claudekit-skills: A 40+ Skill Plugin Marketplace for Claude Code

If you want many skills at once rather than cherry-picking one, mrgoonie/claudekit-skills ships a Claude Code Plugin Marketplace grouped by domain — install only the categories you need.

Why This Skill Matters

The README frames the project as a Claude Code Plugin Marketplace (it announces a "NEW [Dec 2025]" switch from manual git clones to marketplace installs), with grouped install paths. The skills.json description for this repo is the line "All powerful skills of ClaudeKit.cc!" — that is how upstream categorized it on the leaderboard, not a phrase the README uses to describe itself.

The repository has 2.2k stars on GitHub. Note: the GitHub metadata does not declare a license (the license field is null) — the README states "I've been spending 6+ months to dig into every aspect of Claude Code so you don't have to" and points users at the commercial ClaudeKit.cc bundle for "advanced regulated-industry skills, analytics dashboards, or tailored onboarding". The marketplace skill pack itself is the free tier.

The skill catalog spans many domains. The README groups them by category; the marketplace mirrors this with named plugins:

• Authentication & Security — better-auth
• AI & Agent Development — context-engineering, google-adk-python
• Backend Development — backend-development
• AI & Machine Learning — ai-multimodal, context-engineering
• Design & Aesthetics — aesthetic
• Web Development — web-frameworks, ui-styling, frontend-design, frontend-development, threejs
• Browser Automation & Testing — chrome-devtools, web-testing
• Cloud Platforms & DevOps — bunny, devops
• Databases — databases
• Development Tools — claude-code, mcp-builder, mcp-management, repomix, media-processing
• Documentation & Research — docs-seeker
• Code Quality & Review — code-review
• Debugging & Quality — debugging/defense-in-depth, debugging/root-cause-tracing, debugging/systematic-debugging, debugging/verification-before-completion
• Document Processing — document-skills/docx, document-skills/pdf, document-skills/pptx, document-skills/xlsx
• E-commerce & Platforms — shopify
• Payments & Billing — payment-integration
• Problem-Solving Frameworks — problem-solving/collision-zone-thinking, problem-solving/inversion-exercise, problem-solving/meta-pattern-recognition, problem-solving/scale-game, problem-solving/simplification-cascades, problem-solving/when-stuck
• Advanced Reasoning — sequential-thinking
• Visualization & Diagramming — mermaidjs-v11
• Meta Skills (from Anthropic) — skill-creator

Recent additions (per the README's "Update" notes): an "mcp-manager" subagent plus "mcp-management" skills for context-budget management with MCP servers, and a refactor of the skills themselves (see REFACTOR.md).

Installation

The README's recommended path is the marketplace route, replacing manual clones:

/plugin marketplace add mrgoonie/claudekit-skills
/plugin install ai-ml-tools@claudekit-skills

The marketplace ships plugins as installable bundles — you can grab only the categories you need rather than the whole 40+ skill catalog at once.

The legacy manual route still works but is marked deprecated in the README:

git clone https://github.com/mrgoonie/claudekit-skills
cp -r claudekit-skills/.claude/* /path/to/your/project/.claude/

The README warns manual installs are deprecated in favor of marketplace for automatic update flow.

Real Workflow: Stand Up Context Engineering + MCP Management for a New Project

You start a new project and want the agent to handle context budgets and MCP servers without you hand-curating either.

Step 1. Add the marketplace and install two plugins:

/plugin marketplace add mrgoonie/claudekit-skills
/plugin install ai-ml-tools@claudekit-skills
/plugin install platform-tools@claudekit-skills

Step 2. The context-engineering skill (under ai-ml-tools) covers context fundamentals, degradation patterns, optimization techniques (compaction, masking, caching), compression strategies, memory architectures, multi-agent patterns, LLM-as-Judge evaluation, tool design, and project development. The README points at it for designing agent architectures, debugging context failures, optimizing token usage, implementing memory systems, or building multi-agent coordination.

Step 3. The mcp-management skill (under platform-tools) plus the mcp-manager subagent address the "MCP context bloat" problem — when an agent pulls in many MCP servers, naive tool listing blows up the context budget. The MCP_MANAGEMENT.md doc in the repo spells out the discovery-then-execute pattern the skill teaches.

Step 4. Once both are installed, ask the agent a real prompt and watch it discover MCP capabilities on demand rather than listing every tool's full schema up front.

Real Workflow: Pick a Debugging Skill by Failure Mode

Your tests fail intermittently and you do not know which layer introduced the regression.

Step 1. Look at the four debugging/* skills in the catalog. Each maps to a different failure shape:

• debugging/systematic-debugging — four-phase framework ensuring root cause investigation before fixes
• debugging/root-cause-tracing — trace bugs backward through the call stack to fix at the source, not the symptom
• debugging/defense-in-depth — validate at every layer data passes through
• debugging/verification-before-completion — run verification commands and confirm output before claiming success

Step 2. Install only the skill that matches the failure:

/plugin install debugging-tools@claudekit-skills

Step 3. Phrase the request to the agent so the matching skill triggers: "trace this backward through the call stack" routes to root-cause-tracing; "we never finished verifying before claiming success" routes to verification-before-completion.

Step 4. Combine with code-review when you want a review pass before declaring the fix done — the README lists this as "essential for subagent-driven development, pull requests, and preventing false completion claims".

Tips

• Install only the plugin categories you need — picking the full marketplace inflates the project's discovery surface and slows activation.
• The mcp-management skill plus mcp-manager subagent are the README's recommended fix for MCP-induced context bloat; install together.
• Treat the four debugging/* skills as a small menu, not a stack — install the one that matches your failure shape, not all four at once.
• The README points at REFACTOR.md and MCP_MANAGEMENT.md for recent structural changes; read them before opening older issues.
• For the commercial regulated-industry / analytics bundle, the README points at ClaudeKit.cc — the GitHub repo is the free tier.
• The README's marketplace section names twelve plugin categories, from ai-ml-tools and web-dev-tools to research-tools and specialized-tools — install by category name instead of hunting individual skills.

When Not to Use This

If your agent is not Claude Code, the marketplace plugins in this repo do not apply — they are Claude Code-targeted. If you only want one focused skill (say, a single design or payments helper), cherry-pick a smaller repo rather than installing this 40+ skill bundle. If you need an MIT-licensed skill, this repo's GitHub metadata does not declare a license; check the individual subdirectories before redistribution. The one named provenance in the README is skill-creator, listed as a meta skill from Anthropic — the other entries do not declare theirs.

See the leaderboard for more skills.