Agent skills by task
Written November 6, 2026
Agent skills grouped by the job they do, instead of by name. Every skill below links to its own page, which carries the install command and a trust score from a full code review.
Review and check first
- Give Claude Code a repeatable plan, build and review workflow — install Superpowers: A repeatable plan, build and review workflow for Claude Code.
/plugin install superpowers@claude-plugins-officialTrust score 6/10 - NVIDIA SkillSpector — scan a skill before you install it: Security scanner for AI agent skills — static checks plus an optional model-based semantic pass.
uv tool install git+https://github.com/NVIDIA/skillspector.gitTrust score 6/10
Research
- Get current library docs instead of the agent guessing APIs — install Context7: Current library docs on demand, instead of the agent guessing APIs.
npx ctx7 setupTrust score 5/10 - Give Claude Code live web search and research (Exa) — install guide: Live web search and research for Claude Code.
claude plugin install exa@claude-plugins-officialTrust score 7/10
Remember past work
- Make your coding agent remember past work across sessions — install Claude-mem: Remember past work across sessions.
npx claude-mem installTrust score 5/10
Draft and build
- Get reusable engineering workflows for Cursor and Claude Code — install the Matt Pocock skills: Reusable engineering workflows for Cursor and Claude Code.
claude plugin install mattpocock-skills@claude-plugins-officialTrust score 6/10 - Make AI-generated interfaces look designed, not generic — install the taste skill for Claude Code: Design guidance so AI-generated interfaces look designed, not generic.
npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend"Trust score 6/10 - Universal Modder — mod a PC game you own with an agent: Game-modding skills, the um command-line toolkit, and a shared knowledge base.
uv tool install git+https://github.com/rehan-remade/universal-modderTrust score 5/10
How the trust score works
Each skill page scores the skill out of 10 against the same three checks: what the skill's files tell the agent to do, what happens when you install it, and how exposed it is to being steered by untrusted content. A score is given only where the underlying code was fully reviewed. The full method is on the skills directory.
Back to: Agent skills and plugins, checked before listing · Coding playbooks hub · Also: Agent Skills has a neutral home — and still no security story