What Claude Skills and Subagents Actually Are
Anthropic launched the Agent Skills specification in December 2025 as an open standard for packaging reusable agent capabilities (a skill is roughly a bundle of system-prompt + tool definitions + reference materials, callable from a parent agent). OpenAI adopted the same specification for Codex CLI and ChatGPT shortly after, making Skills the first cross-vendor open standard for agent extensions. As of May 2026, the directory has over 1.2 million Skills. This page consolidates the ecosystem state and explains why Skills matter for brand visibility.
Skills Ecosystem Snapshot (May 2026)
| Metric | Value |
|---|---|
| Specification origin | Anthropic, December 2025 |
| Cross-vendor adoption | Anthropic Claude, OpenAI Codex CLI, ChatGPT |
| Total available Skills (May 2026) | 1.2 million+ |
| Primary distribution channels | Lobehub Skills, SkillsMP, MCP Market Skills, Anthropic registry |
| Skill bundle composition | System prompt + tool definitions + reference docs + sometimes code |
How Skills Relate to MCP and A2A
| Layer | Standard | What It Provides |
|---|---|---|
| Agent capability packaging | Agent Skills | How a parent agent uses a specialised capability internally |
| Agent-to-tool calls | MCP | How an agent talks to external tools (databases, APIs, files) |
| Agent-to-agent coordination | A2A | How peer agents delegate work to one another |
The three standards are complementary. A modern agentic system typically uses Skills to organize its own behaviour, MCP to access external tools, and A2A to coordinate with peer agents. None of the three replaces the others.
Six Things the Skills Ecosystem Tells You
- Anthropic shipped a cross-vendor open standard, which is rare. Most "open standards" in AI are vendor-controlled. Anthropic published Skills as a true open spec and OpenAI's rapid adoption (Codex CLI + ChatGPT) validates the format. With MCP also Anthropic-originated and now industry-standard, Anthropic has shipped two cross-vendor primitives in 18 months.
- 1.2M Skills is a working ecosystem, not vapourware. The directory has been growing approximately 5-8x per month since spec launch. Most Skills are narrow (a coding helper, a search wrapper, a domain-specific assistant); breadth comes from quantity rather than per-skill capability.
- Subagent hierarchies are now the production pattern for complex agents. A parent agent decomposes a task, calls specialist subagents (research, code, finance, vertical-specific) via either Skills (internal capability) or A2A (peer agent). Single-agent architectures are increasingly rare in production agent stacks.
- Skill marketplaces matter for discoverability. Lobehub, SkillsMP, MCP Market Skills, and the Anthropic registry are the primary discovery surfaces. Brands building Skills around their products gain visibility inside parent-agent decompositions; the marketplace listing is the discovery channel.
- Skills carry brand-relevant context. A Skill that helps an agent "research B2B SaaS pricing" includes its own reference doc pointers and authoritative source preferences. Brands listed prominently in widely-used Skills get over-represented in the resulting agent outputs.
- Quality varies sharply. The open-spec, low-barrier-to-publish nature means many Skills are duplicative or low-quality. Discovery is currently a problem; expect curation layers (vetted Skill collections, brand-trust ratings) to emerge through 2026.
What This Means for AI Visibility
Skills are a new brand-discoverability surface that did not exist before December 2025. When a parent agent decomposes a task into "use the research Skill, then use the writing Skill, then use the verification Skill," the Skill itself can carry brand-relevant defaults: which research sources to prefer, which products to surface, which vendors to compare. Brands that author or sponsor widely-used Skills can compound visibility across millions of agent calls. Brands that ignore the Skills layer cede an emerging discovery channel.
Methodology
Skills directory counts aggregated May 14, 2026 from Lobehub Skills, SkillsMP, MCP Market Skills, and Anthropic's public spec announcements. Cross-vendor adoption confirmed via OpenAI's Codex CLI and ChatGPT release notes. Refreshed quarterly as the ecosystem continues to grow.
How Presenc AI Helps
Presenc AI tracks brand-mention rates across agent-mediated workflows, including those where Skills mediate the underlying brand-recommendation logic. When a brand appears prominently in a widely-used Skill, the resulting visibility lift compounds across millions of downstream agent calls. For brands building developer-facing or B2B-prosumer AI visibility strategies, the Skills layer is now a structural component of total AI presence.