The Agent Skills open standard today, and the 2026 research on agents that write their own skills.
TL;DR: In late 2025, "Agent Skills" became a thing — a dead-simple way to teach an AI agent a task: a folder with a SKILL.md file (some instructions in Markdown). It's already an open standard. The wild part is what's coming next: agents that write their own skills. I built a demo where an agent solves a task the hard way once, saves a real SKILL.md, and then reuses it — cutting its total effort almost in half. ~130 lines, no API key.
First, what's a "skill"?
If you've used Claude Code or similar tools lately, you've probably seen SKILL.md files. The idea is refreshingly low-tech. A "skill" is just a folder with a Markdown file that says how to do something:
---
name: csv-to-markdown
description: Turn comma-separated text into a Markdown table. Use when the input looks
like CSV and the user wants a table.
---
# CSV to Markdown
## Instructions
Split the text into rows on newlines and columns on commas. Make the first row the
header, add a `---` divider row, then format every row as `| a | b | c |`.
That's it. No SDK, no config. Anthropic introduced this in October 2025 and then published it as an open standard (): the only required fields are name (1–64 chars, lowercase-with-hyphens, and it must match the folder name) and description (≤1024 chars, saying what it does and when to use it). Everything else — license, metadata, compatibility, allowed-tools — is optional. That's the whole spec. The SKILL.md files my demo writes follow it to the letter, so they'd load unmodified in any compatible CLI.
The clever trick: progressive disclosure
Here's the smart part. If you just dumped 50 skills' worth of instructions into the agent's context, you'd fill it up and leave no room for actual work. So skills load in stages:
Always loaded: just thenameand one-linedescriptionof every skill (tiny).
Loaded only when it matches: the full instructions, once a task actually needs them.
Loaded only if referenced: extra files or scripts the skill bundles.
So the agent can have hundreds of skills installed and barely pay for it — it only reads the short descriptions until one matches, then pulls in the details. My demo shows the math: to use 1 skill out of 3 installed, loading everything costs ~1500 "tokens"; the SKILL.md way costs ~560. That gap gets huge as your library grows.
This is also why people say skills and MCP are teammates, not rivals: MCP is how an agent connects to tools; a skill is how an agent knows the procedure for using them.
The frontier: agents that write their own skills
Today, humans write SKILL.md files. The 2026 research is about agents that write their own — and get better over time as their skill library grows. This goes back to Voyager (2023), an agent that played Minecraft and saved working code as reusable skills, getting dramatically faster at the game. The new wave makes it general:
(2026) stores skills as Markdown files that double as the agent's evolving memory, and turns task failures into new skills automatically.
, and the open standard at and — Anthropic's CLI, where the format started.
— Block's open-source agent.- Plus Cursor, GitHub Copilot, and 30+ others.
Write the skill once, use it everywhere. The future bit my demo points at: instead of you hand-writing that file, the agent writes it for itself after solving the task the first time — and from then on, your repo quietly accumulates a library of skills your agent earned.
The 10-second version (my demo)
Same stream of 7 tasks. "Cost" is how much effort each one took.
| No-skills agent | Skill-writing agent | |
|---|---|---|
| What it does | re-solves everything from scratch | learns a task once, saves a SKILL.md, reuses it |
| Total cost | 35 | 19 |
| Both correct? | 7/7 | 7/7 |
[5] csv-to-markdown learned it and wrote SKILL.md
[5] slugify learned it and wrote SKILL.md
[1] csv-to-markdown reused skill 'csv-to-markdown' ← cheap now
[5] extract-emails learned it and wrote SKILL.md
[1] slugify reused skill 'slugify'
[1] csv-to-markdown reused skill 'csv-to-markdown'
[1] extract-emails reused skill 'extract-emails'
It writes real SKILL.md files into a ./skills folder you can open. The first time it sees a task it pays full price; after that, it finds its own saved skill and reuses it for cheap.
Why this matters
Two big reasons engineers should care:
Agents stop repeating themselves. Right now most agents re-derive the same thing over and over, paying for it every time. A skill library means "figure it out once, then it's free" — like a teammate who writes things down instead of relearning them daily.
A whole new ecosystem. There are already 65,000+ shared skills and a scramble to build "the npm of agent skills" — registries and marketplaces where you install a skill like a package. Skills are becoming a unit of shareable expertise: a senior engineer's know-how, packaged in a folder, that any agent can pick up.
Tools tell an agent what it can do. Skills tell it how to do things well — and soon, agents will write that part themselves, and trade it with each other.
Try it
git clone https://github.com/Shridhar-2205/living-software
cd living-software/06-agent-skills
python demo.py
cat skills/csv-to-markdown/SKILL.md # a skill the agent wrote itself
Honest note: this is a POC. Real systems decide when a new skill is worth saving, test it, and refine it over time (that's exactly what the 2026 papers above tackle). Mine keeps that part simple so the core idea — learn once, save a SKILL.md, reuse it — is easy to see.
Written by **Shridhar Shah, Senior Software Engineer at Outshift by Cisco — I work on AI agents, search, and how they "think." Part 6 of "Toward Living Software."
Sources: Anthropic, "Equipping agents for the real world with Agent Skills" (2025) and the Agent Skills open standard (agentskills.io); Voyager (arXiv:2305.16291); MUSE-Autoskill (arXiv:2605.27366); Memento-Skills (arXiv:2603.18743); Skill-Pro (arXiv:2602.01869); MemSkill (arXiv:2602.02474).
SOCIAL SHARE CARD GENERATOR