What you are looking at

The tree is a complete skill in the open format: one required file, three optional directories, relative paths one level deep from SKILL.md. It is original work for this site, labeled CS-SF-001. It is not a listing from Claude’s Skills directory and it is not a claim that any particular product will execute the script.

Three loads, not one dump

Progressive disclosure is the reason a folder this small can carry a checklist, a script, and a template without paying for them on every turn. The spec’s own budget is the interaction you just walked:

Discovery

Metadata only, about 100 tokens, at startup for all installed skills. This specimen’s name and description estimate at 40 tokens (words × 1.3, not a model tokenizer).

Activation

The selected SKILL.md. Keep it under 500 lines and under about 5,000 tokens. This specimen is 36 lines and estimates at 326 tokens.

Execution

Scripts, references, and assets only when the workflow names them. The whole folder estimates at 913 tokens if you did the wrong thing and loaded it all.

"Agents load skills progressively, pulling in more detail only as a task calls for it."Agent Skills specification, Progressive disclosure

This is not Claude’s Skills directory

Claude Has Skills is the right place to browse Claude-specific, installable skills. Convergent Skills does not copy that catalog. The job of this page is the portable contract underneath: a folder an author can hand from Claude to Codex to Copilot to an MCP development workflow without rewriting the idea, even when the runtime around the folder changes.

Open Claude Has Skills when you want a Claude skill to install. Stay here when you want to understand the format you would author into.

How to read a skill you did not write

  1. Read name and description as the activation contract.
  2. Read the body for when not to use it, then for the workflow.
  3. Follow only the paths the body names. If a file is never named, it is dead weight.
  4. Treat scripts as optional. If the client cannot run them, the skill should still degrade to instructions.
  5. Check compatibility before you assume the folder will run as-is.

To draft one of your own, use the Skill Scaffold Generator. To test whether the description will fire, use theDescription Linter. The prose guide isAnatomy of a skill.

Questions

What is a portable agent skill folder?

A directory with a required SKILL.md file: YAML frontmatter for discovery, then Markdown instructions for the workflow. Optional scripts/, references/, and assets/ sit beside it. Agents load the name and description first, the full SKILL.md when the skill activates, and supporting files only when the current case needs them. That directory shape is the open Agent Skills format.

Is SKILL.md the whole skill?

It is the only required file. A useful skill often still needs scripts for fragile checks, references for policy, and assets for templates. Those files are optional in the spec and should stay out of context until SKILL.md points at them.

Where is Claude's official Skills directory?

At Claude Has Skills (claudehasskills.com). Convergent Skills does not duplicate that catalog. This site owns the cross-ecosystem folder format: how SKILL.md, scripts, references, and assets are supposed to load, and how far that shape actually travels.

Can this folder run on every agent?

The directory shape travels. Runtime behavior does not, automatically. Script execution, sandboxing, install paths, and product packaging vary by client. Compatibility belongs in frontmatter when the skill actually needs a particular environment.

How should I start a skill of my own?

Start from one repeated workflow. Write the description before the body so activation is a contract. Keep SKILL.md short. Point at supporting files with relative paths one level deep. Test with prompts that should fire and prompts that should not. The Skill Scaffold Generator on this site will draft the file; it will not invent the judgment.