📚
Skill packs
Bundles of slash-callable workflows. superpowers is the canonical example.
A plugin packages skills, agents, MCP servers, and configuration so a whole workflow installs in one command. Field guide to what's worth installing — and what to think about before you do.
A skill is one workflow. A plugin is a collection of skills, agents, MCP servers, hooks, and settings — packaged together so a teammate (or a fresh laptop) gets the whole kit with one command.
SKILL.md into ~/.claude/skills/<name>/. If you have a dozen workflows, an agent or two, and an MCP server you all want shipped together, that's a plugin.Two paths: the official marketplace built into Claude Code, or third-party registries like claudepluginhub.com.
/plugin. Outside it, the public hub is claudepluginhub.com./plugin install <name>. Some plugins ship as npm packages and use npx <package> install instead.~/.claude/plugins/cache/<marketplace>/<name>/. The installed-plugin index is at ~/.claude/plugins/installed_plugins.json — open it once so you know what's actually there.Ranked by install count from the public hub, May 2026. Numbers are directional — they capture momentum, not quality. Read past the rank.
The flagship skill pack. TDD cycles, git-worktree discipline, plan generation, systematic debugging, brainstorming, parallel-agent dispatch. Each skill is a checklist Claude is required to follow when the situation matches. Single biggest source of installed skills on the network.
300+ specialist agents covering JavaScript, Python, Rust, mobile, infra, and tooling. The "everything bagel" approach — install once and you have a sub-agent for almost any task. Heavy, but powerful when you do not want to build your own roster.
50+ design systems with curated font pairings, color palettes, and component templates for React, Vue, and Flutter. Reaches for a polished look without you having to write the brief. Frontend-leaning teams gravitate to it.
Cost optimizer. Forces terser responses, aggressive sub-agent delegation, and trimmed context. Reported ~75% token reduction on long sessions. Worth a look if your monthly Claude spend is the line item you keep staring at.
Browser automation from inside the CLI — navigation, form fill, scraping, screenshotting. Use it for QA flows, lightweight web research, or pulling data from sites that have no API.
Pull tickets, update status, and comment on issues without leaving the editor. Closes the loop between "I just shipped this" and "the ticket is now Done." First-party Anthropic integration.
500+ service integrations under one umbrella — Gmail, Slack, GitHub, Notion, Drive, and on. The right pick when you want one plugin to authorize against many apps instead of installing a separate connector per service.
Long-term session memory and preference persistence. Auto-captures and compresses session context into a local SQLite + vector store, then injects it into future sessions. Powerful — but see the caution below if you already have a memory workflow you trust.
Headless browser automation tuned for QA — full Playwright API exposed as Claude tools. Use for end-to-end testing, regression checks, or visual diff workflows.
Fast, reliable web fetching for agents that need to read live pages. Returns clean markdown instead of raw HTML. Pairs well with research and content-pipeline workflows.
Bundles of slash-callable workflows. superpowers is the canonical example.
Pre-loaded with palettes, fonts, and component patterns. frontend-design, ui-ux-pro-max.
Pre-built specialist sub-agents. everything-claude-code ships hundreds.
Wire external services in. Linear, Slack, Gmail, Drive, custom internal APIs.
Bias toward terseness, sub-agent delegation, and cache-friendly patterns. caveman.
Persist context across sessions automatically. Claude-Mem is the well-known one.
Plugins currently installed across the WholeTech network of sites. Light footprint on purpose.
v5.1.0. The TDD, debugging, planning, and brainstorming skills are load-bearing — invoked dozens of times a day across content pipelines, security scans, and feature work.
Invoked for any new UI: hero sections, page rebuilds, components. Strong design taste baked in, which keeps the network of sites distinctive instead of generic.
Everything else — the session-close checkpoint, the warranty-claim playbook, the news-fetch pipeline — is a hand-written skill in ~/.claude/skills/, not a plugin. Custom skills win when the workflow is specific to one operation.
A plugin’s value shows up in what it ships. Each row below is a live site on the WholeTech network plus the plugin combo that produced its visual register. The design page has a richer gallery if you want to study the surfaces themselves — this section maps each surface back to the toolchain.
Stack: frontend-design for the visual brief, superpowers/brainstorming for the section outline, caveman for token discipline during the build pass. Cream-paper card surfaces, oversized display headlines, athletic palette. Owner-confirmed brilliant and live the same day — 2026-04-27.
Stack: frontend-design with explicit reference to luxury-RE editorial vocabulary; ui-ux-pro-max as a font-pairing seed. Dense typography, calm authority. Zero-keyword-competition Bergeron spoke, paired against thebergerongroup’s navy/gold hub.
Stack: frontend-design + an Artifacts iteration round. Calculator UI built from scratch in the same session. Aimed at retiree search intent — design vocabulary chosen to feel reassuring rather than aspirational.
Stack: frontend-design with a hand-fed period-correct screenshot reference. CSS scanlines, monospace stack, real database underneath. Sister to the austinspring BBS archive — same vocabulary, modern web app.
Stack: frontend-design for the hero + Provenance section, superpowers/plan for the public-vs-private split, session-close custom skill for B2 mirror + audit log on every change. Cormorant Garamond display + Source Serif body on cream paper. Two surfaces: public service-pitch + auth-gated CRM.
Stack: frontend-design with a Regency-literary-magazine reference. connect-apps for affiliate-tag wiring (single Amazon associate id). Storefront layout that reads like a curated bookshelf, not a shop.
Stack: frontend-design + a Flask-handler co-build. Audit form on the right rail submits to a private admin dashboard. Demonstrates the “site + backend + admin” build pattern in one Claude Code session, ~3 hours start to finish.
Stack: same scaffolding as atxaisites with a swapped brand.json. frontend-design regenerated the hero against a Hot Springs / Lake Hamilton reference, leaving the data shape identical. Sibling-site pattern — one codebase, two visual registers.
Stack: frontend-design with explicit instructions to break from the sister site (girlhoop.com’s reference register). Cover-story-grade hierarchy, serif display, oversized hero quotes. The two basketball sites split editorial vs. database labor.
Stack: frontend-design for the parent visual system, then a templated rollout across 25 niche subdomains (lake / luxury / retire / airbnb / equestrian / barndo / hunting / etc.) via brand.json swap. The strongest demonstration of one-Claude-session-becoming-network-scale design.
Stack: frontend-design for the dashboard layout, Playwright plugin earlier in the same network for the security-fleet baseline that fed columns. CSV-backed live audit of all 187 sites. The dashboard register — dense tabular, monospace metrics, cyan-accent dark theme — is a different vocabulary than any of the marketing sites.
Stack: frontend-design with three print-ready style references (gig poster, almanac, art deco). Browser-print, full-bleed, three distinctly different visual languages on the same hub. Demonstrates that the design plugin scales to print, not just web.
frontend-design for the visual brief and superpowers (or a hand-written skill) for the workflow scaffolding. The leaderboard plugins above are useful, but the meaningful design diversity comes from how Claude is briefed inside frontend-design, not from stacking more plugins on top.Not every popular plugin is right for every setup. The clearest example, drawn from this network's own install log:
The setup. A hand-curated memory system already exists — an indexed MEMORY.md with topic files for user profile, project state, feedback, and references. It is paired with a session-close skill that runs a 9-step checkpoint at the end of every working session: inventory uncommitted work, security-scan, push to the droplet, mirror to two backup locations, verify the live site, update the index.
The temptation. Install Claude-Mem, because the leaderboard says it is popular and "long-term memory" sounds like an upgrade.
The actual outcome, predicted before install. Two memory systems would compete for the same job. Claude-Mem auto-compresses session context into a local SQLite + vector store and injects summaries into future sessions. The hand-curated system already does this — but transparently, with the operator in control of what gets persisted. Running both would mean duplicate artifacts, conflicting context injection, and uncertain precedence on collisions.
The decision. Skip the install. Keep the system that is auditable.
If a skill set is worth sharing, package it. The marketplace accepts plugins from any GitHub repo with the right manifest.
plugin.json manifest at the root, skills in skills/, agents in agents/, MCP config in mcp.json./plugin install /path/to/repo./plugin install <name> for everyone else.