Diagnose a slow home network end-to-end and recommend the highest-leverage fix. Use when the internet or Wi-Fi feels slow — "make my home network/Wi-Fi faster", "speeds are way below my plan", "slow at my desk", "should I buy an extender / new router / mesh?", "bufferbloat", "lag on calls", «зроби домашню мережу швидшою», «інтернет повільний», «повільний Wi-Fi». Runs a repeatable playbook: measure from the machine (throughput, idle-vs-loaded latency/bufferbloat, packet loss, Wi-Fi signal & negotiated rate), localize the bottleneck across four layers (ISP plan → router → Wi-Fi path → client), inspect the router admin UI, then rank fixes — wired backhaul first, a second wireless extender never. Not for public/office networks you don't administer, and not for server-side app latency (that's profiling, not this).
Write, structure, and review great Agent Skills (SKILL.md files) so an agent reliably triggers and follows them. Covers the description/frontmatter that decides triggering, the three kinds of skill (knowledge / task / workflow), progressive-disclosure structure, when to bundle scripts/references/assets, named anti-patterns, how to eval a skill, and portability across both Claude and Codex. Use when creating, writing, editing, improving, reviewing, shortening, or debugging a skill / SKILL.md / plugin skill — when a skill won't trigger, is too long, or must work on both hosts. Triggers include «як написати скіл», «створити/покращити скіл», «чому скіл не тригериться», "write a skill", "make a SKILL.md", "my skill never triggers". Do not use for plugin PACKAGING — manifests, version bumps, marketplace wiring, README tables (that is plugin-dev); this skill is about the SKILL.md itself.
Build and maintain a great AI-agent skills marketplace and the plugins that live in it. Covers the git-repo layout, the Claude (.claude-plugin) + Codex (.agents / .codex-plugin) DUAL-manifest packaging that lets one repo serve both hosts, the version-bump discipline the marketplace auto-update keys off, how to add a plugin or a skill, and what makes a plugin/skill actually worth installing. Use when: creating a plugin marketplace, adding/removing a plugin, packaging a skill for Claude Code / Claude Cowork / Codex, writing or fixing a marketplace.json or plugin.json, or debugging why a newly-added plugin or skill doesn't show up after an update. Triggers include "set up a plugin marketplace", "add a plugin", "package this skill for Claude and Codex", "marketplace.json", "my new plugin isn't showing up". Do not use for writing the SKILL.md prose itself (that's a skill-authoring guide) — this skill is about the marketplace + plugin PACKAGING.
Bootstrap a NEW AI-agent skills marketplace, or bring an existing one up to date, from the canonical learn-yy-skills structure + its meta plugin — always fetched from GitHub at the LATEST version, never a copy that happens to be on this machine. Use when the user wants to "create/start my own skills marketplace", "clone/bootstrap a marketplace like this one", "scaffold a new plugin marketplace repo", or "update my marketplace's meta plugin / structure to the latest" — «створи свій маркетплейс скілів», «онови meta-плагін маркетплейсу». Runs the bundled Bun script clone-marketplace.ts (needs Bun; else run install-bun): --new scaffolds a fresh repo (both marketplace manifests + meta plugin + CLAUDE.md/AGENTS.md + README + LICENSE + git init); --update refreshes an existing repo's meta plugin + scaffolding in place, leaving your plugins untouched. Do NOT use to add one plugin/skill to a marketplace you already run (that is plugin-dev), nor to write SKILL.md prose (skill-authoring).
Install the Bun runtime — this marketplace's scripts (the meta plugin's SessionStart health hook and clone-marketplace) run on Bun + TypeScript, so a machine without Bun can't run them. Use when Bun is missing: the session-start check prints "the marketplace-health check runs on Bun, which isn't installed", a `bun` command fails with "command not found", or the user asks to "install bun" / "set up bun" / «встанови bun». Covers the official installer, Homebrew, and npm, plus verifying it and fixing PATH. Present the command for the USER to run (installing pipes a remote script to the shell) — don't run it silently. Not for installing other tools, nor for authoring skills (that's skill-authoring).
Turn ON auto-update for an installed plugin marketplace — Claude Code or Codex — so it stays current on its own. On Claude Code, set extraKnownMarketplaces.<name>.autoUpdate=true in settings.json (or the /plugin UI toggle); on Codex, git-sourced marketplaces already auto-update, so the fix for a stale one is re-adding it from its Git source or running `codex plugin marketplace upgrade`. Use when the user asks to "enable/turn on marketplace auto-updates", "keep my plugins up to date automatically", "stop my marketplace going stale", "make learn-yy-skills auto-update", or right after marketplace-health reports auto-update OFF — «увімкни авто-апдейт маркетплейсу», «хай плагіни оновлюються самі». This is the ACTION counterpart to marketplace-health (which only diagnoses); do not use it to check status, nor to author/version a marketplace (that is plugin-dev).
Check whether an installed plugin marketplace — on Claude Code or Codex — is HEALTHY — (1) the latest published version of each plugin is actually installed (not a stale cached copy), and (2) auto-update is active so it stays current (Claude Code's extraKnownMarketplaces.autoUpdate flag, or a Codex git vs local source_type). Use when the user asks "is my marketplace up to date / am I on the latest?", "why didn't my new skill/plugin show up after I pushed?", "are plugin auto-updates on?", "check marketplace health", "did my marketplace update", "is my Codex marketplace current?", or after publishing a plugin change to confirm clients will pull it — «чи оновлений маркетплейс», «чому не підтягнулась нова версія скіла», «увімкнені авто-апдейти плагінів». Do not use to AUTHOR a marketplace/plugin or bump versions (that's plugin-dev) — this only diagnoses an already-installed one.