SkillBuddy Tackles the Growing Mess of AI Skills Scattered Across Agents
When a developer runs Claude Code, Codex, Cursor, and Gemini CLI side by side, the same Skill often lives in four directories with four subtly different versions. SkillBuddy replaces that manual folder-diving with a single dashboard that detects drift, syncs a trusted baseline, and keeps project-specific rules from leaking into global config — a workflow that any team adopting AI-assisted coding will need within weeks.
AI Skills have exploded from a niche concept into a sprawling ecosystem. Vue, React, and component-library teams now ship official Skills, while community efforts cover everything from photo-editorial layouts to commercial-grade HTML slide decks. The immediate problem is no longer discovery — it is management. Developers routinely run half a dozen agents, each with its own hidden Skills directory, producing duplicate copies that silently drift apart as rules are tweaked in one tool but not another.
SkillBuddy scans local agent directories, surfaces every installed Skill with its source and scope, and flags content mismatches across agents. It separates user-level, project-level, and plugin Skills, lets you bundle related Skills into reusable packs, and backs up personal configs to a private Git repo. For teams, a shared Git repository becomes the source of truth for reviewed Skills, MCP definitions, and role-based packs, keeping the existing branch-and-PR workflow.
The tool also pulls in MCP Server configs, which suffer the same fragmentation. A local-first desktop app was chosen deliberately — no account, no cloud lock-in, and Git credentials stay with the system’s own SSH agent or credential manager. The project is MIT-licensed and in public preview, with builds for Apple Silicon, Windows x64/ARM64, and Linux x64/ARM64.
Skills management is following the same arc as dependency management: once a project depends on dozens of them, ad-hoc folder copies break down and a package-manager-like layer becomes necessary.
The distinction between personal experimentation and team-audited Skills is not cosmetic — without it, teams cannot safely adopt AI-assisted workflows at scale, because unreviewed prompts become a vector for bad practice or security gaps.
SkillBuddy’s decision to stay format-agnostic and read native agent directories means it avoids creating a new walled garden; if you uninstall it, your Skills remain exactly where each agent expects them.