The Best Claude Code Skills Overall in 2026: Ranked and Compared

Anthropic ships /code-review as a bundled skill, meaning it is part of the official Claude Code workflow surface rather than a community convention. The command reviews the current diff for correctness bugs and cleanup opportunities, and it can optionally apply fixes. It sits at the end of almost every branch and compounds with human review rather than replacing it — a weak code-review skill adds noise, while a good one catches the bug you would otherwise ship. Claude Code is included in the Pro plan ($17/month with annual subscription discount, $20 billed monthly) and Max plans (from $100/month). Independent testing of 16 review tools on 12 pull requests found Claude Code with no tool installed found all 8 planted defects with no false alarm, and six tools matched it — so the bundled skill's value is consistency and a fixed report format on top of already-strong baseline review.

pricing 0

• + Official Anthropic skill — part of the documented Claude Code workflow, not a community convention
• + Zero setup: ships with Claude Code
• + Pays off on almost every branch; compounds with human review
• + Free — included with the Claude Code plans that already include it

• − Adds a fixed checklist and report format rather than better detection — testing found it equals Claude Code with nothing installed for finding defects
• − Generic standards may not cover team-specific coding rules

https://claude.com/pricing

write-tests is the most common 'every team re-builds it' Claude Code skill pattern. 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 in the project. Project-local versions tend to be more valuable than a generic public one because they encode the actual conventions a team uses. Test-writing is described as the fastest path from 'helpful assistant' to 'real leverage' in day-to-day coding. As a pattern rather than one canonical package, it lives in .claude/skills/ and is committed to git so everyone who clones the repo gets it. Skills are cheap when idle: only the name and description (roughly 50–100 tokens) load at session start, with the full SKILL.md (around 500 tokens) loading only when a task matches.

pricing 0

• + Encodes your team's actual test conventions, which generic public skills miss
• + Saves real time on backfilling coverage before refactors and first-pass tests for new features
• + Free to author — a SKILL.md is one markdown file, and the Skill Creator skill can write it for you
• + Repo-local and versionable via git

• − Not one canonical package — you must build or adapt it yourself
• − Quality depends entirely on how well it encodes your conventions
• − An 'encoded preference' skill needs maintenance as your stack changes

/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, making it a parallel-work orchestration surface inside Claude Code rather than a simple shortcut. It ranks high because 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. Typical uses include big migrations, repeated changes across many packages, and 'same edit in five places' tasks. Like all bundled skills it is free with Claude Code, which is included in the Pro plan at $17/month with annual discount ($20 monthly) and Max plans from $100/month. Skills can also run in an isolated subagent context via the context: fork frontmatter option, keeping heavy parallel work from crowding your main session.

pricing 0

• + Changes how much work Claude Code can do in one session — parallel orchestration, not a prompt shortcut
• + Official and documented by Anthropic
• + Free and pre-installed with Claude Code
• + Ideal for migrations and repeated multi-package edits

• − Only relevant when a task is genuinely large — overkill for small linear changes
• − Parallel worktrees consume more usage against plan limits

https://claude.com/pricing

Superpowers is the most popular community skill collection of 2026, with 293,000+ GitHub stars, sitting atop an ecosystem where Anthropic's official skills repositories hold 179,000+ stars. Community frameworks like this one are a fast way to fill .claude/skills/ with dozens of capabilities at once. The caution comes from hands-on testing: one reviewer installed all 47 skills from a popular collection and found 40 of them made output worse, and a broader review of 200+ skills across dozens of repositories concluded most publicly available skills actively hurt — adding tokens, adding latency, and injecting constraints that narrow the output. The useful lens is the capability-uplift vs encoded-preference split: capability-uplift skills are doomed by the next model release as the base model catches up, while encoded-preference skills keep their value. Bulk collections blur that distinction, which is why curation and benchmarking matter before installing.

pricing 0

• + The most-starred community collection — broad coverage of workflows
• + Free and open-source
• + Fast way to explore what skills can do before writing your own

• − Testing of a 47-skill popular collection found 40 skills made output worse
• − Bulk installs add token cost and latency even when idle-heavy skills never fire usefully
• − Capability-uplift skills lose value as base models improve
• − Community-maintained — no vendor guarantee of quality or security

The code-reviewer skill on Agensi is one of the most-used skills in the marketplace. It runs a structured review that checks for security vulnerabilities, logic errors, edge cases, performance issues, and style violations, with findings grouped by severity: Critical, Warning, and Suggestion. What sets it apart is the systematic approach — it follows a defined checklist rather than free-form analysis, so it catches the same categories of issues every time. It is particularly strong on security: SQL injection, XSS, authentication bypasses, and insecure data handling are all on its radar. It works with Claude Code, OpenClaw, Codex CLI, and Cursor, and is free to use through a connected agent. The trade-off versus Claude's built-in review is consistency versus fit: a dedicated skill gives Claude a fixed framework, but generic standards will not cover team-specific rules, which is why building your own SKILL.md (about 15 minutes of work) is recommended when your team has specific standards.

pricing 0

• + Ready-to-use with zero setup — no authoring required
• + Consistent checklist produces the same categories of findings every time
• + Strong security coverage: SQL injection, XSS, authentication bypasses, insecure data handling
• + Works across multiple AI coding agents, not just Claude Code

• − Generic standards will not cover team-specific coding rules
• − A fixed checklist adds a report format, not better detection than baseline Claude review
• − Marketplace skill — quality and maintenance depend on the publisher

https://www.agensi.io/learn/best-code-review-skills-claude-code

SkillzBench is AIQ Labs' free, independent directory of AI agent skills, plus a free MCP tool that connects any agent to the skills you choose. It exists because agent skills are everywhere but nobody can tell which ones actually help and which are risky — published research found ~26% of skills carry at least one vulnerability. SkillzBench is building five pillars: Discover (continuously crawl and dedupe skills from GitHub), Verify (pin every version to a commit SHA, validate, license-check, security-scan, and quality-judge), Benchmark (test every skill in a sandbox with the same model, with and without the skill, and publish the Lift in percentage points), Equip (one free MCP server plus per-agent loadouts, where the agent only sees the skills and tools you toggle), and Stay current (track upstream changes, re-scan, re-benchmark, notify). A trust score from 0 to 100 combines security, quality, license, maintenance, and community signals; Lift is always shown separately from ratings, and money never touches scores. SkillzBench is pre-launch and actively being built — the waitlist is live, with the directory, benchmarks, and MCP to follow. Everything on the platform will be free; revenue comes from advertising around blog articles, never from charging users.

pricing 0

• + Everything free — the directory, the MCP connector, benchmarks, and alerts
• + Addresses the ~26% vulnerability problem with security scanning and version pinning to commit SHAs
• + Lift shown separately from ratings; money never touches scores
• + Per-agent loadouts mean the agent only sees the skills you toggle, cutting token bloat
• + Tracks upstream changes and re-benchmarks so skills stay current

• − Pre-launch — the directory, benchmarks, and MCP are actively being built, not yet public
• − Independent directory, not a skill itself — you still install skills elsewhere

https://skillzbench.com