Agent skills, beyond one vendor

Convergent Skills

Agent skills are portable folders of capability: a required SKILL.md with discovery metadata and workflow instructions, plus optional scripts, references, and assets that load only when the task needs them. That folder shape is the open format. Claude’s Skills directory is at Claude Has Skills; this site is the cross-ecosystem craft. Open the Skill Folder to inspect a specimen the way an agent would load it.

Open the Skill FolderUse free toolsCompare skills, tools, and MCP

Signature experience

The Skill Folder

Walk a labeled specimen the way a compatible agent loads it. Layer 1 is name and description. Layer 2 is the rest of SKILL.md. Layer 3 is the checklist, script, and report template, only when the workflow names them. Original work for this site. Not a Claude catalog listing.

Inspect the folderClaude Skills directory
release-readiness/SKILL.mdscripts/verify-release.shreferences/release-checklist.mdassets/report-template.md

Free tools

Build the first draft.

Browser toolSkill Scaffold Generator

Build a format-aware SKILL.md scaffold from a name, trigger description, workflow steps, resources, and validation rules.

Browser toolSkill vs MCP Decider

Answer a few design questions and get a practical recommendation: skill, MCP server, or both.

Browser toolSkill Description Linter

Score a skill description for trigger coverage, verbs, boundaries, length, and missing activation cues.

The thesis

Skills are capability memory.

A skill is not just a prompt, and it is not the same thing as a tool. It is a portable packaging pattern for capability: what the agent should know, how it should decide that the knowledge applies, which files or scripts support the workflow, and how to validate the result. The format matters because capable agents increasingly need repeatable craft, not another copy-pasted instruction block.

The useful mental model is a three-part split. Skillspackage task knowledge. Tools execute actions. MCP standardizes how hosts connect to external systems that expose tools, resources, and prompts. These pieces overlap in practice, but they solve different problems.

This site is intentionally cross-ecosystem. For Claude-only browsing, examples, and directory-style discovery, see Claude Has Skills. Convergent Skills focuses on the broader pattern: how reusable skills move across Claude, Codex, GitHub Copilot, MCP development workflows, and whatever agent runtimes adopt the same basic contract next.

If you are ready to make one, start with the Skill Scaffold Generator, check the description in theDescription Linter, or use theSkill vs MCP Decider when the architecture is unclear.

Map

Read the system.

The Skill FolderInteractive anatomy of a portable skill directory: SKILL.md, scripts, references, assets, and the three-stage load.Free toolsClient-side generators, checklists, and linters for authoring better portable agent skills.Skills vs tools vs MCPA taxonomy for prompts, tools, MCP servers, resources, and reusable skills.Anatomy of a skillThe SKILL.md file, frontmatter, scripts, references, assets, and progressive disclosure model.AuthoringHow to write, test, distribute, and maintain reusable skills without stuffing context.Claude skill templateA source-backed SKILL.md template with frontmatter, trigger guidance, workflow sections, and portability notes.Skill description best practicesHow to write descriptions that trigger at the right time without becoming too broad or too vague.PortabilityWhere skills travel today, where they do not, and what portability really depends on.ExamplesReusable skill patterns and case studies across document work, engineering, MCP development, and teams.GlossaryConcise definitions for agent skills, MCP, tools, resources, prompts, progressive disclosure, and related terms.SourcesAnnotated bibliography, source map, and machine-readable llms.txt links.FAQShort answers to the common questions: what skills are, how they differ from tools and MCP, and how to start.

Source-backed facts