Zum Hauptinhalt springen

Must-Have Skills for Claude Code Agents

Agent Skills are specialized instruction files (SKILL.md) that extend what a coding agent can actually accomplish in production. Without skills, agents fall back to "statistically average" output: generic UI, shallow research, ad-hoc database design, no security validation. With the right skills installed, the same model produces work that ships.

This guide collects 10 skills every Claude Code setup should have in 2026, why each one matters, and how to install it. It is based on the community write-up 10 Must-Have Skills for Claude and Any Coding Agent in 2026.

Why skills, not prompts?

Skills load on demand via progressive disclosure: only the YAML frontmatter is always in context. The body, scripts and references are pulled in when the agent decides the skill is relevant. That means you can bundle large playbooks with no ongoing context cost.

TL;DR — The 10 Skills at a Glance​

#SkillFixes the gap…
1Frontend DesignGeneric, "AI-looking" UI components
2Browser UseCannot verify deployed features end-to-end
3Code ReviewerShips duplicated logic, dead code, missing error handling
4RemotionNo way to produce demo or release videos
5Google WorkspaceManual OAuth + client setup for Gmail/Docs/Sheets
6ValyuWeb search misses authoritative domain data
7Antigravity Awesome SkillsReinventing common engineering playbooks
8PlanetScale Database SkillsBad schema decisions discovered too late
9ShannonSecurity holes that automated scanners miss
10Excalidraw Diagram GeneratorArchitecture exists only in someone's head

1. Frontend Design​

Purpose: Generates visually distinctive UI components instead of the cookie-cutter hero/three-card/CTA layout every model defaults to.

Why it matters: Without a design skill, an agent picks the statistically-most-likely React component for a given prompt. The result looks like every other AI-generated landing page on the internet. A frontend-design skill injects an explicit design system, typographic scale, spacing tokens and intentional aesthetic constraints — turning the agent into an opinionated collaborator rather than a Tailwind autocomplete.

Use it for:

  • Production landing pages and marketing sites
  • User-facing product surfaces where polish matters
  • Component libraries with consistent look-and-feel
Default behaviour shift

The most valuable skills are the ones that shift the agent's defaults. Frontend Design is the canonical example: you should never need to manually invoke it — it fires whenever the agent reaches for a <Hero> or a <Pricing> component.

2. Browser Use​

Purpose: Gives the agent control of a headless browser, so it can navigate, click, fill forms, and read JavaScript-rendered content.

Why it matters: "It compiles" is not the same as "it works." A browser-use skill closes the loop: after deploying a feature, the agent can actually open the staging URL, log in, exercise the flow, and report what it sees — including runtime errors that no static analysis catches.

Use it for:

  • End-to-end verification of deployed changes
  • QA and regression checks against staging environments
  • Form-filling and data extraction from JS-heavy sites
  • Research tasks that need post-render content

3. Code Reviewer​

Purpose: Runs a structured review pass on the agent's own diff before presenting it to the user.

Why it matters: Agents are good at writing code and bad at noticing the smell of their own code. A reviewer skill enforces a checklist: duplicated logic, unnecessary abstractions, inconsistent naming, missing error handling at boundaries, dead code left behind from earlier attempts. It is the single highest-leverage skill for raising baseline quality.

Use it for:

  • Self-review before handing a diff back to the user
  • Extracting genuinely reusable utilities (vs. premature abstraction)
  • Catching N+1s, leaked secrets, and forgotten TODOs
Pair with the built-in /review

Saurus already ships a /review slash command. The Code Reviewer skill complements it by running before the diff is shown — so most issues are fixed silently rather than discovered in review.

4. Remotion​

Purpose: Programmatic video creation through React components — the agent writes JSX, Remotion renders MP4.

Why it matters: Video used to be an entirely separate workflow (Final Cut, After Effects, a human editor). With this skill, "ship a 30-second demo of the new feature" becomes a task the agent can complete end-to-end alongside the code change.

Use it for:

  • Product demos for release notes
  • Animated changelogs and README headers
  • Onboarding and explainer videos
  • Internal training clips

5. Google Workspace (GWS)​

Purpose: Unified access to 50+ Google Workspace APIs (Gmail, Calendar, Drive, Docs, Sheets, Tasks).

Why it matters: Every team eventually needs the agent to "send the report," "add it to the calendar," or "drop it in the shared sheet." Doing that without a skill means a fresh OAuth dance, picking a client library, and writing scope boilerplate. The GWS skill handles all of that and exposes a single high-level surface.

Use it for:

  • Email drafting and triage
  • Calendar management and meeting prep
  • Status tracking in Sheets
  • Document generation and templating in Docs

6. Valyu​

Purpose: Connects the agent to specialized data sources beyond general web search — SEC filings, PubMed, ChEMBL, clinical trials registries, economic databases.

Why it matters: Generic web search is a noisy proxy for the underlying authoritative source. For domains where correctness matters (finance, biomed, regulatory) the agent needs the actual primary source, not the third-party blog post that summarized it.

Use it for:

  • Financial research grounded in filings
  • Biomedical and pharmacological analysis
  • Patent and prior-art searches
  • Economic dashboards using authoritative time series

7. Antigravity Awesome Skills​

Purpose: A curated library of 1,200+ compatible skills spanning common engineering domains.

Why it matters: Most "skills you would write" already exist. Architecture planning, debugging methodology, API design checklists, security review, PR-description generation — there is a battle-tested community skill for it. The Antigravity collection is the index that prevents you from writing the 1,201st version of "review my Dockerfile."

Use it for:

  • Bootstrapping a new project's skill set
  • Reference implementations for your own custom skills
  • Filling gaps you didn't know you had
Browse before building

Before writing a custom skill, check the Antigravity Awesome Skills repo. Saurus's skills authoring guide covers what to do when you genuinely need a new one.

8. PlanetScale Database Skills​

Purpose: Teaches the agent serverless MySQL schema design, branching workflows and index-aware query analysis.

Why it matters: Schema decisions are the most expensive class of mistake an agent can make: they look fine in dev and fail under production load months later. This skill makes the agent ask the questions a senior DBA would — what queries hit this table, where are the indexes, what does the migration path look like — before the migration is written.

Use it for:

  • Generating schemas that won't need to be re-done
  • EXPLAIN-aware query optimization
  • Migration planning with database branching
  • Reviewing existing schemas for hot spots

9. Shannon​

Purpose: Autonomous penetration-testing agent that runs real exploits in authorized environments.

Why it matters: Static SAST tools produce noise; humans produce slow, expensive results. Shannon claims a 96.15% exploit success rate with zero false positives — meaning every reported finding is something an attacker could actually use. That changes triage from "is this real?" to "fix it now."

Use it for:

  • Validating staging deployments before promotion
  • Confirming IDOR and broken-access-control issues
  • Proving SQL-injection risks with reproducible payloads
  • Continuous security validation in development environments
Authorization required

Only point Shannon at systems you own or have explicit written authorization to test. "Real exploits" means real legal exposure if used against third parties.

10. Excalidraw Diagram Generator​

Purpose: Generates production-quality architecture diagrams from natural-language descriptions, with automatic layout fixes.

Why it matters: "Where does the request go?" is the question every new engineer asks on day one, and the answer usually lives in the head of one person who joined three years ago. This skill makes diagram generation cheap enough that you can keep a diagram next to every non-trivial system — and regenerate it when the system changes.

Use it for:

  • Architecture documentation in READMEs and ADRs
  • System-design communication in PRs and RFCs
  • Data-flow explanations during onboarding
  • Refactoring discussions ("here's the before/after")

Installing Skills in Claude Code​

Skills are distributed via the Anthropic skills marketplace and via plugin repositories. The general workflow:

# Add the official Anthropic marketplace once
/plugin marketplace add anthropics/skills

# Install an individual skill
/plugin install <skill-name>@anthropic-agent-skills

Skills land in one of two locations:

    • frontend-design/
    • code-reviewer/
    • ...
LocationScopeWhen to use
~/.claude/skills/PersonalSkills you want everywhere (review, design)
.claude/skills/ProjectSkills tied to this codebase (db, deploy)
Claude.ai and the API

Skills also work in Claude.ai (paid plans, uploaded via settings) and in the Claude API (container parameter with the skills-2025-10-02 beta flag). The underlying SKILL.md format is identical across all three.

Picking Your First Five​

If you're starting from zero, install in this order — each one pays for itself within a single afternoon:

  1. Code Reviewer — fixes bugs you would have shipped today
  2. Browser Use — closes the verification loop
  3. Frontend Design — every UI task improves immediately
  4. Antigravity Awesome Skills — gives you 1,200+ more for free
  5. The domain skill that matches your stack — PlanetScale if you're on MySQL, GWS if your team lives in Google, Valyu if you do research

The remaining five (Remotion, Shannon, Excalidraw, plus whichever domain skills you skipped) are situational: install them when the situation arrives.

Further Reading​