topskills

Claude Code guide

Best agent Skills for Claude Code

The best Skills to install in Claude Code first are frontend-design (the highest-scoring Skill in our catalogue, 277K+ installs) for any UI work, superpowers for multi-session engineering work that needs a plan to survive, and simplify for a reuse-and-clarity review pass before merge. Claude Code is the deepest Skill target there is — it has filesystem access, shell execution, and subagents, so every catalogued Skill is verified here and action-taking Skills reach their full capability.

Updated August 26, 2026 · first published August 19, 2026

Install first

frontend-design

The single highest-scoring Skill in the catalogue and the most-installed design Skill in the ecosystem. It forces a committed visual direction before code is generated, which is the difference between output that looks intentional and output that looks generated.

Best when: Any greenfield UI — landing pages, marketing sites, dashboards — where the interface is part of the product rather than scaffolding.

Install model

How Skills install on Claude Code

npx skills add github.com/anthropics/skills --skill frontend-design
  1. 1Pick the Skill and copy its Claude Code install command from its page on this site.
  2. 2Run the command, or place the Skill folder in ~/.claude/skills/ for user-wide scope, or .claude/skills/ inside the repo for project scope.
  3. 3Start a new Claude Code session — frontmatter is read at session start, so a newly added Skill is not visible mid-session.
  4. 4Confirm it loaded by asking Claude which skills are available, then trigger it with a task that matches its description.

Claude Code reads each SKILL.md frontmatter at session start (~100 tokens per Skill) and loads the full instructions only when a task matches the description. Project-scoped Skills in .claude/skills/ are committed with the repo, so every contributor and every agent session inherits the same set — usually what you want for conventions, while user-wide scope suits general-purpose Skills.

Ranked

The shortlist

Ordered by what to install first for the common case, not by raw score. Every entry here carries verified Claude Code compatibility — open the skill page for the composite score, licence, and full install commands.

SkillSourceScoreBest when
frontend-designOfficial92Any greenfield UI — landing pages, marketing sites, dashboards — where the interface is part of the product rather than scaffolding.
superpowersCommunity63Multi-file features and refactors that will outlive a single session. Overkill — and pure token cost — on a one-file fix.
simplifyOfficial50Right before opening a PR, on a diff you already believe is correct.
skill-creatorOfficial48The second time you paste the same instructions into a session. That is the signal it should have been a Skill.
pdfOfficial52Any task where a real PDF is the input or the deliverable, rather than text about a PDF.
debugOfficial51A failure you cannot immediately explain — especially an intermittent one.

See every skill verified on this agent, ranked by composite score, on the Claude Code page.

Why these

Pick by pick

  1. 1
    frontend-design

    The single highest-scoring Skill in the catalogue and the most-installed design Skill in the ecosystem. It forces a committed visual direction before code is generated, which is the difference between output that looks intentional and output that looks generated.

  2. 2
    superpowers

    Imposes a full brainstorm → spec → plan → build → review pipeline with test-driven development enforced at each stage. On work spanning many files and several sessions, the expensive failure is losing the plan; this is the Skill that prevents it.

  3. 3
    simplify

    Anthropic-authored and bundled: a quality-only pass for reuse, simplification, and efficiency. It is the complement to a bug hunt, not a substitute — the two reviews find different things, and most teams only run one.

  4. 4
    skill-creator

    The compounding pick. Claude Code's real advantage is that your own conventions become installable, and skill-creator writes the SKILL.md properly — including the description field, which is the single biggest reason a hand-written Skill never triggers.

  5. 5
    pdf

    Deterministic scripts for the format most business workflows actually run on — extraction, form fill, watermarking, encryption. It handles the edge cases improvised Python quietly gets wrong, like encrypted files and scanned pages.

  6. 6
    debug

    Turns "it is broken" into a hypothesis-and-test loop instead of a sequence of speculative edits. Cheap to install, and it changes how the agent behaves on the failure modes that waste the most time.

Limits

What this agent cannot do

Claude Code has no ceiling worth flagging on Skill support — it is the reference implementation, and all 191 catalogued Skills are verified here. The real constraints are yours to set: Skills marked action-taking can run shell commands, hit paid APIs, and touch live systems, so read the capability level before installing one, and prefer project-scoped installs for anything that encodes repo conventions. Note also that a large installed Skill set costs frontmatter tokens at every session start; install deliberately rather than exhaustively.

FAQ

Common questions

How many Skills should I install in Claude Code?

Enough to cover the jobs you actually repeat — typically five to fifteen. Every installed Skill costs roughly 100 tokens of frontmatter at session start whether it triggers or not, and a crowded set also makes triggering less precise because several descriptions compete for the same task.

Do Skills work with Claude Code subagents?

Yes. Skills can be invoked inside the Agent and Workflow primitives, including in parallel across several subagents. A subagent inherits the installed Skill set, so orchestration Skills and task Skills compose.

Which Skills ship with Claude Code already?

Several Anthropic-authored ones are bundled, including code review, simplify, verify, init, security review, and the document Skills (docx, pdf, pptx, xlsx). Check your session's available-skills list before installing a duplicate.

Should Skills be committed to the repo?

Commit the ones that encode conventions for that codebase — review standards, docs structure, deploy checklists — under .claude/skills/. Keep general-purpose Skills user-wide in ~/.claude/skills/ so they follow you between projects.

Why is a Skill not triggering?

Almost always the description. The agent matches the task against the description text, so vague verbs and no concrete triggers give it nothing to match. Rewrite the description to name the artefacts and situations that should fire it, then start a new session.

Other agents

Same question, different agent

New to Skills? What is a Claude skill · How to install · How to audit a skill first