Skills Manager

Skills Manager is a desktop application for developers who use skills across several AI coding agents and want a central library instead of manually maintaining many folders. It organizes imported and local skills, presets and per-agent deployments. Supported tools still need their own installation and configuration; this manager does not include a paid AI account or execute every skill merely by listing it.

Library and deployment

Import supported skill sources, tag and inspect them, choose presets and deploy the selected skills to agent or project workspaces. The publisher describes a shared core for the GUI and command-line interface, update tracking and deployment ownership checks. Skills are instructional files and may include scripts or external dependencies. Review the source, commands and rights before enabling them for an agent that can access your machine.

Preset membership, update tracking and deployment state are maintained centrally. Moving the library or changing an agent path can affect links and existing deployments. Recent fixes protect folders that no longer match a recorded deployment and preserve nested imports. Keep backups and inspect a dry run before modifying an important existing skills setup; a manager is not a guarantee that third-party instructions are safe.

Installation and Mac compatibility

The official skills-manager_1.40.3_aarch64.dmg is 19,977,412 bytes. Read-only inspection confirms 1.40.3 and an arm64 executable. Its plist declares 10.15, but the main Mach-O deployment command specifies macOS 11.0. The selected ARM binary therefore uses the 11.0 deployment floor, not a claim that Apple-silicon hardware runs Catalina. Separate Intel artifacts exist in the official releases and must be selected independently.

Download the matching Mac release and follow its installation workflow, or use brew install --cask skills-manager. The publisher says releases from 1.29.0 onward are Developer ID signed and notarized. No additional formula dependency is declared by the cask; npm, Rust and build instructions in the repository are development routes rather than universal requirements for the packaged desktop app. Individual skills and Git operations may still require their own tools.

Optional backup and synchronization

The Backup page can version the library in a Git repository. The recommended GitHub device-flow sign-in creates a private backup repository, while advanced setup can use another HTTPS or SSH Git URL. The README states that the token is held in the OS keychain and that ordinary metadata excludes configured secrets and machine wiring from backup.

The backup includes skills, tags, presets and per-agent toggles; that means a secret accidentally written into a skill file is still content to review before a push. Skills larger than 100 MB remain local and are excluded by the documented workflow. Automatic synchronization can commit, push and merge changes after editing. Disconnecting a machine, revoking authorization and deleting a remote backup are different operations. Keep an independent copy before a restore or migration.

Permissions, accounts and network

Local library work does not require purchasing an AI subscription. Source downloads, update checks and remote backup require network access; GitHub or another Git service account is needed only for the selected remote workflow. Access is needed to the library and chosen agent or project directories so the manager can write skills or links. Keychain authorization applies to configured credentials. No workflow described here requires granting all macOS privacy permissions by default.

The app publishes its bundled CLI for selected agent workflows. CLI commands can change the same database, files and deployments as the GUI, so authorize agent access deliberately. A skill may request extra software or accounts; those are the skill’s own requirements and not proof that the manager supports or trusts every instruction.

License, fees and risk

The project is MIT licensed. The app’s license does not change licenses for imported skills or entitlements for Claude, Codex, Cursor or other services. Remote hosting and AI-provider charges remain separate. Review updates before deploying changed instructions widely and keep credentials out of skill content and public logs. Filesystem links, copies, concurrent processes and Git conflicts can affect availability; verify the actual deployed result after migration or restore.

Sources: Official README, Exact releases, MIT license.