skillshare
skillshare.runkids.cc- Category
- AI Tools
- Rank
- No. 2765Tools index
Previous survey · No. 2712 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Agent Building
- Interfaces
- CLI
- Builder
- runkids
- GitHub
- 3.0k stars
- Latest release
- v0.29.1
- Date
About
A CLI tool that provides a single source of truth for AI CLI skills, enabling seamless synchronization across multiple AI development environments like Claude, Cursor, OpenCode, and 50+ other tools. It offers bidirectional sync, security auditing, and team collaboration features for managing AI workflows.
What it does
Every AI coding tool keeps its own folder of skills, agents and rules, and copies drift apart. skillshare keeps one master folder on your machine (or one inside a project) and links or copies its contents into each tool's native location. Symlinks on macOS and Linux, NTFS junctions on Windows, with a copy mode when links break. It also installs skills from any Git host, writes one MCP (Model Context Protocol) server definition into each tool's own config format, scans skills for prompt injection before use, and can commit and push the master folder to a Git remote.
Why it's ranked here
The problem is real and gets worse with every agent you add, and skillshare treats it as declarative state rather than a pile of install commands. It ships as one Go binary with no runtime dependency, no registry and no telemetry, under an MIT license. Coverage goes past skills: agents, rules, plugins and MCP connections all route through the same source. The dashboard and the command line call the same domain code, so the two surfaces do not drift, and the repository carries a separate integration test suite and scripted end-to-end runbooks.
What's good
Dry-run exists where it matters: MCP and plugin syncs can be previewed before they touch any tool's config. Machine-readable output is treated as a contract, so JSON modes print only structured data and skip the update check. Per-target include and exclude filters decide which skills reach which tool. Hub indexes can live on private Git hosts over SSH, and the index reader refuses paths that escape the cloned repository. The local dashboard server sets read, write and idle timeouts and shuts down cleanly on interrupt.
Tradeoffs
The one-line installer pipes a tarball into tar and moves it into a system binary folder, asking for sudo if needed, without checking it against the published checksums file. Plugins sync separately and are left out of the sync-everything command, which is easy to miss. Hub search is a plain substring match sorted by name, with no relevance ranking. The dashboard reloads configuration from disk on nearly every API request, which is simple but means a broken config surfaces as a server error. Symlinks can misbehave on some setups, which is why copy mode exists.
How to use it well
Best for someone running two or more coding agents who wants one versioned folder of skills and rules. Run init, let it detect installed tools, then sync. Keep project skills in the repository so teammates get them on clone, and track a shared team repository for organization-wide ones. Run the audit before syncing anything installed from a third party. Use Homebrew, or fetch the binary yourself and compare checksums, rather than the curl pipe. It manages files; it does not write skills for you or judge whether they are any good.
Technical notes+
cmd/skillshare/main.go dispatches through a flat commands map (init, sync, audit, mcp, plugin, hub, ui and more) and runs legacy directory migrations on startup, including a Windows move to AppData. wiki/architecture.md states handlers in internal/server/server.go reuse the same internal domain packages as the CLI; that file registers routes on a ServeMux wrapped with per-request config reload (skipped for health and config endpoints) and optional base-path stripping. internal/search/index.go loads hub indexes over HTTP with a 15-second timeout, from local files, or via shallow SSH clone, rejecting absolute or parent-traversal paths. install.sh never reads the checksums that .goreleaser.yaml produces. schemas/config.schema.json defines merge, symlink and copy sync modes.
Observed
- License
- MIT
- Language
- Go, with a React and Vite web dashboard
- Packaging
- Single static binary built with cgo disabled; shell and PowerShell installers, Homebrew tap, GitHub Action
- Platforms
- Linux, macOS and Windows on amd64 and arm64
- Interfaces
- CLI, terminal UI, local web dashboard backed by an HTTP API
- Configuration
- Published JSON Schema for the global config file
- Install script
- install.sh does not verify archive checksums
- Tests
- Separate CLI integration test directory and end-to-end runbooks described in the repository docs
Read from README.md, go.mod, LICENSE, cmd/skillshare/main.go, internal/server/server.go, internal/search/index.go, wiki/architecture.md, wiki/audit.md, wiki/testing.md, schemas/config.schema.json, install.sh, .goreleaser.yaml.
What it can do
Synchronize AI CLI skills across development environments
AI CLI skills and configuration data → Synchronized skills across multiple AI tools
Manage AI workflows in a centralized repository
AI workflow definitions and skills → Centralized skill repository and management system
Perform security auditing on AI CLI skills
AI CLI skills and code → Security audit reports and vulnerability assessments
Enable bidirectional synchronization between AI tools
Skills and data from multiple AI environments → Synchronized data flows between AI development tools
Facilitate team collaboration on AI workflows
Team member permissions and shared AI skills → Collaborative workspace for AI workflow management
Connect and integrate with 50+ AI development tools
Tool configurations and authentication credentials → Integrated connections to multiple AI platforms
Tags
Tech Stack
Media

Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.