Organising Claude Code Skills Into Plugin Marketplaces

Aight, so, continuing with the Claude Code ‘bits’ and my toolchain I’m putting together for working with Claude Code. Claude Code is very much integrated into my workflow with skills for this and that! Thing is the skills were dotted around in differing repos and I’d be copy pasting from one to the other and sometimes they were outdated and I’m scurrying around trying to find the last place I updated the specific skill I needed etc!

The svelte-claude-skills repo I made when trying to work out How to Make Claude Code Skills Activate Reliably now has ~140 GitHub stars, but, the skills in there I’ve not touched since mid December. 😅

Essentially I needed these skills and ‘bits’ tracked in git and to be able to sync them across projects and pull them when I needed them, rather than setting up sync between them all there’s plugins you can make for Claude Code to consume.

So, I spent some time reorganising the skills I use into some plugins available via two marketplaces!

• claude-code-toolkit
• svelte-skills-kit

Why two marketplaces?

“Why not just one?” and honestly, I considered it. But as I started grouping them a natural split emerged:

General Claude Code workflow ‘stuff’ vs svelte-specific skills.

Some tools are useful for me regardless of Svelte, the others were around performance optimisations, analytics, the forced-eval hook that makes skills actually activate. Others are specific to Svelte and SvelteKit development.

Bundling everything together would mean Svelte devs downloading a bunch of stuff they don’t need, and non-Svelte devs getting confused by runes documentation. Separation of concerns, innit?

claude-code-toolkit

This is the general productivity toolkit when I’m using Claude Code for any project.

• toolkit-skills - The forced-eval hook (84% skill activation vs 20% without it), plus skills for ecosystem guidance, research patterns, skill creation, session reflection, and plugin development
• performance - SQLite FTS5 indexing that dropped file search from 28ms to 3ms with BM25 ranking
• mcp-essentials - Setup guide for my recommended MCP servers
• analytics - Query your Claude Code usage from ccrecall’s database

# In Claude Code
/plugin marketplace add spences10/claude-code-toolkit
/plugin install toolkit-skills
/plugin install performance

svelte-skills-kit

This is where all my Svelte and SvelteKit knowledge lives. If you’re building with Svelte 5, this is the stuff I wish I’d had when I started.

• svelte-runes - $state, $derived, $effect, $props, $bindable and migration from Svelte 4
• sveltekit-data-flow - Load functions, form actions, fail(), redirect(), serialisation rules
• sveltekit-structure - Routing, layouts, error boundaries, SSR
• sveltekit-remote-functions - command(), query(), form() patterns
• svelte-template-directives - {@attach}, {@html}, {@render}, {@const} patterns
• svelte-deployment - Adapters, Vite config, pnpm, PWA setup
• layerchart-svelte5 - Chart patterns with Svelte 5 snippets
• svelte-components - Bits UI, Ark UI, Melt UI patterns

These skills get loaded when Claude detects you’re working on something relevant. Ask about form actions? sveltekit-data-flow activates. Working with $state? svelte-runes kicks in.

# In Claude Code
/plugin marketplace add spences10/svelte-skills-kit
/plugin install svelte-skills

SQLite

I’ve been reflecting on my tooling lately, and SQLite has become my go-to for nearly everything AI workflow related.

The performance plugin uses SQLite FTS5. The analytics plugin queries SQLite. My conversation logging (ccrecall) stores everything in SQLite - I’ve got a 186MB database tracking 2,555 AI sessions, 70k messages, and 19M tokens. 😅

Why SQLite for AI work?

1. Zero config - No server, just a file
2. Portable - Copy it anywhere, query it with any tool
3. Fast - FTS5 queries in 3ms, bulk inserts in 96ms for 5k records
4. Queryable - SQL is the universal language for data exploration
5. Everywhere - Python, Node, Bash, even MCP servers can use it

The pattern I keep seeing: AI tools generate tons of structured data (tokens, costs, tool calls, file operations). SQLite makes it a doddle to store, query, and analyse.

The ecosystem

These marketplaces are part of a broader set of tools I’ve built:

Tool	What it does
mcp-omnisearch	Unified search (Tavily, Kagi, GitHub)
mcp-sqlite-tools	Safe SQLite operations via MCP
ccrecall	Sync Claude Code transcripts to SQLite
mcpick	Toggle MCP servers on/off dynamically

They all work together. Use ccrecall to sync your transcripts, query them with mcp-sqlite-tools, search the web with mcp-omnisearch, and manage which MCPs are active with mcpick.

Getting started

If you want to try these:

For general productivity:

/plugin marketplace add spences10/claude-code-toolkit
/plugin install toolkit-skills

For Svelte development:

/plugin marketplace add spences10/svelte-skills-kit
/plugin install svelte-skills

Pro tip: Install toolkit-skills alongside svelte-skills. The forced-eval hook in toolkit-skills makes the Svelte skills actually activate when they should (84% vs 20% without it - I tested this extensively, see my post on skill activation).

Conclusion

I now have all my skills I use git tracked in a central place (two marketplaces) so I can have them in use at a top level (for every project I work on) and be activated reliably with the forced eval hook!

Using ccrecall and the reflect skill I’l able to improve and refine these skills as I use them in projects along with creating new skills and using things like the claude-skills-cli to enforce best practices creating them!