Skip to content

The Best Claude Code Skills to Make Claude Work Like a Senior Engineer

A lean, audited checklist of the Claude Code skills worth installing — the ones that make Claude plan, test, review, and ship like a senior engineer.

Tuan Tran Van
10 min read
Updated 31/07/2026
Contents (16 sections)
  1. Step 1 — Measure your context budget before installing anything else
  2. Step 2 — Install one skill and confirm the engine loaded it
  3. Step 3 — Install skill-creator before any other skill (for members)
  4. Step 4 — Pick your skills from the catalog below (for members)
  5. Step 5 — Schedule a review every 30 days (for members)
  6. Catalog: engineering discipline (for members)
  7. Catalog: Matt Pocock's skills (for members)
  8. Catalog: frontend and design (for members)
  9. Catalog: testing, security, and safe execution (for members)
  10. Catalog: memory across sessions (for members)
  11. Catalog: extending Claude Code itself (for members)
  12. Catalog: work beyond coding (for members)
  13. Catalog: community bundles and directories (for members)
  14. Catalog: maintaining your skill library (for members)
  15. Troubleshooting (for members)
  16. What to do next (for members)

By the end of this page your ~/.claude/skills/ folder holds only the Claude Code skills you actually use, every description fits inside the context budget, and you have a 30-day review on the calendar so the folder stops growing back.

The catalog in the second half is where you pick those skills.

What you need first: Claude Code installed and signed in, a terminal, and write access to ~/.claude/. If you have no skills installed yet, skip Step 1 and start at Step 2.

Cover image: a lean Claude Code toolbox — around 8 curated skills instead of dozens installed

Step 1 — Measure your context budget before installing anything else

Run this inside a Claude Code session:

/doctor

It estimates what the skill listing costs you in context and names the biggest contributors. The listing gets a character budget of roughly 1% of the model's context window, and each entry is truncated at 1,536 characters — that is description plus when_to_use combined, while description alone is capped at 1,024 characters in SKILL.md. Go past the budget and the engine shortens, then drops, descriptions altogether, starting with the skills you invoke least. Both numbers are configurable: skillListingBudgetFraction for the budget, skillListingMaxDescChars for the per-entry cap.

The skill names stay in the list, but stripped of their descriptions they silently stop auto-triggering — you will assume the skill is broken rather than that you installed too many.

Diagram of the context tax: each skill's description fills the context window, contrasting a crowded junior setup of 30+ skills with a lean senior setup of about 8

A bloated skill folder charges you on every turn, whether or not a given skill fires:

  • Wasted budget — every description is reloaded into the system prompt each turn, spending tokens even when nothing fires.
  • Wrong tool picked — similar skills confuse the model, which attempts a database migration when you asked to format a file.
  • Shallower thinking — redundant tool schemas crowd out working memory, hurting bug detection and cross-module understanding.

Step 2 — Install one skill and confirm the engine loaded it

Pick one path and skip the other.

The official path: marketplace

bash
/plugin marketplace add anthropics/claude-code
/plugin install skill-creator@claude-plugins-official

The community path: clone into your skills folder

bash
git clone https://github.com/disler/agent-sandbox-skill.git ~/.claude/skills/agent-sandbox

Then ask in the session: "What skills are available?". The skill you installed must appear in the list; if it doesn't, the engine hasn't compiled it. Claude Code watches ~/.claude/skills/ live, so you never restart a session after adding or removing a folder.

Share this article

X / TwitterFacebookLinkedIn
← All AI applications