Local skill manager for AI-assisted development environments.
skillbase is a Node.js CLI that empowers you to manage skills for your AI agents using symbolic links (symlinks) or local installations per workspace, preventing context saturation.
When developing with AI assistance (using tools like Cursor, Copilot, or custom agents), having a global directory with hundreds of skills in .agents/skills/ destroys model precision:
- Hallucinations: The LLM gets confused by having too many irrelevant tools at its disposal.
- Token Consumption: Sending the manifest of hundreds of skills in every prompt is slow and expensive.
- Lack of Focus: One project doesn't need the same skills as another.
skillbase maintains a single global registry on your machine (~/.skillbase/skills) and allows you to install only what you need in your current project.
Imagine you're working on the frontend of an e-commerce site. No need for the AI to see DevOps or Backend tools; you only need your React skills and maybe some SEO skills. With skillbase, you give your agent exactly that context.
Install the package globally:
npm install -g @ariasbruno/skillbase
# Or clone and link locally:
# npm link1. Initialize your project Detect technologies and suggest compatible skills:
skillbase init2. Add necessary skills
skillbase add seo-analyzer
# Or open the interactive selector:
skillbase add3. Install from manifest
If you already have a skillbase.json, recreate the environment:
skillbase install| Command | Alias | Description |
|---|---|---|
skillbase ls [-g] |
l [-g] |
List skills in local project. Use -g for global registry. |
skillbase init [--hard] |
Detect technologies and suggests skills (use --hard for deep analysis). |
|
skillbase add [<skill>] [-s] |
a |
Install global skill. Without name, opens interactive selector. -s for symlinks. |
skillbase install |
i |
Install from skillbase.json. Supports -r (remote) and -f (force). |
skillbase install <ref> -r |
i -r |
Install remote skill (URL or GitHub). Requires -k <name> if it's a Git repo. |
skillbase remove [<skill>] [-g] [-a] |
rm |
Remove skill from project. Without name, opens interactive selector. Use -a to remove all, or -g for global registry. |
skillbase check [-r] |
c |
Check for updates. With -r, searches only in remote sources. |
skillbase update [<skill>] [-r] [-f] |
up |
Update one or all skills. -r for remote, -f to force. |
skillbase migrate [-p] [-y] |
m |
Migrate (~/.agents) or promote local skills with -p. Use -y for auto-mode. |
skillbase config [sources|autoupdate|lang] |
cfg |
Configure skill sources, auto-updates, and language. |
skillbase lang <en|es> |
Change CLI language (alias for config lang). |
-h,--help: Show detailed help.-s,--sym: Create a symbolic link instead of copying files (useful for development).-r,--remote: Indicates that the operation should consult external sources (GitHub or skills.sh API).-f,--force: Ignore "already exists" errors and overwrite files/configurations.-k,--skill: Name of the specific skill to extract when installing from a GitHub repository.-g,--global: Operate on the global registry (forlsandremove).-y,--yes: Perform automated migration without prompts (installs and overwrites everything).-a,--all: Removes all installed skills within a specific context (used withremove).
For speed, you can use initials:
l (ls), a (add), i (install), rm (remove), c (check), up (update), m (migrate), cfg (config).
skillbase organizes your tools efficiently:
Your Computer
βββ ~/.skillbase/skills/ <-- (Your physical global registry)
β βββ seo-analyzer/ <-- (Real source code)
β βββ react-helper/
β
βββ /Projects/my-great-app/ <-- (Your current workspace)
βββ skillbase.json <-- (Project manifest)
βββ .agents/skills/ <-- (Clean context for the AI)
βββ seo-analyzer/ <-- (Link or local copy)
You can change the global registry location by setting the SKILLBASE_HOME environment variable.
Install skills directly from GitHub repositories:
skillbase install <repo-url> --remote
# Example:
skillbase install https://github.com/user/my-skills --remote --skill analyzerContributions are welcome! If you have ideas to improve context management, open an Issue or a Pull Request.
MIT
