Ready for builders
17From the tool companies themselves, or from publishers with a public track record, with clean or explained audit ratings. Read the commands and risk lines on each one.
Ready for buildersSuperpowers
Prime Radiant
A full working method: brainstorm, written plan, test-first coding, sub-agent work, code review, then finish the branch.
- Runs on
- Claude Code (official marketplace), Codex, Cursor, Gemini CLI, OpenCode
- Licence
- MIT
- Audits
- test-driven-development: Pass/Pass/Pass
- Commands
- Yes. A SessionStart hook injects its bootstrap; repo has hooks/ and scripts/. A logo request reports the version; SUPERPOWERS_DISABLE_TELEMETRY turns it off
- Risk
- Low; read the hook first
- Notable
- ~316k stars; installs via superpowers@claude-plugins-official; tdd skill 275.6K installs
- Vendors
- Partly: Claude Code plan mode, /code-review, /verify
Ready for buildersMatt Pocock's skills
Matt Pocock
Interview-style planning (grill-me), specs and tickets, TDD, bug diagnosis, code review, hand-offs.
- Runs on
- Claude Code (official marketplace), Codex, Copilot, Gemini CLI, others
- Licence
- MIT
- Audits
- grill-me: Pass/Pass/Pass; tdd: Pass/Pass/Pass
- Commands
- Mostly instructions; reads the codebase. One setup command per repo
- Risk
- Low
- Notable
- ~321k stars; holds about 20 of the top 100 named places on the skills.sh all-time chart; grill-me 1.5M installs (first seen 13 Mar 2026)
- Vendors
- Partly: Claude Code /code-review, plan mode
Ready for buildersAgent Skills
Addy Osmani
24 lifecycle skills (spec, plan, build, verify, review, ship) plus a router skill, reviewer personas and checklists.
- Runs on
- 70+ agents via npx skills; Claude Code marketplace; Cursor, Gemini CLI, Codex and others
- Licence
- MIT
- Audits
- frontend-ui-engineering: Pass/Pass/Pass
- Commands
- Yes: lifecycle hooks for Claude Code; 13 validation and eval scripts
- Risk
- Low. Not a Google product: the site never mentions Google, and the README draws on Google's public engineering references
- Notable
- ~115k stars; GitHub Trending Oct 2026
- Vendors
- Partly, as above
Ready for buildersfrontend-design
Anthropic
Picks a distinct visual direction before coding and avoids stock fonts, palettes and layouts.
- Runs on
- Any SKILL.md tool; Claude Code
- Licence
- Open (anthropics/skills)
- Audits
- Pass/Pass/Pass
- Commands
- No
- Risk
- Low
- Notable
- 1.1M installs (first seen 19 Jan 2026); the most-installed AI-vendor builder skill
- Vendors
- Is the vendor skill
Ready for builders/design (Claude Code)
Anthropic
Give it a brief and Claude publishes a canvas of editable UI artboards as a Claude Design artifact; pick one, adjust it, then have Claude implement it. Also screen flows, landing pages and posters; /design-sync brings in your design system.
- Runs on
- Claude Code CLI and Claude Code Desktop; needs a claude.ai login and the Anthropic API
- Licence
- Built in
- Audits
- Not applicable (built in)
- Commands
- Built in; publishes artifacts to your claude.ai account
- Risk
- Launched as a research preview; the docs still call templates beta. Not available on Bedrock, Google Cloud or Foundry, or where CMEK, HIPAA or zero data retention is on. On Enterprise an owner must enable the Design template
- Notable
- Announced in Anthropic's Claude Code digest for 17–21 Aug 2026 (Pro, Max, Team and Enterprise); brings Claude Design's artboard workflow into the coding tool. The standalone claude.ai/design closes 14 Dec 2026
- Vendors
- Is the vendor feature
Ready for buildersweb-artifacts-builder
Anthropic
Scaffolds and bundles multi-component Claude artifacts (React, Tailwind).
- Runs on
- Claude, Claude Code
- Licence
- Open
- Audits
- Pass/Warn/Pass
- Commands
- Yes: init-artifact.sh and bundle-artifact.sh
- Risk
- Low; the Socket warn is unexplained, but the scripts are the stated purpose
- Notable
- 118.1K installs
- Vendors
- Is the vendor skill
Ready for buildersLive artifacts and Claude Data
Anthropic
Cowork live artifacts are dashboards and trackers that refresh from your connected apps and files each time they open. Claude Data (formerly Claude Dashboards), made from a claude.ai chat or the Artifacts tab, connect to a data warehouse or CRM, answer plain-language questions and keep the dashboard current, with each number linked to its query.
- Runs on
- Claude Desktop (Cowork) and claude.ai; paid plans
- Licence
- Built in
- Audits
- Not applicable (built in)
- Commands
- Built in; reads connected apps
- Risk
- Sharing of connector-backed artifacts is limited to your organisation. Data dashboards are off by default on Enterprise until an owner enables them
- Notable
- Launched in beta as Claude Dashboards on 8 Oct 2026; renamed Claude Data and generally available on all paid plans from 10 Nov 2026, alongside Claude Sheets and Whiteboards (beta). Connects to Redshift, BigQuery, ClickHouse, Databricks, Snowflake or Salesforce
- Vendors
- Is the vendor feature
Ready for buildersweb-design-guidelines
Vercel
Audits UI code against Vercel's Web Interface Guidelines (design, accessibility, UX) and reports file:line findings.
- Runs on
- Any SKILL.md tool
- Licence
- Open
- Audits
- Pass/Warn/Pass
- Commands
- Fetches the latest guidelines from Vercel's GitHub before each review
- Risk
- Remote content each run (from Vercel's own repo)
- Notable
- 846K installs
- Vendors
- Partly: Lovable /accessibility
Ready for buildersFigma skills (figma-use and 23 more)
Figma
Let an agent edit real Figma files with your components and variables, generate diagrams, link code to components.
- Runs on
- Claude Code, Codex, Cursor (via the Figma plugin/MCP)
- Licence
- Downloadable from GitHub
- Audits
- figma-use: Pass/Pass/Pass (10 Nov 2026)
- Commands
- Through Figma's MCP server
- Risk
- Some skills need paid plans or a Full seat; figma-spec is in open beta. Copies of figma-use exist from other publishers; install only Figma's own
- Notable
- use_figma MCP tool entered beta 24 Mar 2026
- Vendors
- Is the vendor skill
Ready for builderswebapp-testing
Anthropic
Tests a local web app in a real browser with Playwright scripts, starting servers as needed.
- Runs on
- Claude Code; any SKILL.md tool with Python
- Licence
- Open
- Audits
- Pass/Warn/Warn
- Commands
- Yes: Python Playwright scripts; with_server.py starts servers
- Risk
- The warns match the purpose (it runs scripts); same pattern as the docx/xlsx teaching example
- Notable
- 202K installs
- Vendors
- Partly: Claude Code /verify, /run
Ready for buildersPlaywright CLI skill
Microsoft
Teaches agents to drive a browser via playwright-cli, which uses fewer tokens than the Playwright MCP server.
- Runs on
- Claude Code, Copilot, Cursor, any skills agent
- Licence
- Open (Apache 2.0, Playwright)
- Audits
- Pass/Pass/Warn
- Commands
- Yes: installs @playwright/cli from npm; runs a browser
- Risk
- Use the scoped @playwright/cli, not the old unscoped playwright-cli package
- Notable
- 212.9K installs (first seen 23 Jan 2026)
- Vendors
- Is the vendor skill; OpenAI's playwright skill wraps the same CLI
Ready for buildersdeploy-to-vercel
Vercel
Deploys a site to Vercel; preview by default, production only on explicit request.
- Runs on
- Claude Code, Codex, claude.ai (no-auth fallback)
- Licence
- Open
- Audits
- Pass/Pass/Warn
- Commands
- Yes: git and Vercel CLI commands
- Risk
- Needs a Vercel login for the normal path
- Notable
- 188.3K installs (first seen 6 Mar 2026)
- Vendors
- Is the vendor skill
Ready for buildersSupabase skill
Supabase
Database, auth, row-level security, storage and edge functions, with a security checklist.
- Runs on
- Any SKILL.md tool; Supabase MCP
- Licence
- Open
- Audits
- Pass/Pass/Warn
- Commands
- Yes: Supabase CLI for schemas and migrations; optional MCP for live queries
- Risk
- Can change a live database; use a dev project
- Notable
- 361.7K installs (first seen 5 Apr 2026); companion Postgres skill 480.6K
- Vendors
- Is the vendor skill
Ready for buildersskill-creator
Anthropic
Interviews you, drafts a SKILL.md, then tests it with and without the skill and shows the comparison.
- Runs on
- Claude, Claude Code; OpenAI and Supabase publish their own versions
- Licence
- Open
- Audits
- Pass/Warn/Pass
- Commands
- Yes: a Python review viewer; background test runs
- Risk
- Low
- Notable
- 428.8K installs
- Vendors
- Lovable /skill-creator; OpenAI skill-creator
Ready for buildersLovable Skills
Lovable
Save instructions as SKILL.md skills; prebuilt /redesign, /accessibility, /SEO-review; upload skills or export them as .zip.
- Runs on
- Lovable
- Licence
- Built in
- Audits
- Not on skills.sh (built in). Lovable validates imports but states no security review; Enterprise has an audit log
- Commands
- Built in
- Risk
- Admins and owners manage workspace skills; plan gating not stated
- Notable
- Launched 18 May 2026
- Vendors
- Is the vendor feature
Ready for buildersReplit Agent skills
Replit
Installs skills into a project's /.agents/skills from Replit's skill picker (Replit and partner skills, audited by Replit) or from skills.sh (community, not audited).
- Runs on
- Replit Agent
- Licence
- Built in
- Audits
- Not on skills.sh. Replit says skills in its picker are audited for safety; externally installed skills are not
- Commands
- Depends on the skill
- Risk
- Vet any skill you add from outside Replit's picker
- Notable
- Docs current Nov 2026; no launch date shown
- Vendors
- Is the vendor feature
Watching: read before installing
9Popular or fast-rising builder packs that show where things are going, but that install a lot, send data elsewhere, or raise terms-of-service and IP questions. Not recommendations.
Watchinggstack
Garry Tan
"Virtual engineering team": office hours, CEO/design/engineering plan reviews, QA in a real browser, security review, ship and deploy, retros.
- Runs on
- Claude Code; Codex, Cursor, Gemini CLI via flags
- Licence
- MIT
- Audits
- gstack (root): Pass/Warn/Warn; review: Pass/Pass/Pass; skillify: Gen Warn
- Commands
- Heavily. ./setup builds compiled tools and a bundled Chromium browser; team mode adds a required hook and commits to .claude/ and CLAUDE.md; some skills drive your real logged-in browser (setup-browser-cookies); "outside reviews" send code to Codex/Claude by default; update checks on by default; telemetry opt-in
- Risk
- Data leaves the machine by default. Windows 11 Smart App Control can block its unsigned binaries; Windows support is "minimal"
- Notable
- ~141k stars; Product Hunt #3 of the day, 13 Mar 2026; 77 skills
- Vendors
- Yes, partly: Claude Code /verify, /run, /code-review
Watchingponytail
Publisher not named
Makes the agent reuse existing code and the platform before writing new code; adds review, audit, debt and cost commands.
- Runs on
- Claude Code, Codex, Copilot CLI, Gemini CLI, Pi, OpenCode, Devin CLI
- Licence
- MIT
- Audits
- ponytail: Pass/Pass/Pass
- Commands
- Yes: two lifecycle hooks; a status band that imports past session data into ~/.config/ponytail/
- Risk
- Benchmarks are the project's own ("53% less code"); local data only, per the site
- Notable
- ~188k stars within five months (first seen 12 Jun 2026); 406K installs across 6 skills
- Vendors
- Partly: Claude Code /simplify
WatchingFour-rule coding guidelines (CLAUDE.md)
Publisher not named
One CLAUDE.md of four rules: think before coding, keep it simple, make surgical changes, define a goal and check it.
- Runs on
- Claude Code (plugin or CLAUDE.md), Cursor rule
- Licence
- MIT
- Audits
- Sample skill: Pass/Pass/Pass
- Commands
- No scripts; install is a curl or plugin add
- Risk
- Low. A single CLAUDE.md; read it and copy the rules into your own rather than installing
- Notable
- ~232k stars; 45.7K installs (first seen 28 Jan 2026)
- Vendors
- Partly: Anthropic's best-practices page covers the same ground
WatchingECC (Everything Claude Code)
Affaan Mustafa
Very large harness: 71 sub-agents, 314 skills, hooks, memory, rules, a default Chrome DevTools connector and the AgentShield scanner.
- Runs on
- Claude Code, Codex, Cursor, OpenCode
- Licence
- MIT (core); paid GitHub App
- Audits
- security-review: Pass/Pass/Pass (first seen 19 May 2026)
- Commands
- Heavily. npx ecc-universal installer or install scripts; optional hooks; MCP connector; /astra-review sends diffs to OpenAI
- Risk
- Sprawl: the docs warn mixing install methods duplicates hooks. Sponsor and affiliate links; the README warns that third-party mirrors may carry malware
- Notable
- ~301k stars; top skill only 8.8K installs
- Vendors
- Partly
WatchingImpeccable
Paul Bakaus
Design vocabulary, 20+ commands (polish, typeset, audit), a detector for AI "slop" patterns, PRODUCT.md/DESIGN.md context, live mode.
- Runs on
- Claude Code, Codex, Cursor, Gemini CLI, Copilot, Replit Agent and others
- Licence
- Apache 2.0 (site does not state; secondary sources do)
- Audits
- Pass/Pass/Warn
- Commands
- Yes: a launcher runs a self-contained binary that ships with the skill or is downloaded on first run
- Risk
- Binary download
- Notable
- ~107k stars; 398.5K installs (first seen 8 Apr 2026); extends Anthropic's frontend-design
- Vendors
- Partly: frontend-design, /design
WatchingLinkedIn Agent skill (11 skills)
Jake Schincariol (OpusJake)
Drafts posts, comments, replies, carousels, DMs and a weekly plan; scores profiles. Drafts only; never posts.
- Runs on
- Claude Code, Cursor, Codex, Copilot, any SKILL.md tool
- Licence
- MIT
- Audits
- li-post: Pass/Pass/Pass (first seen 8 Sep 2026)
- Commands
- No posting or login. The README says browser automation breaches LinkedIn's User Agreement
- Risk
- Drafting only: skip its humanizer skill (/li-human) and never connect it to an account. LinkedIn's terms prohibit automation.
- Notable
- 3.4k stars; ~3.7K installs
- Vendors
- Partly: general writing in every assistant
WatchingInstagram Agent skill (13 skills)
Jake Schincariol (OpusJake)
Drafts reels, captions, carousels, stories, comments and DMs; plans and audits. "These skills write. You post."
- Runs on
- Same as above
- Licence
- MIT
- Audits
- ig-caption: Pass/Pass/Pass (first seen 14 Sep 2026)
- Commands
- /ig-viral reads public content through your own browser "at human speed"; never logs in
- Risk
- Drafting only: skip its humanizer skill (/ig-human) and never connect it to an account. The browser-reading skill sits close to Instagram's automation terms.
- Notable
- 5.8k stars
- Vendors
- Partly
WatchingReplica (11 skills)
Jake Schincariol (OpusJake)
Studies an existing app from public pages, screenshots and store listings, then plans, rebuilds, tests, rebrands, launches and deploys your own version.
- Runs on
- Claude Code (plugin) or skill folders
- Licence
- MIT
- Audits
- replica-recon: Pass/Pass/Warn (first seen 3 Oct 2026)
- Commands
- Yes: six standard-library Python tools (image diff, parity score, contrast check, name/colour sweep); the README says they make no network calls
- Risk
- IP, trademark and terms exposure. The README states clean-room rules: no source, assets, logos or copy; no scraping behind logins or against terms. It says "This is not legal advice" and recommends a lawyer before launch
- Notable
- 8.8k stars; ~14.3K installs
- Vendors
- No
Watchingfind-skills
Vercel
Searches skills.sh and installs other skills on request.
- Runs on
- Any agent using the skills CLI
- Licence
- Open
- Audits
- Pass/Pass/Warn
- Commands
- Yes: runs npx skills add, installing third-party code on your behalf
- Risk
- Ranks by install count and stars, which section 1 shows are noisy
- Notable
- #1 all-time on skills.sh (4.3M installs)
- Vendors
- Partly: /plugin Discover tab with "Anthropic verified" marks
Referencegetdesign.md / awesome-design-md
VoltAgent team
550+ DESIGN.md files (colours, type, spacing, components) that reproduce well-known brands' styles for agents; follows Google's DESIGN.md spec
- Note
- Trademark and trade-dress caution. It lists Apple, Stripe, Claude and others with no non-affiliation note. Use for learning, not for client sites that imitate a brand. Some paid extras ($29 templates)
ReferenceLandingfolio
Operator not named on the site
961 landing pages and 4,652 section examples; Pro tier needed for colour filters. Also offers seven free MIT agent skills and a free MCP server (section ideas, landing-page audit, DESIGN.md extraction, mobile pass, screenshots, share images)
- Note
- One of its skills, humanize-copy-pass, is a humanizer: excluded. The other skills are not audited here
ReferencePricingPages.design
Operator not named (footer "by FS")
Curated pricing-page examples, filterable by component (tier cards, FAQs, comparison tables) and pattern (usage calculators, enterprise)
- Note
- Free to browse as far as shown
ReferenceSite of Sites
Operator not named (Instagram @siteofsites)
Gallery of 613 websites filterable by industry, style and platform (Wix, Webflow, Framer)
- Note
- No AI features
ReferenceClaude Code best practices (official)
Anthropic
Official guidance: tell Claude what it can't see, give it a check to run, enforce hard rules
- Note
- The vendor version of the "best practices warehouse"
Referenceclaude-code-best-practice (community)
Publisher not named
Reading collection: 83 tips, workflows, settings, hooks, cross-model workflows
- Note
- ~68.9k stars; sponsor-supported; read it, don't install it
Referenceawesome-claude-code
Publisher not named
Hand-picked list of skills, agents, plugins and tooling
- Note
- ~57.7k stars; the page states no vetting policy
ReferencePlugin security and trust
Anthropic
What a plugin can run on your machine, and a four-step review before installing
- Note
- Pair this with every "Watching" row
Reviewed 10 Nov 2026; the shelf is checked monthly. Items marked † rest on a single source in part.
This map is a by-product of the work. Keeping track of what each of these actually does, and which ones are worth your time, is part of how I advise founder-led companies on AI. If you want that applied to your business, the first conversation is free.
Book a 30-minute conversation →
