Skills Collection — Curated Agent Skills for Every Tool

A curated, vetted catalog of agent skills (SKILL.md) that work across Claude Code, OpenCode, Gemini CLI, Codex, Cursor, and 40+ other agents. Each skill includes install tabs, full source, and instructions for verifying it yourself.

August 27, 2026
skillsagent-skillsskills-collectionskill-packsagentskills-io
Skills Collection — Curated Agent Skills for Every Tool

Skills Collection

Agent skills are the open standard for packaging specialized knowledge into portable folders. A skill is just a SKILL.md plus optional scripts, references, and assets — installable in Claude Code, OpenCode, Gemini CLI, Codex, Cursor, and 40+ other agents. One skill, every tool.

This collection is a curated catalog, not a registry. The skills.sh marketplace has thousands of unaudited skills; ~26% of them carry security issues ranging from prompt injection to credential exfiltration. Discovery is no longer the bottleneck — judgment is. Every skill here has been reviewed against the open standard and tagged with what we checked, which models reviewed it, and when. That review is transparency, not a trust seal: repos change silently, so we ship instructions on every page for you to verify the skill with your own models before installing.

What's in the collection

[find-skills](/skills/workflow/find-skills)

The meta-skill for discovering other skills (1.8M+ installs). Start here.

[skill-creator](/skills/workflow/skill-creator)

Build your own skills from recorded workflows. Bridges to skill-packs.

[Superpowers](/skills/coding/superpowers)

Obra's full skills framework for coding agents. Pairs with superpowers-6.

[Anthropic Official Skills](/skills/coding/anthropic-official)

Reference implementations of the agentskills.io standard. Best for learning the anatomy.

[Gemini Skills](/skills/coding/gemini-skills)

Google's official repo. Documented 28%→96% API-code pass-rate lift.

[HF Skill Pack](/skills/coding/hf-skill-pack)

Hugging Face skill packs for ML/agent workflows on Spaces and beyond.

[Taste-Skill](/skills/design/taste-skill)

Anti-slop frontend skill. Bring taste and craft to AI-generated sites.

The open standard

Skills conform to agentskills.io, the open format originally developed by Anthropic and now adopted by ~40 products. Every skill is a folder with a SKILL.md:

my-skill/
├── SKILL.md          # Required: frontmatter + instructions
├── scripts/          # Optional: executable code
├── references/       # Optional: documentation
└── assets/           # Optional: templates, resources

SKILL.md frontmatter has six fields (name, description, license, compatibility, metadata, allowed-tools) and the body is loaded only on skill activation — agents use progressive disclosure so installing 20 skills doesn't cost 20 instruction sets upfront. The same skill file runs unmodified in Claude Code, OpenCode, Codex, and Gemini CLI.

Install it once, run it anywhere

Use the skills.sh CLI for the broadest ecosystem (~76 agents supported):

npx skills add <github-org>/<repo>

This installs the skill into whichever agent's local skill directory is detected. The published skill is whatever the repo's SKILL.md points to.

How to read a skill page

Every skill page follows the same template so you can compare them at a glance:

  1. Overview — what it does, who maintains it, adoption signal
  2. When to use / when not — the realistic job
  3. How it works — the SKILL.md anatomy, scripts, progressive disclosure
  4. Install tabsnpx skills add, Claude Code, OpenCode, Codex
  5. Full SKILL.md — copyable source after the vetting gate passes
  6. "Verify it yourself" — the condensed red-flag checklist + a copyable review prompt you can paste into your own model
  7. Review transparency — what we checked, which models, when
  8. Source + credits + license — where it came from, what you can do with it

Before you install anything

Note:

The skills ecosystem is young. The "Do Not Mention This" prompt-injection attack wave and ~26% vulnerable-skill rate from the Aug 2026 sweep mean you cannot treat a skill as inert text. It is an instruction set that will be loaded into an agent with real tool access. Always run the verification block on the page before installing — or, at minimum, paste the SKILL.md into a model that has no tools, no file access, and no network and ask for a behavior review.

Start with the Vetting Guide to learn the methodology, then find-skills to discover more.