Browse
A quiet pill appears when a page has installable skills — GitHub repos, curated lists, marketplaces. No skills, no pill, no clutter.
Works with Claude Code, Cursor, Codex, Gemini CLI and 10+ more
Skill Installer finds skills on GitHub and skill marketplaces, shows you exactly what they’ll do, and installs them into every agent you use. One click — no terminal, no copy-paste.
Coming soon to the Chrome Web Store · Free · No account · No telemetry
One click, every major coding agent
An Agent Skill is a folder with a SKILL.md file in it — instructions that teach Claude Code, Cursor, Codex or any other compatible agent to do something specific, the way you want it done. The format is an open standard, and the ecosystem grew around it fast: curated lists, dedicated marketplaces, and skills published across GitHub by everyone from Anthropic to the person who just solved your exact problem.
Installing one has not kept up. There is no skill install command. You find something useful in a README, then you switch to a terminal, clone a repository you only wanted one directory from, copy that directory into ~/.claude/skills, and do the whole thing again for every other agent you use. Most people give up somewhere around the second step.
Skill Installer closes that gap. It notices skills on the page you are already reading, shows you what is inside one before anything touches your disk, and writes the folder into every agent you picked — pinned to the exact commit you reviewed.
Read the format explainedgit clone https://github.com/…/skills.git
cp -r skills/document-skills/pdf \
~/.claude/skills/
cp -r skills/document-skills/pdf \
~/.agents/skills/
# hope it was the version you readThree steps between “that looks useful” and a skill your agent can actually use.
A quiet pill appears when a page has installable skills — GitHub repos, curated lists, marketplaces. No skills, no pill, no clutter.
Before anything touches disk: the skill’s instructions rendered, its file tree, and security flags — bundled scripts, tool grants, external URLs.
One click writes the folder atomically into every agent you picked, pinned to an exact commit. Claude Code picks it up instantly; updates show a diff first.
Skills carry instructions — and sometimes scripts — that your agent may act on. Skill Installer treats that with the seriousness it deserves.
Nothing installs without the preview: rendered instructions, file tree, security flags. Flags inform; they never block.
Every file is fetched at a resolved commit SHA — a branch that moves mid-install can’t swap content under you.
The helper writes only inside skill folders you’ve approved, validates every path, and never marks anything executable.
No account, no telemetry. Update checks read GitHub’s public commit feeds straight from your browser.
How to install skills in each agent, what the format actually is, and where MCP servers fit alongside them.
Manual copy, npx, plugin marketplaces, one click — and how to tell which one you want.
One gives your agent knowledge, the other gives it reach. When to use which.
One directory, roughly ten agents — plus the tools that do their own thing.
A folder, a markdown file, some frontmatter — and how an agent decides to read it.
Agent Skills are folders of instructions — a SKILL.md file plus optional scripts and reference files — that teach AI coding agents to do something specific, well. The format is an open standard published at agentskills.io, and it is read by Claude Code, Cursor, Codex, Gemini CLI and dozens of other tools. Agents load a skill on demand: they see its one-line description, and pull in the full instructions only when the task matches.
Manually, you copy the skill’s folder into ~/.claude/skills for personal use, or into .claude/skills inside a repository so it travels with the project. With Skill Installer you click the pill that appears on the page you found the skill on, review the preview, and the folder is written for you. Claude Code hot-loads new skills, so there is no restart.
Claude Code (~/.claude/skills) and Cline (~/.cline/skills) have dedicated targets, and a single Universal install covers every tool reading the shared ~/.agents/skills directory — Cursor, Codex, Gemini CLI, OpenCode, Amp, Goose and GitHub Copilot among them. You can also register custom directories and per-project folders.
A skill is procedural knowledge: static markdown telling the agent how to do something well. An MCP server is a running program that gives the agent live access to a system it otherwise could not reach — a database, a private API, anything behind authentication. Skills cost almost no context until they are used; MCP tool definitions occupy the context window for the whole session. Most substantial setups use both.
Treat a skill like software: it carries instructions your agent may act on, and sometimes scripts it may run with your permissions. That is why every install here starts with a preview showing security flags — bundled scripts, tool grants, external URLs, and instructions that try to talk past you — why downloads are pinned to an exact commit SHA, and why the helper writes only inside directories you have approved. Flags inform; they never block.
A small native helper, registered once on your machine, does the writing — a browser extension cannot touch your filesystem on its own. Chrome spawns the helper per request and it exits immediately, so it is never resident. It accepts messages only from this extension, validates every path, writes atomically, and never marks anything executable.
You still get detection and the full preview. The install button becomes a copy button that produces the equivalent npx skills add command for you to run in a terminal — which already puts you ahead of copying folder paths by hand.
Not yet — skills come first. MCP servers and subagents are on the roadmap and will use the same preview-first flow. They are a genuinely different job underneath: installing a skill writes a folder, while installing an MCP server means merging an entry into a config file you did not write, without disturbing what is already in it.
Yes. It is free, there is no account, and there is no telemetry of any kind. Update checks read GitHub’s public commit feeds directly from your browser.