Every AI Coding Agent That Supports SKILL.md in 2026

Most AI coding agents still can't read a standardized skill definition file. You copy-paste documentation, explain contexts repeatedly, and hope the model remembers what you need three prompts later.

SKILL.md changes this. It's a markdown specification that lets AI agents understand what they can do, how to do it, and when to use specific capabilities. Instead of explaining your project structure every session, you write it once in a format every compatible agent reads.

The adoption happened faster than expected. What started as an experiment in 2024 became the de facto standard by late 2025. Now in 2026, nearly every major AI coding agent supports SKILL.md files.

Claude Code (Anthropic)

Claude Code leads SKILL.md adoption. Makes sense since Anthropic helped design the initial specification. The agent reads SKILL.md files from your workspace root automatically.

claude-code --skill-file ./backend.skill.md --task "optimize database queries"

Works with nested skill files too. You can have separate SKILL.md files in subdirectories, and Claude Code merges them intelligently. The context window handles up to 50 skill definitions simultaneously.

Best part: Claude Code respects the constraints section religiously. If your SKILL.md says "never modify migrations directly," it won't. Other agents sometimes ignore constraints when under pressure.

Cursor

Cursor integrated SKILL.md support in version 0.42. The implementation focuses on real-time skill discovery while you code.

When you open a file, Cursor scans for nearby SKILL.md files and suggests relevant skills in the sidebar. Type a comment like // need to add authentication and it automatically references your auth skills without prompting.

# Cursor reads from .cursor/skills/ by default
mkdir .cursor/skills
cp auth.skill.md .cursor/skills/

The autocomplete gets scary good with SKILL.md files. It knows your coding patterns, preferred libraries, and project constraints. Sometimes it suggests solutions before you finish typing the problem.

GitHub Copilot

Copilot added SKILL.md support reluctantly. Microsoft wanted to push their own format initially, but developer pressure forced compatibility.

The integration feels bolted-on compared to native implementations. Copilot reads SKILL.md files but doesn't surface skills proactively. You need to reference them explicitly in comments.

# Using the database-migration skill from SKILL.md
def migrate_user_table():
    # Copilot now knows your migration patterns
    pass

Still works well for code completion. Once Copilot reads your skills, the suggestions improve dramatically. Just don't expect the elegant workflow you get with Cursor or Claude Code.

Codex CLI

OpenAI's command-line tool supports SKILL.md through the --skills flag. Simple but effective for batch operations.

codex --skills ./api.skill.md --input src/ --output optimized/

Codex CLI excels at applying skills across multiple files. Want to refactor an entire codebase using your established patterns? Point it at your skills directory and let it work.

The tool respects skill dependencies too. If skill A requires skill B to be applied first, Codex CLI handles the ordering automatically.

Gemini CLI (Google)

Google's entry supports SKILL.md but adds their own extensions. The gemini-skills section lets you define Google Cloud-specific capabilities.

# In your SKILL.md file
gemini-skills:
  - cloud-functions
  - vertex-ai-integration

Gemini CLI reads standard SKILL.md files perfectly. The extensions only matter if you're deep in the Google ecosystem. For most developers, it behaves identically to other agents.

Performance is solid. Gemini CLI processes large skill files faster than Copilot but slower than Claude Code.

Smaller players making moves

Several newer tools built SKILL.md support from day one. Aider reads skills naturally and applies them during git-aware refactoring. Cody by Sourcegraph treats SKILL.md files as first-class documentation for enterprise codebases.

Replit's AI agent supports skills but calls them "templates." Same concept, different naming. The agent reads SKILL.md files and converts them internally to their template format.

What SKILL.md actually contains

A SKILL.md file defines what an AI agent can do in a specific context. Think of it as a contract between you and the model.

The specification includes several key sections:

Purpose: What this skill accomplishes
Context: When to use it
Inputs: What data the skill needs
Outputs: What it produces
Constraints: Hard limits and requirements
Examples: Sample usage patterns

Creating your first skill takes about 10 minutes. Most developers start with a simple database interaction skill or API integration pattern.

Skills vs other approaches

SKILL.md files differ from MCP servers in important ways. MCP servers provide live data and external integrations. Skills define repeatable coding patterns and project-specific knowledge.

You might use an MCP server to fetch current stock prices. You'd use a skill to define how your application handles financial data validation and storage.

Rules files overlap with skills but serve different purposes. Rules constrain behavior globally. Skills provide positive capabilities for specific tasks.

Compatibility gotchas

Not every agent implements SKILL.md identically. Claude Code and Cursor handle skill inheritance differently. If you define a base skill and extend it, test the behavior across your toolchain.

File naming matters. Some agents look for SKILL.md, others accept skill.md or *.skill.md. Stick with uppercase SKILL.md in your project root for maximum compatibility.

Version differences exist too. The SKILL.md specification is on version 1.2, but some agents still implement 1.0. Check your agent's documentation for supported features.

The momentum keeps building

SKILL.md adoption accelerated because it solves a real problem. Developers waste enormous time re-explaining context to AI agents. A standardized skill format eliminates most of that friction.

The specification continues evolving. Version 1.3 will add skill composition and dependency management. Version 2.0 might include executable skill validation.

For now, pick an agent that supports SKILL.md well and start writing skills for your common coding patterns. The initial investment pays off within days as your AI interactions become more precise and productive.