Anthropic Skills

Executive Summary

anthropics/skills is Anthropic's official public repository of Agent Skills — self-contained instruction packages that teach Claude how to perform specialized tasks. The project is actively maintained and accelerating: the repo sits at ~149k GitHub stars and ~17.6k forks as of June 11, 2026 (up from ~73k stars in February 2026), with commits as recent as June 9, 2026 (GitHub). The key innovation is progressive disclosure: skills load context incrementally so the agent only reads what it needs.

The bigger story since this profile was first written: Agent Skills stopped being a Claude feature and became an ecosystem standard. Anthropic published the format as an open standard at agentskills.io on December 18, 2025 (AI Business), and as of June 2026 the spec's client showcase lists roughly 40 adopters — including GitHub Copilot, VS Code, Cursor, OpenAI Codex, Gemini CLI, Goose, OpenCode, Roo Code, Spring AI, Databricks, and Snowflake (agentskills.io).

The SKILL.md Format

Every skill is a folder with a required SKILL.md file:

---
name: my-skill-name
description: A clear description of what this skill does and when to use it
---

# My Skill Name

[Instructions Claude follows when this skill is active]

## Examples
- Example usage 1

## Guidelines
- Guideline 1

Optional subdirectories extend the skill:

• scripts/ — Executable code (Python, Bash, etc.)
• references/ — Documentation loaded as needed
• assets/ — Templates, fonts, icons used in output

Progressive Disclosure

This is the core design principle. Rather than dumping all context upfront:

1. Level 1 — Frontmatter only: Claude sees skill name + description (a few tokens)
2. Level 2 — SKILL.md body: If relevant, Claude reads the full instructions
3. Level 3 — Supporting files: Only loaded when the task requires them

This is why skills scale better than stuffing everything into a system prompt. An agent can have dozens of skills registered but only pay the token cost for the ones actually used.

Key Skill Categories

Document Skills (Source-Available)

The production skills that power Claude's native document capabilities:

• docx — Word document creation and editing
• pdf — PDF extraction and generation
• pptx — PowerPoint creation
• xlsx — Excel spreadsheet generation

Example Skills (Apache 2.0)

Open-source demonstrations across domains:

• Creative & Design — Art, music, design workflows
• Development & Technical — Web app testing, MCP server generation
• Enterprise & Communication — Branding, communications, workflows

What's New Since February 2026

• Repo growth: ~73k → ~149k stars; ~17.6k forks; last push June 9, 2026 (GitHub API).
• Open standard governance: The spec, published December 18, 2025, now lives at agentskills.io with open development on GitHub (agentskills/agentskills) and a community Discord (agentskills.io).
• Cross-vendor adoption: As of June 2026, the official client showcase lists ~40 skills-compatible products — notably OpenAI Codex, GitHub Copilot, VS Code, Cursor, Gemini CLI, JetBrains Junie, Goose, OpenCode, Amp, Factory, Kiro, Mistral Vibe, Spring AI, Databricks Genie Code, and Snowflake Cortex Code (agentskills.io).
• Enterprise controls and partner directory: Anthropic added central admin settings for Team/Enterprise plans (provision/default-enable skills per org) and launched a directory of partner skills from Atlassian, Canva, Cloudflare, Figma, Notion, Ramp, and Sentry (AI Business).
• Tooling: Claude Code 2.1.0 (January 2026) added skill hot-reloading, making skills the primary mechanism for teaching Claude repeatable workflows (The New Stack).

Platform Support

Cautions

• "Just markdown" skepticism: A vocal contingent of developers argues skills are a design pattern, not a technology — packaging prompt text and calling it a standard (Hacker News).
• Supply-chain risk: Skills execute bundled scripts; malicious skills targeting Claude Code users have already been reported and discussed in early 2026 (Hacker News). Treat third-party skills like any unaudited dependency.
• Standards-capture concern: Analysts frame agentskills.io as Anthropic's "next bid to define AI standards" after MCP — adopters are betting Anthropic stewards it neutrally (The New Stack).

What Developers Say

From the launch discussion on Hacker News (October 2025):

"Claude Skills are awesome, maybe a bigger deal than MCP." — simonw (Simon Willison)

"Skills are cool, but to me it's more of a design pattern/prompt engineering trick than something in need of a hard spec." — pants2

"Fundamentally you're getting hyped over a framework to append text to your prompt?" — rafaelmn

"It may be conceptually simple, but that doesn't mean it isn't novel." — pseudosavant

Analysis

Agent Skills formalize something the agentic engineering community has been doing informally — giving agents structured instructions in files. What Anthropic adds is:

1. A standard format (SKILL.md with frontmatter) that's simple enough to adopt
2. Progressive disclosure as a first-class concept, not an afterthought
3. Platform integration across CLI, web, and API surfaces
4. A marketplace model via Claude Code plugins and a partner directory

The question this profile originally posed — whether agentskills.io gains cross-vendor adoption or remains Claude-specific — has been answered decisively. With OpenAI Codex, GitHub Copilot, Cursor, and Gemini CLI all shipping support (agentskills.io), skills are repeating the MCP playbook: an Anthropic-originated format becoming the ecosystem default.

Bottom Line

Recommended for any team building with coding agents: skills are now the portable, vendor-neutral way to encode repeatable workflows, and writing them costs almost nothing (it's markdown in folders). Not recommended as a moat or a product in itself — the format's simplicity means there is no defensible technology here, and unaudited third-party skills carry real supply-chain risk (Hacker News). Outlook: Agent Skills has crossed from Claude feature to de facto cross-vendor standard in under nine months; expect spec governance, skill registries, and security scanning to be the battlegrounds through late 2026.