6 Claude Code Skills You Should Actually Install (2026 Edition)

Claude Code now has dozens of skills and plugins available, both official and community-built. Installing them indiscriminately backfires: when several skills cover the same territory, Claude has to guess which one to use, and that guesswork shows up as unpredictable behavior. Based on what's available as of July 2026, this post narrows the field down to 6 skills that actually move the needle in day-to-day work, along with the selection criteria behind the picks and the install commands. It also covers what to watch for when rolling these out to a team. The 6 skills at a glance A typical dev cycle runs design → implementation → review → security → testing, then loops back to design. Each of the 6 skills below maps to one of these phases. flowchart LR A[Designskill-creator] --> B[Implementation UIfrontend-design] B --> C[Reviewcode-review] C --> D[Securitysecurity-guidance] D --> E[Testingplaywright] E --> A Skill Phase What it does superpowers Development discipline Forces steps like "clarify requirements before writing code" and "identify root cause before fixing a bug" skill-creator Design (meta: builds skills) Walks you through a Q&A to generate a custom SKILL.md for your own workflows frontend-design Implementation (UI) Generates frontends with a deliberate aesthetic direction instead of generic "AI-looking" UI security-guidance Security Detects roughly 25 vulnerability patterns in real time on every edit playwright Testing Drives a browser to run E2E tests and verify behavior discord Notifications / remote control Sends run notifications and lets you operate Claude Code remotely from your phone Three criteria guided this selection: Keep exactly one skill per phase. Installing multiple skills that cover the same phase makes Claude hesitate over which one to invoke, which leads to unpredictable behavior. Prefer official or well-established plugins. A SKILL.md someone wrote as a side project is riskier to depend on than something maintained by Anthropic or widely used across the ecosystem (high weekly install counts, for example) — the latter is more likely to keep being maintained. Start with these 6, add more only when you actually need to. Add anything beyond this baseline only once you hit a task you find yourself repeating almost identically every time. Things like test generation, changelog formatting, or company-specific checklists are worth building yourself with skill-creator — but only after you've used the core 6 enough to know exactly what you need. The 6 skills in detail 1. superpowers (enforces development discipline) What it does: Claude Code tends to jump straight into implementation, skipping requirements clarification and root-cause analysis. Superpowers forces steps like "clarify requirements before adding a feature" and "identify the root cause before fixing a bug." Built by developer Jesse Vincent, it was added to the official plugin marketplace in January 2026. Install /plugin install superpowers@claude-plugins-official Usage: No special invocation is needed after install — just ask Claude to implement something as usual, and it will automatically walk through clarify requirements → design → implement → test. If you only want a specific phase, like brainstorming or test-driven development, you can call the relevant sub-skill directly, e.g. /superpowers:brainstorming. 2. skill-creator (a skill for building skills) What it does: Walks you through an interactive Q&A to build a SKILL.md tailored to your own team's workflows. This is an official Anthropic plugin, and one of the most widely used skills in the ecosystem. Install /plugin install skill-creator@claude-plugins-official Usage: Typing /skill-creator starts an interactive Q&A. Answer questions about the goal of the skill you want, expected inputs, and expected outputs, and it generates a SKILL.md template. Save it under .claude/skills/, then call it with /skill-name to test and refine it. It's best to build one skill per purpose — e.g., a standardized meeting-notes format or a code review checklist — rather than one catch-all skill. 3. frontend-design (frontend UI generation) What it does: Instead of generic "AI-looking" defaults, this generates UI with a deliberate aesthetic — brutalism, glassmorphism, and so on. It's an official Anthropic skill with a large install base. Install /plugin install frontend-design@claude-plugins-official Usage: When asking for UI, be explicit about both the skill and the style you want — e.g., "Using the frontend-design skill, build an admin dashboard with a brutalist aesthetic." Follow-up nudges like "make it more monochrome" or "give it more whitespace" refine the result while keeping the original direction. 4. security-guidance (security scanning) What it does: On every code edit, it scans for roughly 25 vulnerability patterns — injection, unsafe deserialization, dangerous DOM API usage, and more — using regex-based real-time detection. It also layers in an LLM-based diff review at the end of each turn and a cross-file review at commit time, for a three-tier check. This is a free, official Anthropic plugin. Install /plugin install security-guidance@claude-plugins-official Usage: No explicit invocation is needed — a background scan runs automatically on every edit. When it flags a vulnerability, it proposes a fix on the spot, which you can review and choose to apply. If you'd rather batch a review before committing, you can also ask it to actively review a specific scope of changes. 5. playwright (browser automation) What it does: Drives a browser to handle navigation, form input, screenshots, and console log inspection. There are two ways to install it: Microsoft's official Playwright MCP, or a CLI+Skill setup (playwright-cli install --skills) that's more token-efficient. Install # MCP approach claude mcp add playwright # CLI+Skill approach playwright-cli install --skills Usage: With the MCP approach, register it with claude mcp add playwright, then ask Claude to test something like a login form — it will drive the browser and return screenshots and results. With the CLI+Skill approach, call /playwright after install and specify the test scenario step by step. 6. Discord (notifications / remote control) What it does: Sends Claude Code's run results to Discord, and lets you pair with a bot via DM to remotely generate code or manipulate files from your phone. It combines an MCP server with Agent Skills. Install /plugin install discord@claude-plugins-official Usage: After installing, DM the bot on Discord and it replies with a pairing code. Enter that code in Claude Code to complete the link. From there, the /discord command handles run notifications and lets you remotely generate code or manipulate files via Discord. Rolling this out to a team Commit skills to the project-level .claude/skills/ directory rather than your personal home directory (~/.claude/skills/), so the whole team shares the same 6 skills. For production repos where you want to avoid Claude auto-triggering a skill at an unintended moment, set disable-model-invocation: true in the SKILL.md frontmatter. This disables Claude's automatic invocation from conversational context, restricting it to explicit /skill-name calls only. security-guidance is designed with the assumption that it will produce some false positives, so it's worth agreeing as a team on how much to trust its findings. Discord involves managing bot tokens and notification channels, so decide in advance who's allowed to notify or operate through which channel. It's also worth restricting which marketplaces are allowed for installs — official (claude-plugins-official) or internally vetted ones only. Wrapping up Once these 6 skills cover discipline, design, implementation, security, testing, and notifications, the next step is to use skill-creator to turn your own team's repeated workflows into SKILL.md files. The generic skills above won't catch your company's specific review standards or release process — and that's exactly where building your own skills pays off the most. For more on how skills work under the hood, see the official Claude Code docs. 📌 This post reflects information as of July 2026. Claude Code updates frequently, so check the official docs for the latest specifics. This article was edited with AI assistance. *Originally published in Japanese on EdgeHUB.