How to install an agent skill (60 seconds)
- Claude Code: copy the skill folder into
.claude/skills/in your project, or install plugins via/pluginfrom the official catalog. - Codex / Gemini CLI / others: most skills are plain instructions plus scripts — drop them where your agent reads project context (
AGENTS.md, config folder), or follow the repo's README. - Cursor: use the plugin registry (
cursor/pluginsabove) or point Rules at the skill's instructions. - Start with one or two skills, not ten. Skills interact; keep only the ones that measurably improve output.