What Are Agent Skills?

Agent Skills are a lightweight open format for giving AI agents specialised knowledge and workflows. A skill is a folder with a SKILL.md file — name, description and instructions — plus any scripts or reference material the agent needs. Agents load skills on demand, so specialised know-how travels across products without stuffing every prompt.

That definition matters more this month than it did last year. On 22 October 2026, the Agentic AI Foundation took Agent Skills as a hosted project after Anthropic donated the standard. Anthropic said more than 70 agent platforms had already integrated the format, including ChatGPT, Cursor, GitHub Copilot and VS Code, with about 1 million skills published and nearly 280 million installs. The name, logo, GitHub organisation and agentskills.io domain moved with the donation.

If you have been asking what Agent Skills are, start there: a portable procedure folder, not a new model and not another chat product.

How a skill actually loads

The Agent Skills overview describes progressive disclosure in three stages:

1. Discovery. At startup the agent loads only each skill's name and description — enough to know when it might apply.
2. Activation. When a task matches a description, the agent reads the full SKILL.md into context.
3. Execution. The agent follows the instructions, and may run bundled scripts or open referenced files.

That design keeps many skills available without burning the whole context window on day one. A typical layout looks like this:

my-skill/
├── SKILL.md          # required: metadata + instructions
├── scripts/          # optional executable code
├── references/       # optional docs
└── assets/           # optional templates and resources

Developers on X have been shipping that pattern as the AAIF news landed — for example this write-up of the donation and coverage of the AAIF project pack.

Skills vs chat prompts vs MCP

A chat prompt is a one-off instruction. A skill is a version-controlled folder you reuse and share. That is the difference that shows up when the same task has to run the same way next week.

MCP (Model Context Protocol) solves a different problem: how an agent connects to tools and data. Agent Skills package how to do a job once those tools are available. Plenty of stacks will use both. An agent harness is the runtime that loops plan → act → check; skills are the specialised playbooks that harness can pull in. For how work moves between agents, see AI agent workflow.

Why the AAIF donation matters

Before the donation, Agent Skills lived as an Anthropic-led open standard. Moving it under AAIF puts governance next to MCP and the foundation's other agent projects, with maintainers from more than one company. Anthropic's post is clear that you can keep using and contributing to the standard without joining AAIF.

For builders, the practical upside is portability: write a skill once, run it in any client that speaks the format. For buyers, the question shifts from "which vendor owns this procedure?" to "does this skill still work after we change models or tools?"

When skills are enough — and when they are not

Skills shine when you have repeatable procedures: a review checklist, a research pipeline, a formatting standard, a deployment ritual. They fail when the work is open-ended coordination across many missions, or when production software has to leave as reviewed pull requests rather than as a local folder of instructions.

If you are assembling a personal software factory, skills are one useful layer inside it — not the whole line. Factories still need sandboxes, review gates and someone who owns the merge.

If your work isn't code: a squad instead of a skill library

Most knowledge workers do not want to author SKILL.md files. They want research done, drafts written and admin cleared, with a human still on the review step.

xSquads for Work gives each knowledge worker their own squad of specialised agents to delegate to and review. The squad runs on your computer, works on your own files and folders, and moves each mission across a shared board from Backlog to To Do to In Progress to Review to Done. Bots propose, you approve, and anything irreversible is escalated to you instead of executed. It is on Mac and Windows, and free with Pro agents.

If you need production software: a managed software factory

When shipping code is the job and you would rather not build or staff the agent line yourself, use a managed software factory. On this site that offer is xSquad operating as your code factory: agents open pull requests you can read, a person reviews before code is merged, and nothing merges without your approval. You add or remove agents as the workload changes. Every output passes through quality gates — automated tests, code reviews, QA validation and human approval. The fuller description is in what a managed AI software factory does.

Which one fits you?

• Author Agent Skills if you build agents or need portable, version-controlled procedures across compatible clients.
• Use xSquads for Work if the work is research, writing, planning and admin in your own files, and you want a squad whose output waits for your review.
• Use a managed code factory if production software has to ship as pull requests a person reviews, and you'd rather not assemble the line.

Where these facts come from

Read on 4 November 2026.

• Donating Agent Skills to the Agentic AI Foundation (Anthropic, 22 October 2026)
• Agentic AI Foundation Announces Four New Projects (AAIF, 22 October 2026)
• Agent Skills overview
• Posts on X: mrfokki, mojeskoqq

Frequently Asked Questions

What are Agent Skills?

Agent Skills are a lightweight open format for giving AI agents specialised knowledge and workflows. A skill is a folder with a SKILL.md file (name, description and instructions) plus optional scripts, references and assets the agent can load when the task matches.

What is a SKILL.md file?

SKILL.md is the required file inside a skill folder. It holds metadata (at least name and description) and the instructions the agent follows. Agents usually load only names and descriptions at startup, then read the full file when a task matches.

Who donated Agent Skills to AAIF?

Anthropic donated the Agent Skills standard to the Agentic AI Foundation on 22 October 2026, including the name, logo, GitHub organisation and agentskills.io domain. Anthropic said more than 70 agent platforms had integrated the format, with about 1 million skills published and nearly 280 million installs.

How are Agent Skills different from MCP?

MCP is about how agents connect to tools and data sources. Agent Skills package procedural knowledge and workflows the agent loads on demand. Many stacks use both: MCP for connectors, skills for how to do a job.

Do I need Agent Skills if I don't write code?

Not as a format you author yourself. If your work is research, writing, planning and admin on your own files, a ready-made squad of agents fits better than assembling skill folders. xSquads for Work gives each knowledge worker that squad to delegate to and review.

When should I use a managed software factory instead?

When production software has to ship as pull requests a person reviews, and you would rather not build or staff the agent line yourself. On xsquads.ai that offer is xSquad as your code factory: agents open PRs, nothing merges without approval, and you scale agents with the workload.

Related Reading

• What is MCP and what is an agent harness.
• AI agent workflow and personal software factory.
• Managed software factory and the code factory.
• xSquads for Work.

Raina Rufus, Founder of xSquads

xSquads for Work is free. Join the waitlist and we'll send the download link.