A skill is the careful balance of deterministic utilities with non-deterministic intelligence.
Today we are releasing the Agent Skills Marketplace, an open-source catalogue of portable agent skills that we maintain at the Institute.
We are releasing 11 skills across three plugins (under Apache License), and all of them install into both Claude Code and Copilot CLI.
Why Agent Skills Are Gaining Popularity
The reason why agent skills are gaining popularity is because most of the capability that people are building with coding agents today is ephemeral. It lives in a prompt someone pasted into a chat window, in a paragraph buried in a project instruction file, or in the head of the one engineer who worked out how to make the agent do the thing reliably. It can't be reviewed, it can't be copied to another team, and it usually disappears with the next tool change.
A skill writes that same capability down as a folder with a SKILL.md file at its root. The file holds a short description and a procedure. The agent matches the description against the task in front of it, and once it matches, it follows the procedure. Because the skill is a file in version control, it can be read in a pull request, corrected in public, and copied to another team.
Our objective with the marketplace is to make that the normal way capability gets shared, for some utilities that we have found helpful ourselves.
The Same Argument, One Level Higher
Since 2017 we have argued that systems making consequential decisions should be inspectable by the people affected by them. That argument has produced the Nine Principles for AI Alignment & Safety, over 30 policy contributions, and a set of open-source projects that give the principles something to run on.
Agent skills are the same argument applied, but one level higher in abstraction. The instructions we hand to agents are now a real part of how software gets built. We believe those instructions deserve the same treatment we ask for everywhere else: written down, versioned, reviewed, and open to correction.
That's the mandate we operate the marketplace under. It's a place where a working procedure can be published, validated and improved by anyone who uses it, and it doesn't compete with the assistants it installs into.
The Mechanics: Skills, Plugins and Validation
We kept the mechanics deliberately small, as we don't aim for this to be a complete list of skills, but a curated list.
Skills and Plugins
A skill is a directory containing a SKILL.md. The frontmatter carries a name and a description, and the body carries the procedure. The description is doing real work here, as the agent matches it against the task. A vague description means the skill never fires, and an overreaching one means it fires when it shouldn't.
A plugin bundles one or more related skills, and it's the thing an assistant actually installs. We didn't add any packaging layers beyond that.
Installing from Either Client
You register the marketplace once and then install plugins from it by name. Claude Code takes the slash-command form:
/plugin marketplace add EthicalML/agent-skills-marketplace
/plugin install dev-utilities@agent-skills-marketplaceCopilot CLI takes the shell form and reads the same catalogue from the same repository:
copilot plugin marketplace add EthicalML/agent-skills-marketplace
copilot plugin install dev-utilities@agent-skills-marketplaceWe also wrote INSTALL.md in the repository to do the prerequisite check, the registration and the installation for you, in case you'd rather hand the whole thing to your assistant.
Validation Before Review
We make sure that every change runs scripts/validate.sh in CI. The script checks manifest structure, skill frontmatter, skill references, naming and plugin freshness. A submission either satisfies the format or gets told exactly why it doesn't, before a human reviewer spends time on it.
What We Are Releasing Today
Each of the 11 skills is a tool we use in our own day-to-day work, across this site's repository, the KAOS project and the marketplace itself.
| Plugin | Skill | What it does |
|---|---|---|
dev-utilities | writing-skills | Writes or revises a SKILL.md so it reads as executable instructions rather than prose. |
dev-utilities | writing-agents-md | Cuts a repository's agent instruction files down to what every session actually needs. |
dev-utilities | explain-code-walkthrough | Turns a commit, pull request, diff or git range into a succinct walkthrough of what changed and why. |
dev-utilities | standardize-agent-instructions | Consolidates a repository's agent instruction files so one set works across Copilot, Claude Code, Codex, Cursor and Gemini. |
codebase-automations | dependabot-fix | Diagnoses and fixes one failing Dependabot pull request end to end, and merges it only when the checks prove it safe. |
codebase-automations | dependabot-fix-all | Processes every open Dependabot pull request on autopilot, one isolated fix at a time. |
codebase-automations | release-repo | Scores a repository's release readiness and executes a semantic version release end to end. |
workflow-automations | site-capture | Records a scripted screen capture of a site as video or GIF, with human-paced scrolling, clicks and a visible cursor. |
workflow-automations | create-agent-harness | Builds a skill-driven Python agent harness with deferred capability loading against any OpenAI-compatible endpoint. |
workflow-automations | create-streamlit-app | Scaffolds a local Streamlit and Polars data app with pluggable file, SQL, HTTP or snapshot datasources. |
workflow-automations | verify-streamlit-app | Drives a headless browser against a running Streamlit app to confirm it genuinely renders, from a manual loop to a pytest suite. |
The first of those is worth calling out because it's the skill that governs the rest. writing-skills encodes our opinionated view of what a skill should look like, and we have written it up in full in How to write an agent skill.
standardize-agent-instructions is the one that tends to surprise people. Most repositories that have seen more than one agent tool now carry three or four instruction files, and those files have diverged from each other. The skill collapses them into a single source of truth with symlinks, and that removes an entire category of "the agent ignored the rules" bug.
Contributing Your Own Skills
Contributions arrive as pull requests against the repository. Please run scripts/validate.sh before you open one, and read CONTRIBUTING.md for what the validator expects and what a good skill looks like.
Our bar for a new skill is that it should be a procedure you actually run, portable across the supported runtimes, and short enough that someone can read it and disagree with it. If you have one of those, we want it.
- Browse the marketplace on the site.
- View the repository on GitHub.
- Read the install guide if you want your assistant to set it up.