The best Claude Code skills in 2026, ranked by day to day usefulness: code review, test writing, parallel batch work, debugging, release flows, and one visual bundle from Nimbalyst.
The best Claude Code skills in 2026 are the ones that remove repeated setup from work you already do every week: review a diff, write tests, debug a failure, run work in parallel, verify the app, and ship cleanly. One important detail is that “skill” in Claude Code does not just mean one marketplace package. Some of the best skills are bundled by Anthropic, some are repo-local skills in .claude/skills/
, some still live in .claude/commands/
, and some ship through plugins.
This is why the strongest Claude Code skill libraries look a little uneven. A good team usually has a mix of official bundled skills, project-specific skills, and one or two visual or domain-specific bundles. Below is the ranked list I would actually keep in rotation.
Best Claude Code Skills: Quick Answer #
Best Claude Code skill overall:/code-review
. It pays off on almost every branch and Anthropic ships it as a bundled skill.Best custom skill to add to almost any repo:write-tests
. Teams keep re-inventing this one because it saves real time.Best official skill for parallel work:/batch
. It breaks a large task into units and runs them in isolated worktrees.Best visual skill bundle: Nimbalyst’s visual skills bundle. It is the cleanest example of when a coding workflow needs mockups, diagrams, or data models instead of more text.
One caveat before the ranking: several of the names below are common skill patterns, not one universal package with one canonical install command. That is normal in Claude Code. The best skills are often repo-local and tuned to the team’s actual workflow.
Claude Code Skills Comparison Table #
| Rank | Skill | Type | Best for | Typical source |
|---|---|---|---|---|
| 1 | /code-review | Bundled skill | Catching bugs and cleanup before PR review | Anthropic bundled skill |
| 2 | write-tests | Custom skill pattern | Fast, repo-aware test generation | Project-shipped or community |
| 3 | /batch | Bundled skill | Large parallel changes across a codebase | Anthropic bundled skill |
| 4 | /debug | Bundled skill | Install and runtime troubleshooting | Anthropic bundled skill |
| 5 | refactor | Custom skill pattern | Safer multi-step cleanup work | Project-shipped or community |
| 6 | playwright | Custom skill pattern | E2E and browser workflow coverage | Project-shipped or community |
| 7 | investigate | Custom skill pattern | Root-cause work before a fix | Project-shipped |
| 8 | commit | Custom skill pattern | Consistent commit messages and commit hygiene | Project-shipped or community |
| 9 | /verify | Bundled skill | Confirming the change works in the running app | Anthropic bundled skill |
| 10 | track | Custom skill pattern | Turning work into linked bugs, tasks, or decisions | Project-shipped |
| 11 | release-alpha | Custom skill pattern | Repeatable release flow with less checklist drift | Project-shipped |
| 12 | Nimbalyst visual skills bundle | Plugin bundle | Mockups, diagrams, and data models in the same workflow | Nimbalyst plugin bundle |
What makes a Claude Code skill worth keeping? #
Three filters matter more than novelty:
It removes repeated setup. If you type the same setup instructions every week, that is a skill candidate.It knows something the base model does not. File paths, repo rules, release steps, test helpers, and domain conventions are where skills earn their keep.It triggers at the right time. A broad skill that fires on everything is worse than no skill.
The Ranked List #
1. /code-review
If you only install or learn one Claude Code skill deeply, make it /code-review
.
Anthropic ships /code-review
as a bundled skill, which matters because it means the command is not a random community convention. It is part of the official Claude Code workflow surface. The command reviews the current diff for correctness bugs and cleanup opportunities, and it can optionally apply fixes.
Why it ranks first: it sits at the end of almost every branch, and it compounds with human review rather than replacing it. A weak code-review skill adds noise. A good one catches the bug you would otherwise ship at 6 p.m.
Best for: Any branch that is about to become a PR.
2. write-tests
write-tests
is the most common “every team re-builds it” Claude Code skill pattern for a reason.
The useful versions do not just say “write tests.” They read the repo, infer whether the project uses Vitest, Jest, Playwright, RSpec, or something else, inspect nearby tests for style, then generate tests that look like they belong there. Project-local versions tend to be more valuable than a generic public one because they encode the actual conventions a team uses.
Why it ranks second: test-writing is the fastest path from “helpful assistant” to “real leverage” in day to day coding.
Best for: Backfilling coverage before a refactor or generating the first test pass for a new feature.
3. /batch
/batch
is the official bundled skill for work that is too large for one linear pass.
Anthropic’s docs describe it as a skill that decomposes a large change into independent units and runs them in separate git worktrees. That makes it more than a shortcut. It is a parallel-work orchestration surface inside Claude Code.
Why it ranks this high: many skill lists over-index on tiny helper prompts and under-rate the skills that actually change how much work Claude Code can do in one session.
Best for: Big migrations, repeated changes across many packages, and “same edit in five places” tasks.
4. /debug
/debug
is another official bundled skill that is easy to overlook until you need it.
Anthropic ships it for install and runtime troubleshooting, which keeps it grounded in a real workflow instead of vague “debugging tips.” It is especially useful when the problem is not your application logic but Claude Code’s environment, config, or runtime behavior.
Why it ranks fourth: many coding sessions do not need it, but the sessions that do need it really need it.
Best for: Claude Code install trouble, broken runtime behavior, confusing environment issues, and session log analysis.
5. refactor
refactor
is a strong example of a skill that is usually better as a pattern than as a single named package.
A good refactor skill breaks work into safe steps, keeps diffs reviewable, runs tests between major edits, and avoids the “rewrite everything at once” trap. Community versions exist, but the best ones are usually tuned to the team’s own code review tolerance and verification rules.
Why it belongs near the top: refactors are where AI is most dangerous when left vague and most useful when constrained.
Best for: File splits, naming cleanups, extracting services, and other multi-step codebase cleanup.
6. playwright
A playwright
skill is one of the clearest examples of a repo-specific skill that beats a generic prompt.
A solid /playwright
command forces precise selectors, existing helpers, sequential workflow tests, and no lazy .first()
escapes. That is exactly what a real Claude Code skill should do: encode local testing rules that the base model would not guess.
Why it ranks sixth: frontend and product teams get repeatable value from it, but only if it is wired to the repo’s actual Playwright conventions.
Best for: E2E coverage, browser regressions, and UI workflows that need structured verification.
7. investigate
investigate
is the planning-first bug skill.
A good /investigate
command gathers evidence, forms a root-cause hypothesis, classifies scope, and only then asks whether to fix or design. That is a healthier default than charging into code on the first symptom.
Why it ranks seventh: it does not ship new capability, but it dramatically improves judgment on ambiguous work.
Best for: Bug triage, regressions, and tasks where the shape of the problem is still unclear.
8. commit
commit
looks small until you use it every day.
Useful versions read the current diff, match the repo’s commit style, write a concise title, and catch avoidable mistakes like staging the wrong file or forgetting a required changelog update. A /commit
command tuned to a repo can quietly improve every PR.
Why it ranks eighth: it is not glamorous, but it improves history quality, review readability, and release hygiene.
Best for: Teams that care about commit message consistency and clean branch handoff.
9. /verify
/verify
is one of the most valuable official bundled skills because it pushes Claude Code past “tests passed” and toward “the change actually works.”
Anthropic documents it as a skill that builds or runs the app and observes the result rather than relying only on static checks. That is a better mental model for AI-assisted coding than treating tests as the whole truth.
Why it ranks ninth instead of higher: it is critical on product work, but it depends more on app setup than review or test-writing skills do.
Best for: UI changes, workflow fixes, and any task where real behavior matters more than green unit tests.
10. track
track
is the skill that turns work into durable project memory.
Coding is not the only job. Teams also need linked bugs, decisions, and tasks that survive beyond a single transcript. If you already have a tracker, a skill like this keeps the agent from treating planning and implementation as disconnected worlds.
Why it ranks tenth: it is not universal, but once multiple people or multiple agents are involved, it becomes hard to live without.
Best for: Teams that want bugs, decisions, and implementation work tied together.
11. release-alpha
release-alpha
is exactly the sort of skill most teams should not pretend is generic.
Release skills are only good when they match the actual release flow: changelog rules, version bumping, prerelease tags, CI checks, and deployment sequence. A project-shipped /release-alpha
command with real guardrails around what gets updated and when is far more reliable than a community version.
Why it ranks eleventh: huge value when it fits, low value if it does not match your release pipeline.
Best for: Teams with a recurring release process that is still too manual.
12. Nimbalyst visual skills bundle
This is the one Nimbalyst entry that belongs on the list.
Nimbalyst bundles a set of visual skill surfaces that are genuinely different from text-only Claude Code workflows:
MockupLM for.mockup.html
UI mockupsDataModelLM for Prisma-backed visual data modelsExcalidraw for diagrams and architecture sketches
What makes this bundle worth mentioning is that some coding work should produce a mockup, a data model, or a diagram instead of another wall of markdown. In a typical Nimbalyst setup, mockuplm
and datamodellm
ship as Claude plugin commands, and excalidraw
ships as a plugin skill.
Nimbalyst is an open-source visual workspace that runs Claude Code and OpenAI Codex side by side, with pluggable agent harnesses. The desktop and iOS apps are MIT-licensed.
Why it ranks twelfth: it is not the first skill bundle most developers need, but it is one of the clearest examples of a skill bundle that expands the workflow beyond plain code generation.
Best for: Product work that needs visual artifacts alongside code.
How to install a Claude Code skill #
The modern default is a folder with a SKILL.md
file:
.claude/skills/my-skill/SKILL.md
This is the preferred Claude Code shape for project-local skills. User-level skills live under:
~/.claude/skills/my-skill/SKILL.md
Claude Code also still supports command-style files like:
.claude/commands/write-tests.md
The older command form still works and is common in real repos. In practice, the best advice is simple:
- Use
.claude/skills/
for new skills. - Expect to see
.claude/commands/
in existing repos. - Treat plugin-contributed skills and commands as a third category, especially for editor-specific workflows.
Claude Code Skills vs MCP Servers #
This is the distinction that matters most:
- A skill tells Claude Code how to do something. - An MCP server gives Claude Code new tools it can call.
If you want Claude Code to follow your test conventions, write a skill. If you want Claude Code to talk to GitHub, Playwright, Slack, or a database, add an MCP server. Good teams use both.
Frequently Asked Questions #
What is the best Claude Code skill overall?
/code-review
is the safest overall pick because it improves almost every branch and ships as an official bundled skill. If you only add one custom skill of your own, make it write-tests
.
Are the best Claude Code skills official Anthropic skills?
Not all of them. Some of the best are official bundled skills such as /code-review
, /batch
, /debug
, and /verify
. Many others are project-local skills because the most valuable instructions are repo-specific.
Where should Claude Code skills live?
Use .claude/skills/<name>/SKILL.md
for project-scoped skills and ~/.claude/skills/<name>/SKILL.md
for personal skills you want everywhere. Older .claude/commands/*.md
files still work and remain common in open-source repos.
What is the best Claude Code skill for teams?
For teams, the best stack is usually /code-review
, write-tests
, track
, and either /batch
or /verify
, depending on whether the team struggles more with scale or with validation.
Are Claude Code skills different from subagents?
Yes. A skill is reusable instruction. A subagent is a separate running Claude session. They work well together: a main session can load a skill and then spawn a subagent for a narrower piece of the work.
Can I use Claude Code skills inside Nimbalyst?
Yes. Nimbalyst is a workspace layer above Claude Code, not a replacement for skills. It is most useful when the output should be visual, multi-agent, or cross-tool rather than just another terminal transcript.
Related Reading #
Related pages #
Nimbalyst vs Claude Code Desktop
Compare a workspace around Claude Code with Anthropic's official desktop surface.
Cursor vs Claude Code Desktop vs Nimbalyst
A 3-way comparison for teams choosing the right surface around Claude Code.
OpenCode vs Claude Code
Helpful context when your Claude Code workflow also includes open-source harnesses.
Related posts #
Claude Code Skills: A Practical 2026 Guide
Claude Code skills explained: what they are, how to write one, the best built-in skills, and how skills differ from MCP servers and subagents.
Claude Code Plugins: A 2026 Guide
Claude Code plugins explained: what they are, how they extend the agent, and how plugins relate to MCP servers, skills, and Nimbalyst extensions.
Claude Code Subagents: A Practical 2026 Guide
Claude Code subagents explained: what they are, when to use them, how to configure them, and how parallel subagents change the agent loop.