Guides
Plain answers to common questions about agent skills and the instruction files AI agents read, each with commands that work and figures from our own data. The data behind the figures is published in full in the reports, and studies built on it are on the blog.
Pin agent skills by content hash, keep the pin in a lockfile, check for upstream changes on purpose and read the diff first. Real commands and output.
What AGENTS.md, CLAUDE.md, SKILL.md, Cursor rules, DESIGN.md and llms.txt do, which agents read each, and how to keep AGENTS.md and CLAUDE.md in sync.
What to check in an agent skill before you install it, a command that finds curl piped into a shell, and what an audit of public SKILL.md files flags.
skills.sh, SkillsMP, ClawHub, gh skill, Anthropic's marketplaces and markdownregistry, compared from their own docs: listing, install, audit, versions.
The two required SKILL.md fields and their rules, how many public skills break them, how skills load, and when to use a skill or an MCP server.
Keep a team on the same agent skills with a committed lockfile, install exact versions, share Claude Code skills, and publish your own. Real output.
From the blog
- How often agent files change: 31,009 files over four weeks
- What goes in a CLAUDE.md: 2,536 files counted
- What goes in an AGENTS.md: 4,687 files counted
- What ships beside SKILL.md: 70,893 skill folders counted
- How SKILL.md files say when to use a skill: 73,322 counted
- DESIGN.md files against the official linter: 1,162 checked
- How llms.txt files follow the format: 438 files counted
- How projects write Cursor rules: 557 files counted
- AGENTS.md and CLAUDE.md in one repo: what 1,203 folders do