Skill Atlas

About the atlas

A public reading room for agent capabilities.

Skill Atlas is a bilingual technical research directory for people deciding whether to install a Claude Code Agent Skill—and for readers learning how skills are constructed.

What belongs here

The catalog begins with publicly available skills whose source can be collected under compatible terms. Each record preserves a path back to its upstream repository. Inclusion is not endorsement, certification, or a claim of official partnership.

Who it is for

Practitioners

Inspect installation commands, resource structure, prerequisites, and risk surfaces before bringing a capability into your workspace.

Learners

See how instructions, linked resources, and workflow choices turn a markdown file into an agent capability.

Researchers

Work from versioned source snapshots with explicit evidence boundaries instead of popularity rankings or unsupported scores.

Provenance over popularity

Stars and update dates describe the repository, not the reliability of a skill. The core record is the artifact itself: where it came from, when it was collected, what it contains, and which methods produced each observation.

How records change

A synchronization pipeline can refresh public source data without paid model calls. Exact token analysis and controlled evaluation are separate, explicit operations. The snapshot timestamp describes the catalog as a whole; each dossier also retains source and review dates.

One record, two reading languages

Interface and research content are available in English and Simplified Chinese. Switching language preserves the page you are reading. If a localized research field is missing, the interface falls back to its English source rather than hiding the content.

Continue upstream

For implementation details and the authoritative source files, consult the official Agent Skills documentation and public repository.

Agent Skills documentationPublic skills repository

A research surface, not a marketplace

There are no download rankings, sponsored placements, or invented quality scores. The purpose is simpler: make a technical artifact easier to inspect before you trust it.