Skip to content
Second Brain

10 Claude Code Skills Worth Installing

The skills ecosystem got crowded fast. As of 2026-07 the ones with real traction fall into four buckets: process (make the agent plan before it codes), memory (survive context loss), domain expertise (frontend, security, testing), and context economy (spend fewer tokens per turn). The consistent community verdict is that skill count is not the win — 8–12 well-scoped skills cover most of a senior dev’s week, and every loaded skill pays a context tax whether it fires or not. Star counts below are from the GitHub API on 2026-07-25.

How installs work

Section titled “How installs work”

Two routes, both current:

  • Native plugin marketplace — /plugin marketplace add <owner>/<repo> then /plugin install <name>@<marketplace>. No extra tooling.
  • skills CLI — npx skills add <owner>/<repo> --skill <name>. Adds -g for global (all projects) instead of repo-local.

Skills execute arbitrary code in your environment. Read the SKILL.md before installing anything from a source you don’t recognise.

The 10

Section titled “The 10”

1. Superpowers — planning and TDD as a forced workflow

Section titled “1. Superpowers — planning and TDD as a forced workflow”

The big one (261k stars, obra/superpowers). Jesse Vincent’s framework makes the agent earn the right to write code: brainstorm, write a spec, get the spec adversarially reviewed by a subagent, write a plan, then execute. Ships ~14 skills including systematic-debugging, test-driven-development, using-git-worktrees, and subagent-driven-development.

Community read is roughly 60/40 positive. Praise: catches design mistakes before they become code. Criticism: over-engineered for small tasks, and it crowds the skill-listing budget — run /doctor after installing.

npx skills add obra/superpowers
/brainstorm "add real-time collaboration to the note editor"

2. handoff — compress a session before you hit the wall

Section titled “2. handoff — compress a session before you hit the wall”

From mattpocock/skills (188k stars, “Skills for Real Engineers”). Dumps the live session into a markdown doc you can feed to a fresh session or a different agent. Solves the single most common daily annoyance: context runs out mid-task and the next session starts blind.

npx skills add mattpocock/skills --skill handoff -g
/handoff wire the new frontmatter hook into CI

How it differs from /compact

Section titled “How it differs from /compact”

Both compress a conversation, but they are not substitutes — they operate on different things.

/compact handoff
Output Rewritten context, in-place Markdown file in OS temp dir
Session Same one continues Current session untouched; you start a fresh one
Trigger Manual or automatic at the context limit Manual only (disable-model-invocation: true)
Target Generic summary of everything Tailored to a stated next goal (the argument)
Portable Claude Code internal Any agent, any model, can read the file

The substantive differences:

  • Compaction is lossy in a direction you don’t choose. Auto-compact fires when you hit the wall and summarizes whatever happens to be in context. handoff takes an argument describing what the next session is for, and compresses toward that goal — “the conversation squeezed down to just its resumable core, so a fresh agent inherits the momentum, not the noise.”
  • handoff is artifact-aware. It explicitly refuses to re-summarize content already living in specs, plans, ADRs, issues, commits, or diffs — it references them by path or URL instead. Compaction has no such notion and will happily burn tokens re-describing a diff that’s already on disk.
  • It emits a “suggested skills” section, so the next session knows which tooling to load rather than rediscovering it.
  • It redacts. API keys, passwords, and PII are stripped before writing, because unlike compacted context this is a file that persists on disk.

Practical rule: /compact when you want to keep going in the same session and just need headroom. handoff when the session is done, poisoned, or about to die — and the next one should start clean with intent rather than with a summary of your dead ends.

3. grill-me — interrogate the plan before any code

Section titled “3. grill-me — interrogate the plan before any code”

Also mattpocock. Interviews you relentlessly about a plan until there’s shared understanding. Lighter-weight alternative to full Superpowers if you want the “think first” discipline without the whole spec/plan pipeline.

npx skills add https://github.com/mattpocock/skills --skill grill-me
Grill me on this feature spec before I start building

4. claude-mem — persistent memory across sessions

Section titled “4. claude-mem — persistent memory across sessions”

thedotmack/claude-mem (89k stars). Captures what the agent did, compresses it with an LLM, injects the relevant slice back at the start of future sessions. Local-only vector storage, works across Claude Code, Codex, Cursor, OpenCode.

Overlaps with Claude Code’s own memory files — worth it when you work across many sessions on the same codebase and keep re-explaining decisions.

npx claude-mem install

5. frontend-design — escape generic AI UI

Section titled “5. frontend-design — escape generic AI UI”

From anthropics/skills (164k stars), the official set. The most-cited single skill for output quality: pushes past default AI slop toward opinionated, production-grade interfaces. Give it an aesthetic direction, not just a component list.

npx skills add https://github.com/anthropics/skills --skill frontend-design
Build a landing page for a productivity app. Typographic focus, dark editorial aesthetic.

6. webapp-testing — drive a real browser

Section titled “6. webapp-testing — drive a real browser”

Official Anthropic skill. Uses Playwright to actually exercise your running app instead of asserting the code “looks right”. Consistently named the first official skill worth adding for anyone doing frontend work.

npx skills add https://github.com/anthropics/skills --skill webapp-testing
Test the login flow at http://localhost:4321 with valid and invalid credentials

7. skill-creator — build your own, properly

Section titled “7. skill-creator — build your own, properly”

Official. Scaffolds, evaluates, and iterates on custom skills. The recurring advice in every roundup: once you notice yourself re-explaining the same workflow, stop installing other people’s skills and write one.

npx skills add https://github.com/anthropics/skills --skill skill-creator
Use skill-creator to build a skill for running database migrations safely

8. Vercel agent skills — React/Next.js and accessibility rules

Section titled “8. Vercel agent skills — React/Next.js and accessibility rules”

vercel-labs/agent-skills (29k stars). web-design-guidelines audits UI code against 100+ accessibility and UX rules; react-best-practices applies 57 performance rules; composition-patterns replaces boolean-prop hell with compound components. Also ships react-native-skills, vercel-optimize, and deploy-to-vercel.

npx skills add https://github.com/vercel-labs/agent-skills --skill web-design-guidelines
/web-design-guidelines src/components/**/*.tsx

9. Trail of Bits security skills — real static analysis

Section titled “9. Trail of Bits security skills — real static analysis”

trailofbits/skills (6k stars). Smaller audience, high signal: an actual security firm’s CodeQL and Semgrep workflows for vulnerability detection and audit, rather than an LLM eyeballing code and guessing.

npx skills add trailofbits/skills
Run a security audit on the authentication module

10. Firecrawl — reliable scraping and search

Section titled “10. Firecrawl — reliable scraping and search”

firecrawl/firecrawl (156k stars). Gives the agent dependable web scraping, search, and browser automation instead of flaky ad-hoc fetching. The go-to when a task needs current external data — docs, changelogs, competitor pages.

npx -y firecrawl-cli@latest init --all --browser
Scrape the Astro changelog and summarize the last 30 days into changelog-summary.md

Honourable mentions

Section titled “Honourable mentions”
  • mcp-builder (anthropics/skills) — scaffolds MCP servers; reported to cut custom integration setup by ~80%.
  • Document skills — pdf, docx, xlsx, pptx from anthropics/skills. Dull until the job is a spreadsheet, then indispensable.
  • context-mode (19k stars) — sandboxes tool output for ~98% reduction; restores session state after a reset.
  • caveman (93k stars) — strips narration from replies, ~65% output token reduction. Already installed here.
  • code-simplifier (anthropics/claude-plugins-official) — readability-only cleanup of freshly written code, no logic changes.
  • coreyhaines31/marketingskills (42k stars) — 32 skills for CRO, copywriting, SEO. Non-engineering, but heavily used.

Try-list

Section titled “Try-list”

Ordered by likely payoff for this setup (Astro/Starlight, solo, frontend-heavy):

  • handoff — cheapest win, solves context death today
  • frontend-design — biggest quality delta on UI work
  • webapp-testing — verify the Starlight site in a real browser
  • web-design-guidelines (Vercel) — accessibility audit on components
  • grill-me — try before committing to full Superpowers
  • Superpowers — only if grill-me proves the “plan first” discipline pays off; run /doctor after
  • skill-creator — once a repeated workflow shows up worth encoding
  • Evaluate claude-mem against plain CLAUDE.md + memory files before adding it

Sources

Section titled “Sources”