What Are Agent Skills? Reuse One SKILL.md Across AI Tools

Are you still copying the same instructions into a separate config file for Claude, Codex, and Gemini CLI every time you want one of them to handle a task your way? The more tools you add, the more versions of those instructions drift apart, and a fix made in one place quietly leaves the others behind.

Agent Skills is an open standard that teaches an AI agent a workflow through a single SKILL.md file. Anthropic published it in December 2025 and, on October 22, 2026, donated it to the Agentic AI Foundation (AAIF), part of the Linux Foundation. According to the announcement, more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, support it. This guide covers how a skill is structured, what the move to a neutral foundation means, how to reuse one skill across tools, how to write descriptions that trigger reliably, a security checklist, and how daglo meeting transcripts can feed a skill.

What are Agent Skills? One SKILL.md that teaches an agent a task

An agent skill is a folder containing a SKILL.md file and any supporting material. The agent knows only each skill's name and description until a request matches, then reads the full instructions and follows them.

• SKILL.md: YAML frontmatter with a name and description at the top, followed by step-by-step instructions in Markdown. This file alone is a valid skill.
• scripts/: Code the agent can run. Anything that must produce the same result every time is better handled by code than by prose.
• references/: Templates, glossaries, and longer docs the agent reads only when needed.
• Progressive loading: Only names and descriptions load up front, so adding more skills barely increases what the agent reads at the start.

If you are new to building skills in Claude Code, start with our Claude Code skills guide. This article focuses on sharing the same skill across tools.

What the move to the Linux Foundation's AAIF means

Moving the standard from one company to a neutral foundation means the skills you write depend less on any single vendor's decisions. The name and logo, the GitHub organization, and the agentskills.io domain all moved to the AAIF.

The AAIF took on other projects the same day, including OpenAI's Open Responses, bringing it to ten projects in its first year. MCP, the standard for connecting agents to external tools, is also an AAIF project, which makes the split clearer: MCP for tool connections, skills for workflows. For how these compare with ChatGPT's building blocks, see our ChatGPT plugins vs. MCP and skills article.

Build a SKILL.md and use it across tools

Create one skill folder, then place it where each tool looks for skills. Here is a hands-on example: a skill that turns a meeting transcript into a follow-up email draft.

1. Create the folder: Name it after the skill and add SKILL.md, a template, and an optional script.
2. Write SKILL.md: Use lowercase letters and hyphens for the name, matching the folder name. Keep the description under 1,024 characters.
3. Place it per tool: Copy or link the same folder into each tool's skills directory.
4. Test it: Send each tool a request that matches the description and compare whether the skill triggers and the output matches.

meeting-followup-email/
├── SKILL.md
├── references/email-template.md
└── scripts/extract_actions.py

---
name: meeting-followup-email
description: Reads a meeting transcript and drafts a follow-up email for attendees. Use when asked for a meeting follow-up or recap email.
---
1. Find decisions, action items with owners, and next dates in the transcript.
2. Draft the email using references/email-template.md.
3. Never add anything that is not in the transcript; mark gaps as [needs check].
4. Do not send the email. Show the draft only.

Keep a single source of truth (for example, .agents/skills) and point the other tools' folders to it with a symlink. One edit then reaches every tool. Supported tools and folder locations change often, so check the client list on agentskills.io and each tool's official docs. To bundle several skills and share them with a team, see our Claude Code plugins guide.

Writing descriptions that trigger, plus a security checklist

Agents decide whether to use a skill based on its description alone, so say what it does and when to use it. Before installing someone else's skill, read every file in the folder yourself.

How to write the description

• Lead with the output: "Drafts a follow-up email from a meeting transcript" is clearer than "Helps with meetings."
• Use the words people actually type: Phrases like "write the follow-up email" help the agent match requests.
• Draw the boundary: Note when not to use it, such as "for full meeting notes, use meeting-summary."

Security checklist

• Check what each script in scripts/ runs and whether it sends data to outside addresses.
• Read the instructions for hidden directions such as "ignore previous instructions."
• For anything hard to undo, like sending email, deleting files, or deploying, have the skill stop at a draft so a person approves it.
• Never put passwords, API keys, or customer data in SKILL.md or reference files.
• Use skills only from sources you trust, and reread the changes after every update.

We cover attacks that hide instructions for an AI in more depth in our prompt injection defense guide.

Feed daglo meeting transcripts into a follow-up email skill

Skills work best when their input has a consistent shape. When you transcribe a meeting with daglo, you get a transcript that shows who said what, plus a summary, which is exactly what a follow-up email skill needs.

• Record and transcribe: Record the meeting in daglo or upload a recording, and you get the full transcript with a summary.
• Names, not numbers: daglo's Speaker ID remembers the voices of the people in your meetings and fills in their real names instead of "Speaker 1" from the next recording on. daglo is the first in Korea to offer this to every user. With names in place, the skill assigns each action item to the right owner.
• Hand it to the skill: Export the transcript as text into your working folder and ask, "Write the follow-up email for this meeting." Any tool that supports the skill produces a draft in the same format.
• Check against the source: Ask Board Chat to "list only the decisions and owners" and compare it with the draft to catch anything missing.

If you want every meeting summarized the same way, save that instruction in daglo's My Templates. daglo remembers the prompt and applies it automatically each time a recording ends, so the input your skill receives always has the same shape.

In short, Agent Skills is a shared standard for teaching many AI tools the same workflow with one SKILL.md, and the move to the Linux Foundation gives it a vendor-neutral home. Build one source skill, connect it to each tool, and refine it with a clear description and the security checklist. Supported tools and details may change, so check agentskills.io before you start.

#AgentSkills #SKILLmd #ClaudeCode #Codex #GeminiCLI #LinuxFoundation #AIAutomation #daglo

👉 Transcribe your meetings free