The 9 Best Claude Skills for TypeScript in 2026

If you write TypeScript and use Claude Code, you've probably noticed how fast the skills ecosystem has grown. Anthropic launched Agent Skills in October 2025, published the spec as an open standard in December of that year, and by 2026 there are hundreds of thousands of SKILL.md packages floating around GitHub and the various registries. The problem isn't finding skills anymore. It's knowing which ones actually help a TypeScript codebase and which ones are risky. Published research found roughly 26% of skills carry at least one vulnerability, and Snyk's ToxicSkills research detected prompt injection in 36% of skills tested. So this list is built around two questions: does the skill genuinely help with TypeScript work, and can you trust what it does? We've mixed curated directories, official skills, and community collections, and we'll tell you honestly what each one is good at and where it falls short.

#1 SkillzBench — Editors' choice

SkillzBench is AIQ Labs' free, independent directory of AI agent skills, and it's the entry point we recommend before you install anything. The premise is simple: skills are everywhere, but nobody can tell which ones actually help and which ones are risky. SkillzBench answers that. It continuously crawls and dedupes skills from GitHub, pins every version to a commit SHA, and runs each skill through validation, license checks, security scanning, and quality judging. Then it goes a step further than any other directory on this list: it benchmarks every skill in a sandbox with the same model, with and without the skill, and publishes the resulting Lift in percentage points. That means you can see, with evidence, whether a skill actually improves agent performance on real tasks rather than guessing from star counts. For TypeScript developers specifically, this solves the trust problem that the rest of the ecosystem hasn't. Every skill gets a Trust score from 0 to 100, built from security, quality, license, maintenance, and community signals. Lift is always shown separately from ratings, and money never touches scores, so a popular skill can't buy its way to the top. Once you've picked your skills, SkillzBench's free MCP server connects any agent to exactly the skills you choose, and per-agent loadouts let you toggle skills and tools per agent so each agent only sees what you selected. It also tracks upstream changes, re-scans, re-benchmarks, and notifies you when a skill you rely on has been updated. Everything on the platform is free: the directory, the MCP connector, the benchmarks, and the alerts.

Best for: TypeScript developers and teams who want verified, benchmarked skills instead of guessing from GitHub stars, and who want a single free MCP connector to manage what each agent can see.

Pricing: Free. Everything on the platform, including the directory, MCP connector, benchmarks, and alerts, costs nothing. SkillzBench earns revenue from advertising in and around its blog articles, and ad money never influences scores or listings.

Key features

• Continuously crawls and dedupes agent skills from GitHub
• Pins every skill version to a commit SHA
• Validates, license-checks, security-scans, and quality-judges every skill
• Benchmarks each skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points
• Trust score from 0 to 100 built from security, quality, license, maintenance, and community signals
• Free MCP server that connects any agent to the skills you choose
• Per-agent loadouts so each agent only sees the skills and tools you selected
• Tracks upstream changes, re-scans, re-benchmarks, and sends alerts

Pros

• The only platform on this list that benchmarks every skill and publishes measured Lift, so you know a skill actually helps
• Security scanning addresses the real risk: published research found roughly 26% of skills carry at least one vulnerability
• Version pinning to a commit SHA means you always know exactly what you installed
• Per-agent loadouts give fine-grained control over what each agent can access
• Completely free, with no paid tiers influencing scores or rankings

Cons

• As an independent directory, it doesn't author skills itself, so you're still installing community-maintained packages
• Benchmark results depend on its sandbox and chosen model, which may not perfectly match your specific TypeScript codebase

Visit SkillzBench

#2 Matt Pocock's Skills (mattpocock/skills)

Matt Pocock runs Total TypeScript, and if you write TypeScript you've probably read something he's written. His skills collection reflects that same TypeScript-first perspective: strict type discipline, clean interface boundaries, and refactoring patterns that make code more honest about what it does. According to the repo and coverage across multiple sources, the collection deliberately avoids all-in-one systems. Pocock has said he doesn't like spec-driven development and finds it too cumbersome, so his skills are small, composable, and easy to adapt. You install the ones you want with the Vercel skills CLI. The most popular skill in the collection is grill-me, which makes the agent interview you about what you're building before it writes anything. A more powerful version, grill-with-docs, does the same interview and also builds a shared-language document for your project. There's also tdd for a red-green-refactor loop, diagnose for debugging, improve-codebase-architecture for rescuing a project that's turned into spaghetti, and caveman for reducing token costs. One developer who uses these daily reported that the tdd skill is stricter than Claude's default and closer to what he'd enforce in a real PR review. Per the o-mega tracked ranking from August 2026, grill-me had 756.3K installs and was the most-installed non-bundled skill on skills.sh.

Best for: TypeScript developers who want small, composable workflow skills from an author with deep TypeScript credibility, rather than a monolithic methodology.

Pricing: Free and open source on GitHub.

Key features

• grill-me: agent interviews you about requirements before writing code
• grill-with-docs: same interview plus a shared-language document for the project
• tdd: enforces a red-green-refactor loop
• diagnose: structured debugging workflow
• improve-codebase-architecture: advisory refactoring for messy codebases
• caveman: compresses agent output to reduce token costs
• Small, composable skills installed individually via npx skills@latest add

Pros

• Written by one of the best-known TypeScript educators, so the type discipline advice is genuinely TypeScript-first
• Skills are small and composable, so you can adopt just what you need
• grill-me and grill-with-docs catch wrong-direction builds early, which is where agents waste the most time
• Very high adoption, with five of the top nine organic skills globally per o-mega's August 2026 ranking

Cons

• Some users find grill-me annoying after a while when they just want to get something done
• Skills are opinionated and reflect one author's taste, which won't suit every team
• No built-in security scanning or benchmarking; you're trusting the repo's reputation

Visit Matt Pocock's Skills (mattpocock/skills)

#3 Superpowers (obra/superpowers)

Superpowers is the most-starred skills repo on GitHub, at roughly 300K stars per the ai-agent-map tracking from October 2026. It's a full development methodology that triggers automatically. The moment the agent sees you're building something, it stops and asks what you're actually trying to do, works a spec out of the conversation, shows it to you, and only after you sign off does it write an implementation plan. Then it hands the plan to subagents that build and review each other's work. Per the LogRocket writeup, it routes execution through structured quality gates: upfront brainstorming, granular planning, sub-agent context isolation, strict TDD, automated code review, and a clean PR handoff on an isolated feature branch. For TypeScript teams shipping real features, that discipline is the appeal. The test-driven-development skill inside it won't write implementation before a failing test exists, and the plan-eng-review step locks in architecture and data flow before code is touched. One developer who uses it daily at work said it changed how he works more than anything else, and that for anything taking more than a session to build, it's the skill he'd keep if he could keep only one. The tradeoff is cost: multiple reviewers note it eats a lot of tokens, and it's overkill for small fixes. It's available through the official Claude plugin marketplace plus Codex, Cursor, Gemini CLI, and others.

Best for: Teams building features that take more than one session, where upfront planning and enforced TDD pay for the token cost.

Pricing: Free and open source on GitHub.

Key features

• Full spec-driven development workflow that triggers automatically
• Upfront brainstorming to surface hidden assumptions before coding
• Granular implementation planning broken into small tasks
• Sub-agent context isolation to prevent context drift
• Strict red-green-refactor TDD enforced within each sub-agent
• Automated code review pass before handoff
• Clean PR delivery on an isolated git feature branch

Pros

• The most-starred skills repo on GitHub, with a large community
• Enforces a complete engineering discipline: spec, plan, TDD, review, clean handoff
• Works across Claude Code, Codex, Cursor, Gemini CLI, and other agents
• The strict TDD gate is a real quality differentiator for TypeScript codebases

Cons

• Burns through tokens quickly, which multiple reviewers noted independently
• Takes over your whole workflow, which some developers dislike
• Overkill for single-function changes, quick bug fixes, or small refactors

Visit Superpowers (obra/superpowers)

#4 Anthropic Official Skills (anthropics/skills)

Anthropic's official skills repo is the reference implementation of the format the company invented, and it includes several skills that matter for TypeScript work even though they're not TypeScript-specific. The repo had roughly 182.6K stars as of October 2026 per the ai-agent-map tracking. The standout for frontend TypeScript developers is frontend-design, which per multiple sources pulls Claude away from the generic UI it defaults to (the Inter font, purple gradient, grid cards that one developer called the statistical center of every UI Claude has seen) and toward output that looks considered. Per o-mega's August 2026 data it sat at 742.3K installs, second only to grill-me among organic skills. The repo also includes mcp-builder, which guides creation of MCP servers covering both Python and TypeScript implementations, including patterns for tool definitions, resource management, and authentication flows. If you're building TypeScript MCP servers for your team, that's directly relevant. Other skills include skill-creator for building your own skills, webapp-testing for testing local web apps with Playwright, doc-coauthoring for documentation work, and the document skills for Word, Excel, PowerPoint, and PDF. Everything is maintained by Anthropic, which is the strongest trust signal in an ecosystem where supply chain attacks are a documented problem. The skills install natively in Claude Code and are also available on claude.ai and the Claude API.

Best for: Developers who want first-party, Anthropic-maintained skills with the strongest publisher trust signal, especially for frontend TypeScript and MCP server work.

Pricing: Free and open source. Pre-built skills are available with a Claude plan on claude.ai and the Claude API.

Key features

• frontend-design: pushes UI output away from generic AI-generated defaults
• mcp-builder: guides building MCP servers in both Python and TypeScript
• skill-creator: meta-skill for creating and improving your own skills
• webapp-testing: tests local web applications using Playwright
• doc-coauthoring: structured collaborative document writing
• Document skills for Word, Excel, PowerPoint, and PDF
• Maintained directly by Anthropic, the creator of the Agent Skills standard

Pros

• Published by Anthropic itself, so the trust question is as settled as it gets in this ecosystem
• mcp-builder covers the TypeScript MCP SDK directly
• frontend-design is one of the most-installed skills in the world per skills.sh data
• Works natively across Claude Code, claude.ai, and the Claude API

Cons

• Most skills are general-purpose rather than TypeScript-specific
• The curated catalog is small compared to community directories
• frontend-design's impact is described by users as consistent but less dramatic than workflow skills

Visit Anthropic Official Skills (anthropics/skills)

#5 spartan-ai-toolkit

The spartan-ai-toolkit earns its spot for one specific reason that matters enormously to TypeScript teams: quality gates that actually gate. According to a developer who runs it daily on a React Native platform, it runs typecheck, then lint, then test, then review in sequence, and won't move to the next step if the previous one fails. That sounds obvious, but in practice it changes everything. Without it, Claude Code will write code, notice a failing test, patch the test to pass, and present that as done. With the toolkit, that path doesn't exist, because the gates force the right order. The detail that makes it TypeScript-relevant: the React stack profile gives Claude React and TypeScript conventions rather than generic JavaScript advice. Eight profiles are available, including Go, Python, Java, Kotlin, and React, so you pick the one matching your stack. It installs via the npx CLI. This is a narrower tool than Superpowers or the Pocock collection; it doesn't try to own your whole workflow or change how you plan. It just makes sure nothing ships without passing the checks you already care about. For a TypeScript codebase where typecheck is your first line of defense, that's a high-value, low-drama addition.

Best for: TypeScript and React teams whose main pain point is agents declaring victory on code that fails typecheck or tests.

Pricing: Free and open source on GitHub.

Key features

• Sequential quality gates: typecheck, lint, test, review, in that order
• Won't proceed to the next step if the previous gate fails
• React stack profile providing React and TypeScript conventions
• Eight stack profiles including Go, Python, Java, Kotlin, and React
• Installs via npx @c0x12c/ai-toolkit@latest --local

Pros

• The typecheck-first gate is exactly what a strict TypeScript codebase needs
• Eliminates the pattern where the agent patches a failing test to make it pass
• Stack profiles give language-specific conventions instead of generic advice
• Narrow, focused scope means it doesn't fight with your other skills

Cons

• Smaller community and less third-party coverage than the bigger collections
• Only as good as your existing checks; it enforces gates, it doesn't define them
• No TypeScript-specific profile beyond the React one

Visit spartan-ai-toolkit

#6 Addy Osmani's Agent Skills

Addy Osmani's collection is a library of production-grade engineering skills that encode senior-engineer workflows as structured playbooks an agent can follow consistently. According to the project's own materials, it includes 24 skills covering spec writing, TDD, code review, security hardening, CI/CD, and deprecation, plus 8 slash commands (/spec, /plan, /build, /test, /review, /ship, and more) that map to the software lifecycle. The repo had roughly 109.1K stars as of October 2026 per the ai-agent-map tracking. Each skill includes verification gates and an anti-rationalization table so agents can't skip steps, which is a genuinely clever mechanism: it anticipates the ways a model will talk itself out of following instructions. The collection installs into 70+ agents, including Claude Code, Cursor, Codex, Copilot, and Gemini CLI, via the skills CLI. For TypeScript developers, the browser-testing skill is the standout. Per Composio's detailed writeup, it teaches the agent to open the application, interact with it like a user, and collect evidence from the live runtime: DOM inspection, console logs, network requests, screenshots, responsive checks, and performance traces. It includes an explicit rule preventing the agent from claiming success without running the application. If your TypeScript work is web apps, that single skill changes the agent's completion criterion from "the code looks right" to "the workflow was exercised and verified."

Best for: Web-focused TypeScript developers who want disciplined, lifecycle-oriented skills and rigorous browser verification.

Pricing: Free and open source on GitHub.

Key features

• 24 production-grade engineering skills covering the software lifecycle
• 8 slash commands including /spec, /plan, /build, /test, /review, and /ship
• Verification gates and anti-rationalization tables in each skill
• Browser-testing skill with DOM, console, network, and screenshot inspection
• Rule preventing the agent from claiming success without running the app
• Installs into 70+ agents via the skills CLI

Pros

• The anti-rationalization tables directly address how models skip steps
• Browser testing shifts completion criteria from plausible code to verified behavior
• Broad agent compatibility means the same skills work if you switch tools
• Built on documented practices, which the repo credits to Software Engineering at Google

Cons

• General-purpose engineering skills, not TypeScript-specific
• Browser testing doesn't replace unit, integration, accessibility, or security testing
• The full lifecycle approach may be more process than small projects need

Visit Addy Osmani's Agent Skills

#7 vercel-react-best-practices

If your TypeScript work involves React or Next.js, this Vercel-official skill packages the company's React and Next.js performance guidance into an installable skill. Per multiple sources it had roughly 606.7K installs and was the #7 organic skill globally as of August 2026, and one developer reported 176,400 installs via skills.sh for his setup. The skill is designed for writing, reviewing, or refactoring React and Next.js code with better performance patterns. Generic React code can work and still be slow, and this skill directs Claude toward higher-impact problems: unnecessary client-side imports, async waterfalls, avoidable re-renders, and inefficient data-fetching patterns. The concrete difference, per a developer who uses it daily on React Native and Expo: before the skill, Claude generated components that worked in isolation and broke in context. After installing it, output followed patterns he'd actually write himself: proper hook usage, clean prop interfaces, no mystery side effects. The React fundamentals it encodes, component structure, data fetching patterns, and error boundaries, transfer directly to TypeScript React codebases even when you're not on Next.js. It's maintained by Vercel, which gives it a strong publisher signal since the guidance comes from the team that runs the most widely used React deployment platform. Install with npx skills@latest add vercel/react-best-practices.

Best for: TypeScript developers building React or Next.js applications who want performance-aware code generation and review.

Pricing: Free and open source.

Key features

• Vercel's React and Next.js performance guidance packaged as a skill
• Targets unnecessary client-side imports and async waterfalls
• Catches avoidable re-renders and inefficient data-fetching patterns
• Component structure, data fetching, and error boundary patterns
• Useful for writing, reviewing, and refactoring React code
• Maintained by Vercel with an official repo

Pros

• Official Vercel skill, so the guidance tracks current React and Next.js best practice
• Catches real performance regressions, not just style issues
• Very large installed base, so edge cases have been found and reported widely
• Works for both writing new code and reviewing existing React

Cons

• Scoped to React and Next.js; useless for backend-only TypeScript
• Per o-mega's tracking, its install growth (+14% over four weeks) was the slowest of the top cohort
• Doesn't cover TypeScript type discipline itself, only React patterns

Visit vercel-react-best-practices

#8 GStack (garrytan/gstack)

GStack is Garry Tan's Claude Code setup, and at nearly 110,000 stars it's one of the most popular skill collections out there, per the DEV Community tour of the ecosystem. Instead of one mode that handles everything, it gives the agent distinct roles, each behind its own slash command: product vision, designer, engineering manager, release manager, doc engineer, QA, and post-launch retrospective. It's the same genre as Superpowers, a full workflow, but organized around the roles on a software team rather than a sequence of phases. If you think in terms of "who would handle this part," GStack maps to that mental model nicely. For TypeScript teams, that role-based framing is useful when the work spans more than code. The doc engineer role handles the documentation TypeScript projects chronically underwrite, the QA role covers verification, and the release manager role brings discipline to shipping. Per the source writeup, it installs with a clone and a setup script, with flags like --host kiro, --host cursor, and similar for other agents, and it also works with Kiro. The role model means you invoke the specific perspective you need rather than running everything through one monolithic pipeline, which keeps token usage more predictable than Superpowers' automatic triggering.

Best for: Solo developers and small teams who want a full software-team workflow, with different perspectives available on demand.

Pricing: Free and open source on GitHub.

Key features

• Distinct agent roles, each behind its own slash command
• Roles include product vision, designer, engineering manager, release manager, doc engineer, QA, and post-launch retrospective
• Full workflow organized around team roles rather than sequential phases
• Installs via git clone and a setup script
• Host flags for other agents including Kiro and Cursor

Pros

• The role-based model maps naturally to how people actually think about dividing work
• Covers non-code work like docs, QA, and release management that pure coding skills skip
• Explicit slash commands mean predictable invocation and token usage
• Large community at nearly 110,000 stars

Cons

• It's a full-workflow system, so like Superpowers it takes over your process
• Not TypeScript-specific; the roles are language-agnostic
• Installation via clone and setup script is clunkier than npx-based installs

Visit GStack (garrytan/gstack)

#9 Skills Directory (skillsdirectory.com)

Skills Directory is an open directory that indexes agent skills from across GitHub and applies automated security analysis to every listed skill. Per its own site, it has indexed over 670,000 skills, runs detection across 120 patterns in 10 threat categories, and scans for prompt injection, credential theft, data exfiltration, and related issues. It also publishes a notable statistic: 36% of skills in the wild have security flaws, consistent with what Snyk's ToxicSkills research found independently. That framing matters, because it's one of the few directories that treats security scanning as a core feature rather than an afterthought. For TypeScript developers, the value is discovery with a baseline safety net. You can search by category (development, code review, testing, security, databases, and more), read what each skill does, and install from your terminal with npx -y skills find. The directory notes that submissions are reviewed before being published. It's worth being clear about what this is and isn't: it's a discovery and scanning layer, not a benchmarking platform. It tells you a skill is probably not malicious, but it doesn't tell you whether the skill actually improves your agent's output. Pair it with measured results from a benchmarking platform and you get both halves of the trust question answered.

Best for: Developers who want broad discovery with automated security scanning before installing community skills.

Pricing: Free to browse and install skills.

Key features

• Indexes over 670,000 agent skills, continuously updated
• Automated security analysis on every listed skill
• 120 detection patterns across 10 threat categories
• Scans for prompt injection, credential theft, and data exfiltration
• Terminal installation via npx -y skills find
• Category browsing for development, testing, security, databases, and more
• Community submissions reviewed before publication

Pros

• Security scanning is applied to every listed skill, addressing the documented vulnerability problem
• Huge catalog means you'll likely find a skill for niche TypeScript tasks
• Free, with simple CLI-based installation
• Transparent about its scanning methodology and detection patterns

Cons

• Security scanning tells you a skill isn't malicious, not that it's useful; there's no performance benchmarking
• The catalog is largely unaudited beyond automated scanning, so quality varies widely
• Sheer size (670,000+ skills) makes judgment, not discovery, the real bottleneck

Visit Skills Directory (skillsdirectory.com)

Conclusion

The right Claude skills for TypeScript work depend on what's actually breaking in your workflow. If agents ship code that fails typecheck, spartan-ai-toolkit's sequential gates fix that directly. If they build the wrong thing, grill-me or Superpowers force the conversation first. If your React output is slow or generic, vercel-react-best-practices and frontend-design are the established answers. But before you install any of them, the trust question deserves an answer, because published research found roughly 26% of skills carry at least one vulnerability, and popularity doesn't protect you. That's why SkillzBench sits at the top of this list: it's the only platform that scans every skill, pins versions to a commit SHA, and benchmarks each one in a sandbox so you can see the measured Lift before you commit your context window to it. And it's free, including the MCP connector that lets you control exactly which skills each of your agents can see. Browse the directory, check the Trust scores, and look at the benchmark results for yourself at skillzbench.com. Skills, proven.

Frequently asked questions

What makes SkillzBench different from other skill directories?

Most directories stop at listing skills, maybe with a security scan. SkillzBench does three things the others don't: it pins every skill version to a commit SHA so you know exactly what you installed, it runs each skill through security scanning, license checks, and quality judging, and it benchmarks every skill in a sandbox with the same model, with and without the skill, publishing the Lift in percentage points. That last part is the real differentiator. A star count tells you a skill is popular; a benchmark tells you it actually improved agent performance. Lift is always shown separately from Trust scores, and money never touches either. Everything on the platform, including the MCP connector and per-agent loadouts, is free.

Are Claude skills safe to install?

Not automatically. Published research found roughly 26% of skills carry at least one vulnerability, and Snyk's ToxicSkills research detected prompt injection in 36% of skills tested. Skills can include executable scripts that run in your environment, so treat them like any third-party dependency. Practical precautions: review the SKILL.md and any bundled scripts before installing, prefer skills from publishers you can identify (Anthropic, Vercel, and known authors like Matt Pocock carry stronger signals), and use a platform that security-scans skills before you install them. SkillzBench scans every skill it lists and assigns a 0-100 Trust score built from security, quality, license, maintenance, and community signals.

Which Claude skill should a TypeScript developer install first?

It depends on your biggest pain point. If your agent declares success on code that fails typecheck or tests, start with spartan-ai-toolkit, whose sequential quality gates won't let that happen. If your agent builds the wrong thing, grill-me from Matt Pocock's collection makes it interview you first. If you're on React or Next.js, vercel-react-best-practices is the highest-leverage install for code quality. For features that take more than a session, Superpowers brings planning and strict TDD. Whatever you pick, check its Trust score and benchmark results on SkillzBench before installing, so you're choosing based on evidence rather than star counts.

Do Claude skills work with other coding agents like Cursor or Codex?

Yes, in most cases. Anthropic published the Agent Skills specification as an open standard in December 2025, and it's been adopted by roughly 40 products as of mid-2026, including OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, and VS Code, per the agentskills.io showcase. A skill is a directory with a SKILL.md file, and that format now runs across agents without modification. Installation paths vary by agent, but collections like Superpowers, Matt Pocock's skills, and Addy Osmani's library all document installs for multiple tools. SkillzBench's free MCP server works with any agent and lets you toggle which skills and tools each agent can see.

Do Claude skills slow down the agent or waste context?

Installed skills cost very little context when idle. The format uses progressive disclosure: at startup the agent loads only each skill's name and description, roughly 100 tokens per skill, and pulls the full instructions in only when a task matches. So you can install dozens of skills without bloating every session. The real cost shows up when skills run. Workflow-heavy skills like Superpowers deliberately slow the agent down with planning and review phases, and multiple reviewers note they burn a lot of tokens. That's a feature for large features and a bug for small fixes. Skills like caveman exist specifically to compress output and reduce token costs.

Is SkillzBench really free, or is there a catch?

It's genuinely free. The directory, the Trust scores, the benchmark results, the security scanning, the MCP connector, the per-agent loadouts, and the upstream-change alerts all cost nothing. SkillzBench earns revenue from advertising shown in and around its blog articles, and the platform states that ad money never influences scores or listings. That model matters here: when a ranking platform charges the tools it ranks, you have to wonder about the rankings. SkillzBench's position is that Lift is always shown separately from ratings and money never touches scores, so you can verify performance claims against the published benchmark results yourself.