Learn

Skills as Open Standard

In December 2025, Anthropic published Skills as an open standard. The same SKILL.md format that powers Claude Code now works across multiple AI coding tools.

Cross-Platform Compatibility

The SKILL.md format is supported by:

PlatformSupportNotes
Claude CodeFullNative, original implementation
CursorFullSame SKILL.md format in project directory
Gemini CLIFullGoogle’s CLI reads .claude/skills/
Codex CLIFullOpenAI’s CLI reads .claude/skills/
Antigravity IDEFullReads skills from project directory

This means you write a skill once, and it works everywhere. A team using different editors gets the same automated workflows.

How It Works

The standard defines:

  1. SKILL.md format - YAML frontmatter (name, description, triggers) + markdown body (instructions)
  2. Discovery location - .claude/skills/ directory in the project root
  3. Progressive disclosure - frontmatter always loaded, body loaded on match, linked files on demand
  4. Frontmatter contract - required fields (name, description), optional fields (license, compatibility, metadata)
---
name: deploy-service
description: >
  Deploys services to Kubernetes with safety checks.
  Use when user says "deploy", "ship it", "push to staging".
---

# Deploy Service

1. Run pre-deploy checks: `scripts/pre-deploy.sh`
2. Build container: `docker build -t $SERVICE .`
3. Push to registry: `docker push $REGISTRY/$SERVICE`
4. Apply manifests: `kubectl apply -f k8s/`
5. Verify rollout: `kubectl rollout status deployment/$SERVICE`

This skill works identically in Claude Code, Cursor, Gemini CLI, and Codex CLI.

Organization-Level Skills

Shipped in January 2026: admins can deploy skills workspace-wide.

What this enables:

  • Centralized management of team workflows
  • Automatic updates pushed to all team members
  • Consistent behavior across the organization
  • No manual installation per developer

How it works:

  • Admin configures skill sources (GitHub repos, internal registries)
  • Skills are synced to all team members automatically
  • Users see organization skills alongside personal skills
  • Admins can enforce, suggest, or make skills optional

Implications

For Tool Authors

Build one skill, reach users across all major AI coding tools. No platform-specific adaptation needed.

For Teams

Standardize workflows regardless of which AI tool individual developers prefer. A team with Claude Code, Cursor, and Gemini CLI users all get the same PR review workflow, the same deployment process, the same code generation patterns.

For the Ecosystem

Skills create a portable layer of AI workflows above the model and tool layer. MCP standardized how agents connect to tools. Skills standardize how agents execute workflows. Together, they form a complete stack:

Skills     = What to do (workflows, processes)
MCP        = How to do it (tools, resources, data)
Model      = The intelligence (reasoning, generation)

Relationship to Claude-Specific Features

While the SKILL.md format is cross-platform, some features are Claude-specific:

  • Hooks in frontmatter (PreToolUse, PostToolUse, Stop) - Claude Code only
  • Agent/subagent spawning - depends on the platform’s agent capabilities
  • MCP server references - depend on the platform’s MCP support

Write skills for the common denominator, and add platform-specific enhancements via the compatibility field:

---
name: my-skill
description: Works everywhere
compatibility: >
  Hooks require Claude Code 2.x.
  MCP features require a platform with MCP support.
---

Next