About Cult of Claude

What we do

Cult of Claude is an independent directory and marketplace of Claude Code skills and agents, built and operated by FDF Labs. A skill is a SKILL.md folder that teaches Claude Code a procedure; an agent is a Markdown file defining a subagent with its own prompt, tools and context window. We crawl public GitHub for both, score what we find, and publish one page per entry with its source file, the tools it declares and a working install command, so you do not have to read dozens of repositories to find the two worth installing. The guides explain the terminology.

Our Sources

Every public GitHub repository with a skill or agent file is a source; these five contribute the most entries.

Methodology

• 1 Collect skills and agents from multiple open-source repositories
• 2 Deduplicate entries that appear in multiple sources
• 3 Enrich with consistent metadata: categories, difficulty levels, tools used
• 4 Categorize into a unified taxonomy for easy browsing
• 5 Generate static pages for fast, searchable access

How entries are scored

Two separate scores are computed, both out of 100, both from metadata rather than from anyone's opinion. Neither is a security review, and neither measures whether a skill works. They measure documentation and adoption.

The quality gate

Every crawled entry is scored once before it is listed; anything under 40 is quarantined and never gets a page. Points are added for a README over 100 characters (+15), a trigger condition (+10), a long description that differs from the short one (+10, plus +5 when it runs past 400 characters), a description over 40 characters that is not a known template (+10), an official Anthropic origin (+10), a declared tool list (+5), an install command (+5), a source path that looks like a skill or agent location (+5), and GitHub stars (+5 each at more than 0, 50 and 500). Points are taken away for templated descriptions such as bulk "skill for X workflows" dumps (−30) and entries describing an MCP server rather than a Claude Code skill (−25 or −20), a description identical to the name (−25), an author with more than 100 entries in the catalog (−15), a stub where the short and long description are the same and under 25 characters (−15), no README, description, trigger or tools at all (−10), and a description under 20 characters (−10).

The displayed score

The number shown beside entries on the homepage and in search results is a different, simpler formula meant for ranking listed entries against each other: twelve points per order of magnitude of GitHub stars (log₁₀(stars + 1) × 12), plus +10 for a README and +5 each for a long description, a trigger, an install command, usage tips, a declared tool list and an official Anthropic origin. Agents swap trigger and tips for a specialization and use cases, with up to +10 more for the number of use cases. It is capped at 100, and an official entry with no stars of its own is capped at 40 so that Anthropic's example skills do not crowd out community work.

Stars are the repository's stars, not the individual file's, which is a known weakness: one popular repository can lend its stars to every entry inside it. When you see a high score, read the source file anyway.

Contributing

Have a skill or agent to share? Visit the submit page to see how entries are discovered. Already listed? There is a badge for your README.

Who runs this

Cult of Claude is built and operated by FDF Labs, which also runs the crawler, pays for the hosting, and answers for what is listed. The site is free, carries no advertising, and takes no payment for placement. If it is useful, the support page explains how to chip in.