Every Agent Skill · Rated
Cross-ecosystem agent skills directory · Claude Skills, OpenClaw ClawHub, and the generic AgentSkills-compatible format graded on the Gecko Rating ladder, with install snippets for every runtime.
What is an agent skill?
An agent skill is a packaged capability · a SKILL.md instruction file plus optional scripts and dependencies · that an AI agent installs and invokes like a plugin. Skills teach an agent how to do something (read a PDF, run a git workflow, scrape a page) while MCP servers give it the tools to actually execute those actions against real systems.
Drop SKILL.md files into ~/.claude/skills/ · Claude Code, Desktop, and the API auto-discover them on next launch.
Browse Claude SkillsInstall via `openclaw skills install <slug>` from the ClawHub registry · versioned, sandboxed, community-curated.
Browse OpenClaw skillsGeneric SKILL.md + YAML frontmatter format · one skill works across Claude, OpenClaw, and any spec-compatible harness.
Browse cross-ecosystem skillsThe Gecko Rating ladder
· 15 skills gradedMaturity-first quality tier · we score each skill on release maturity, creator type, runnable scripts, cross-client support, and metadata hygiene. Untested is neutral· it means we don't have enough public signal to rate the skill, not that it is bad.
Official + production-ready · 1.0+ release, company-backed, scripts included, cross-client support.
Mature community skill · stable release, public repo, multiple tested clients.
Working prototype · early release, limited testing, early adopters welcome.
Not enough public signal yet · no clear maturity indicator, try at your own pace.
Ecosystem split
One skills concept · 3 ecosystems · every format tracked
Anthropic's Claude Skills · drop SKILL.md files into ~/.claude/skills/ and Claude Code + Desktop + API auto-discover them. Official skills ship from the anthropics/skills repo.
OpenClaw ClawHub registry · skills install via `openclaw skills install <slug>` from the community registry at clawhub.ai. Versioned, reviewable, sandboxed.
Generic AgentSkills-compatible format · SKILL.md + YAML frontmatter + optional scripts. Works with any agent harness that reads the spec · cross-ecosystem by design.
Category constellation
Every skill category · bubbles sized by count
PDF · Word · Excel · PowerPoint · form-filling · document extraction.
Git workflow · code review · API testing · skill authoring · meta-skills.
Brand guidelines · design tokens · screenshots · UI regression · asset generation.
Multi-source web research · citation synthesis · literature review.
SQL introspection · schema discovery · query generation · data pipelines.
Headless browsers · scrapers · HTML parsing · content extraction.
Terraform · Kubernetes · CI/CD · cloud infra · deployment automation.
Best for the job
Top 5 skills for common facets · rankings refresh every time the data does, no editorial picks.
The full leaderboard
· 15 of 15 skillsFrequently asked
Answers pulled from the dataset · updated daily
What is an agent skill?
An agent skill is a packaged capability · a SKILL.md instruction file plus optional scripts and dependencies · that an AI agent installs and invokes like a plugin. Claude Skills live in ~/.claude/skills/, OpenClaw skills install from ClawHub, and the generic AgentSkills format works across any spec-compatible agent harness.
How many agent skills does BenchGecko track?
BenchGecko tracks 15 agent skills across 3 ecosystems and 7 categories · Claude Skills (9), OpenClaw ClawHub (3), and generic AgentSkills-compatible (3). Every skill is graded on the Gecko Rating ladder and refreshed daily.
How does Gecko Rating work for skills?
We score each skill 0-100 on version maturity, creator type, scripts included, cross-client support, and testing signal · then bucket into Frontier (≥80 · 8 skills), Solid (≥60 · 2), Beta (≥35 · 5), and Untested (0). Skills don't have GitHub stars the way MCP servers do, so the ladder leans on release maturity and metadata hygiene.
What is the difference between a Claude Skill and an OpenClaw Skill?
Claude Skills are Anthropic's native format · SKILL.md files that Claude Code, Claude Desktop, and the Claude API auto-discover from ~/.claude/skills/. OpenClaw skills are versioned packages published to the ClawHub registry and installed via `openclaw skills install <slug>`. Both read the same SKILL.md shape · spec-compatible skills work in either ecosystem.
Are agent skills the same as MCP servers?
No · they are complementary layers. Skills are instruction packages the agent reads to know *how* to do something. MCP servers are long-running processes the agent talks to over a protocol to actually *execute* tool calls against external systems. Many skills pair with MCP servers · a "database explorer" skill might use a Postgres MCP server.
What is the highest-rated agent skill right now?
PDF tops the Gecko Rating ladder with 95/100 · an Claude Skills skill by Anthropic for Docs · PDF · Office.
How do I install an agent skill?
Each skill detail page on BenchGecko includes ready-to-paste install snippets for its native ecosystem. Claude Skills copy the SKILL.md into ~/.claude/skills/<slug>/. OpenClaw skills run `openclaw skills install <slug>`. Generic AgentSkills-compatible skills clone the repo into your agent harness skills folder.
Can I submit a skill to BenchGecko?
You don't need to · we sync the official Anthropic Skills repo, the OpenClaw ClawHub registry, and the AgentSkills-compatible community index automatically. If a skill is missing, open a PR on the BenchGecko skills-seed.json file or publish it to one of the upstream registries.
See also
Keep exploring the BenchGecko graph