Agent Skills Joins the Agentic AI Foundation: What's New, Where Skills Run, and Alternatives (2026)
Agent Skills now has a neutral home. On October 22, 2026, Anthropic donated the open standard to the Agentic AI Foundation (AAIF), which sits under the Linux Foundation. The format itself does not change. A skill is still a folder with a SKILL.md file that teaches an agent one task.
This guide covers what was announced and what a skill is. It then covers what was already in place, what the move means for people who use skills, and which apps run them. The last sections cover safe installation and the main alternatives. Every detail comes from official pages read on October 23, 2026.
What changed on October 22
Anthropic's post is titled "Donating Agent Skills to the Agentic AI Foundation." It describes Agent Skills as "the open standard that gives AI agents specialized knowledge and workflows."
The post lists exactly what moved. "We are donating the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain."
It also explains who runs the standard from now on. Engineers from Anthropic, Google, OpenAI, and other companies will maintain it as individuals. David Soria Parra and Mahesh Murag are the Lead Maintainers.
The Linux Foundation announced the move the same day, alongside three other projects:
- Agent Skills, contributed by Anthropic
- Open Responses, contributed by OpenAI
- OpenEnv, contributed by Hugging Face
- OpenSandbox, contributed by Alibaba
The press release says the additions bring the AAIF to 10 hosted projects. It launched with three projects in December 2025. The news came at AGNTCon + MCPCon North America, held in San Jose on October 22 and 23.
The new projects join six existing ones. They include the Model Context Protocol (MCP), AGENTS.md, goose, and the Agent2Agent (A2A) protocol.
What Agent Skills are
The Anthropic post defines a skill as "a folder containing a SKILL.md file that teaches an agent a specific task or domain." The folder can also hold any scripts or reference material the task needs.
The specification at agentskills.io sets out the layout. A skill folder needs a SKILL.md file, and it can add three optional folders:
- scripts/ for executable code
- references/ for documentation
- assets/ for templates and other resources
The SKILL.md file starts with YAML frontmatter. Two fields are required:
| Field | Rule in the specification |
|---|---|
| name | Max 64 characters; lowercase letters, numbers, and hyphens only; must match the folder name |
| description | Max 1024 characters; "Describes what the skill does and when to use it." |
Optional fields cover a license, compatibility notes, metadata, and an experimental list of allowed tools.
The description does most of the work, because it decides when an agent picks the skill. The specification gives a good example: "Extracts text and tables from PDF files, fills PDF forms, and merges multiple PDFs." Its poor example is simply "Helps with PDFs." The good one names the inputs, the actions, and the trigger words.
Agents load skills in three stages, which the agentskills.io overview calls progressive disclosure. At discovery, agents read only each skill's name and description. At activation, a matching task pulls in the full SKILL.md instructions. At execution, the agent follows them and runs any bundled scripts.
That design matters for scale. The overview says agents "can keep many skills on hand with only a small context footprint."
What was already there before
The donation changes ownership, not the format. Most of what users rely on shipped earlier.
| When | What happened, per Anthropic and the Linux Foundation |
|---|---|
| October 2025 | Anthropic launches skills while working on Claude Code |
| Following weeks | Developers publish thousands of skills on GitHub and write loaders for Codex, Cursor, and OpenCode |
| December 2025 | Anthropic publishes Agent Skills as an open standard at agentskills.io |
| December 2025 | The AAIF launches with three projects |
| By October 2026 | "More than 70 agent platforms" have integrated the standard |
| October 22, 2026 | The standard moves to the AAIF |
Anthropic also gives usage figures. "So far, 1 million skills have been published and nearly 280 million installed."
Anthropic donated MCP to the same foundation earlier. The post notes that Anthropic co-founded the AAIF with Block and OpenAI. Skills now follow the same path that MCP took.
What the move means for people who use skills
For most users, nothing breaks. Your existing skills keep working, because the file format is unchanged.
Three things do change.
Governance is shared. No single vendor controls the specification any more. The AAIF calls this "open, vendor-neutral governance."
Participation is open. The post says "You can continue to use the standard or contribute to it without joining the AAIF." Feedback on proposed changes goes through the project's Discord and GitHub.
A roadmap is coming. Anthropic proposed starting with "versioned releases, a formal proposal process, and security guidance for people who write and install skills." Versioned releases should make it easier to know which agents support which features.
One caution stays the same. The VS Code documentation notes that "supported locations and optional features vary by agent." A skill is portable in format, but each app decides where it looks for skills.
Where Agent Skills run today
The Anthropic post names ChatGPT, Cursor, GitHub Copilot, and VS Code among more than 70 platforms. The table below covers apps whose own documentation describes skills support, read on October 23, 2026.
| App | Where skills work, per its documentation |
|---|---|
| Claude | Free, Pro, Max, Team, and Enterprise plans, with code execution turned on; also Claude Code and the API |
| ChatGPT and Codex | ChatGPT desktop app, Codex CLI, and the IDE extension; skills bundled in plugins also work across ChatGPT |
| GitHub Copilot | Copilot in VS Code, Copilot CLI, the Copilot app, and Copilot cloud agent |
| Cursor | Agent in the Cursor editor, from skill directories |
| Powerdrill Bloom | Built-in agent skills on Free; custom agent skills and Claude Skills on Pro |
Claude. The help center says "Skills are available for users on Free, Pro, Max, Team, and Enterprise plans." You add your own in Customize, then Skills. Anthropic also ships its own skills for Excel, Word, PowerPoint, and PDF files.
ChatGPT and Codex. OpenAI's guide says skills "build on the open agent skills standard." It adds that "Standalone skills are available in the ChatGPT desktop app, Codex CLI, and IDE extension."
GitHub Copilot. GitHub's docs describe skills as folders "that Copilot can load when relevant to improve its performance in specialized tasks."
Cursor. Cursor's docs say "When Cursor starts, it automatically discovers skills from skill directories and makes them available to Agent." You can also call a skill by typing a slash in Agent chat.
Our Claude Skills API guide covers using skills from code.
How to write and install a skill safely
The steps below follow the specification and Claude's help center. Other apps use different folders, so check their docs for the install location.
- Create a folder. Name it in lowercase with hyphens, such as
quarterly-report. - Add SKILL.md. Write the name and description in the frontmatter. The description should say what the skill does and when to use it.
- Write the instructions. Use plain steps in Markdown. Add scripts or templates only if the task needs them.
- Install it. In Claude, upload it under Customize, then Skills. In Codex or Cursor, place it in a skill directory.
- Test it on one task. Ask for the job the description names, and check that the agent picks the skill.
Treat skills from others like software. Claude's help center names prompt injection and data exfiltration as the main risks. It advises installing skills only from trusted sources. For skills from a less-trusted source, it says to read the bundled files before enabling, with attention to code dependencies.
On the Enterprise plan, Claude can scan third-party skills for malicious content before they run. Anthropic says most scans finish in one to two minutes.
What to watch next
The donation is a starting point. Four things are worth tracking over the coming months.
Versioned releases. Today, the specification has one current version. Numbered releases would let an app state which version it supports. That would make the "supported locations and optional features" gap easier to read.
Security guidance. The roadmap names guidance "for people who write and install skills." Skills can run scripts, so this part matters most for teams that share skills widely.
Distribution through plugins. OpenAI already packages skills inside plugins. Its guide says plugins "distribute reusable skills and connectors through the universal plugin directory shared by ChatGPT and Codex." Expect more apps to ship skills inside larger bundles.
Team rollout. On Claude Team and Enterprise plans, Owners can provision skills for every member. The help center says these "appear automatically in every team member's skills list." Shared, approved skills are likely to become the normal way companies use them.
If you write skills, keep each one focused on a single job. A short, specific description works better than a long, general one.
Agent Skills alternatives
Skills are one of several ways to give an agent context. Most of these complement skills rather than replace them.
| Option | What it gives the agent | Best for |
|---|---|---|
| Agent Skills | Procedures, scripts, and templates loaded on demand | Repeatable tasks across many agents |
| Model Context Protocol (MCP) | Connections to outside tools and data | Live access to apps and databases |
| AGENTS.md | One instruction file for a code repository | Build steps and coding conventions |
| Claude Projects | Background knowledge loaded in every chat of a project | Stable reference material |
| Custom instructions in VS Code | Coding standards and guidelines | Style rules for one team |
MCP. Claude's help center draws the line clearly. "MCP connects Claude to external services and data sources." Skills, in contrast, carry the procedure. You can use both together.
AGENTS.md. This is another AAIF project. Its site calls it "A simple, open format for guiding coding agents, used by over 60k open-source projects." It works like a README written for agents.
Projects. Claude's help center says "Projects provide static background knowledge that's always loaded when you start chats within them." Skills load only when a task needs them.
Custom instructions. The VS Code docs say custom instructions "primarily define coding guidelines," while skills can include scripts and resources.
Where Powerdrill Bloom fits
Most skills today live in coding tools and general assistants. A growing number target analysis work, such as monthly reports or market research. Our list of agent skills for analysis and reporting covers examples.
Powerdrill Bloom is an AI workspace built for that kind of work. Its plans list skills support directly. The Free plan includes "Use built-in agent skills." The Pro plan adds "Run built-in and custom agent skills" and "Use Claude Skills to run specialized AI workflows for research, analysis, automation, and execution."
Skills sit inside workspaces too. The changelog says "Tune each workspace with its own skills and MCP servers to match exactly how you work." Each workspace also keeps files in memory across chat sessions.
For recurring work, the Scheduled Agent runs a task daily, on weekdays, weekly, or monthly. The AI Report Generator and Excel to PPT pages show the finished outputs.
On price, the pricing page lists Free at $0.00 with 1 scheduled task. Pro is $13.27/mo, with 20 scheduled tasks and Claude Skills support.
To see a skill run on your own data, try Powerdrill Bloom with one spreadsheet you report on every month.
Frequently asked questions
What is the Agentic AI Foundation?
The Agentic AI Foundation (AAIF) is a Linux Foundation group that hosts open projects for AI agents. It launched in December 2025 with three projects. After the October 22, 2026 additions, it hosts 10, including MCP, AGENTS.md, and Agent Skills.
Did Anthropic give away Agent Skills?
Anthropic donated the standard to the AAIF on October 22, 2026. The donation covers the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain. Engineers from several companies now maintain the specification as individuals.
Do my existing skills still work?
Yes. The donation changes who governs the standard, not the file format. A skill is still a folder with a SKILL.md file. Install locations and optional features still vary by app, so check each app's documentation.
Which apps support Agent Skills?
Anthropic says more than 70 agent platforms have integrated the standard, including ChatGPT, Cursor, GitHub Copilot, and VS Code. Claude, Codex, and Powerdrill Bloom also list skills support on their own pages.
What is the difference between Agent Skills and MCP?
MCP connects an agent to outside tools and data sources. Agent Skills give the agent procedures, such as the steps and templates for a monthly report. The two work together, and both are now AAIF projects.
Sources: Anthropic, Donating Agent Skills to the Agentic AI Foundation · Linux Foundation, AAIF announces four new projects · Agent Skills overview · Agent Skills specification · Claude Help Center, What are skills? · Claude Help Center, Use skills in Claude · OpenAI, Build skills · GitHub Docs, Agent skills · VS Code, Use Agent Skills · Cursor Docs, Agent Skills · AGENTS.md. Details read from official pages on October 23, 2026.