skills-manager
Skills Manager
You manage skills using the npx skills CLI. Always use -y (yes flag) to skip interactive confirmation prompts so commands run non-interactively.
Commands
Find skills
npx skills find <query>
Returns a list of matching skills with install counts and URLs. Parse the output: skill identifiers look like owner/repo@skill-name.
Install a skill
# Global (user-level, available across all projects) — the default
npx skills add <owner/repo@skill> -g -y
# Project-level (only current directory)
npx skills add <owner/repo@skill> -y
# Install to a specific agent
npx skills add <owner/repo@skill> -g -a claude-code -y
# Install all skills from a repo
npx skills add <owner/repo> --all -g -y
List installed skills
npx skills list -g # global skills
npx skills list # project-level skills
npx skills list -g -a claude-code # filter by agent
Remove a skill
npx skills remove <skill-name> -g -y # global
npx skills remove <skill-name> -y # project
Check for updates
npx skills check
Update all skills
npx skills update
Typical workflows
User asks to find and install a skill:
- Run
npx skills find <relevant query> - Show the user the results (name, install count, URL)
- Ask which one they want, or if the best match is obvious, suggest it
- Install with
npx skills add <identifier> -g -y - Confirm what was installed and where
User asks to remove a skill:
- If they named it, run
npx skills remove <name> -g -y - If unsure of the exact name, first run
npx skills list -gto find it
User asks what skills they have:
Run npx skills list -g and present the results clearly.
User asks to update skills:
- Run
npx skills checkto show what's outdated - Run
npx skills updateto update everything, or ask if they want to be selective
Notes
- Default to global (
-g) install unless the user is clearly working on a project that should have its own skill set - The
findcommand output is non-interactive when a query argument is given — it prints results and exits - Skill identifiers use the format
owner/repo@skill-name(e.g.,vercel-labs/agent-skills@commit) - After installing a skill, it becomes available in the current session only after restarting Claude Code (let the user know)
More from trtmn/agent-skills
self-improvement
Run the self-improvement agent to review this session and the ~/.learnings/ log files. Use this skill whenever the user explicitly asks to review learnings, promote entries to CLAUDE.md, do an end-of-session review, or analyze GitHub PRs/issues for recurring patterns. Also use when the user says "promote", "review learnings", "what have we learned", or "self-improvement". Do NOT use this skill just for logging — logging happens automatically without the skill (see Passive Logging below). This skill is specifically for the *review and promotion* workflow.
20cowsay
Generates an ASCII cow saying custom text. Use when the user wants "cowsay", "cow say", or a cow to say something.
12unifi-api
Query and control a UniFi network using the `unifi` CLI (a restish wrapper with 1Password auth) or the REST API as fallback. Use this skill whenever the user wants to manage their UniFi network — listing connected clients, blocking/unblocking devices, managing firewall policies, checking WAN health and speed test results, rebooting devices, managing VLANs or SSIDs, reading traffic stats, port forwarding, or any other UniFi network management task. Prefer the `unifi` CLI for Integration API endpoints; fall back to raw curl/python for legacy API endpoints. Trigger even if the user doesn't say "API" or "UniFi" — phrases like "check my network", "block that device", "show me who's connected", "add a firewall rule", "what's my WAN IP", "how's my internet speed", or "what's on the guest network" are all good triggers.
5homebrew-dev
Package and distribute macOS apps, fonts, CLI tools, and arbitrary files using Homebrew formulas and casks. Use this skill whenever the user wants to create a Homebrew formula or cask, set up a personal tap, package a macOS .app bundle, distribute fonts or pre-built binaries via brew, use `brew create`, bump a formula or cask to a new version, submit a package to homebrew-core or homebrew-cask, or publish anything with Homebrew — even if they just ask how to "make something installable with brew", "share my app through Homebrew", "update my formula", or "get my package into Homebrew".
5tailscale-policy-manager
>
5obsidian-cli
>
4