Declare the Agent Skills you want; let Skills Deck reconcile the rest — right inside VS Code.

Skills Deck demo

English

Skills Deck holds a single list of the skills you want, diffs it against what is actually installed on disk, and surfaces only the gaps — what to install, what to remove, what is out of sync across Agents.

Features

Want vs. installed — keep "do I want it" (Wanted / Unwanted) separate from "is it there" (Installed / Missing), so the actions you need are always clear.
One list, many Agents — maintain the declaration once and sync it to Claude Code, Codex, Pi, OpenCode, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Roo, Kilo, Kiro, Continue, Goose, Trae, Amp, or a custom Agent.
Single source of truth — all skills live in ~/.agents/skills; enabled Agents get managed symlinks, so removing a link never deletes the skill and switching Agents costs you nothing.
Bring skills from anywhere — add them from GitHub, skills.sh commands, skillhub.cn, or local directories.
Multi-skill repositories — discover the skills inside a repo, then select, update, or remove them per skill or per repository.
Native VS Code UI — TreeView grouping, search, multi-select, context menus, Details, tooltips, notifications, with the heavy lifting in the background.
Per-Agent diff — spot missing links, broken links, local overrides, and skills an Agent installed on its own. Read-only reconciliation — your local content is never overwritten.
Skill notes — jot down personal evaluations or reminders right from Details.
Quick navigation — jump to a GitHub repo, the local source directory, SKILL.md, an Agent's skills directory, or data.json in one click.
Portable state — desired state lives in a compact data.json; actual install state is always derived from disk, never stored twice.

Quick Start

Open Skills Deck from the Activity Bar.
Configure enabled Agents in the Agents view.
Use Add Skill or Sync to Data (Install State → List) to populate your declaration list.
Use Sync from Data (List → Install) to reconcile the global library and enabled Agents.

Remote operations run through the skills CLI in the background. Local sources use symlinks. Project-local skill management is currently disabled to keep the global model predictable.