Skill: Svelte Core Best Practices
by Svelte
The Svelte team's own guidance on writing modern Svelte 5 — when to reach for each rune, why effects are an escape hatch, and the reactivity traps models fall into.
The official Svelte team skill that wires an agent to the @sveltejs/mcp CLI for Svelte 5 docs lookup and an autofixer that validates components before you ship them.
Add this skill to your favorite AI agent in a few steps.
Learn how to use this skill with different AI agents.
Maintained by the Svelte core team in the sveltejs/ai-tools repository, this skill stops an agent guessing at Svelte 5 syntax by giving it three CLI commands it can actually run.
list-sections enumerates every Svelte 5 and SvelteKit documentation section with its title and path. get-documentation then pulls the full text of the sections that matter — so when the agent is unsure how $derived differs from $effect, it reads the current docs rather than recalling a Svelte 4 pattern from training data. svelte-autofixer analyses a component, either as inline code or a file path, and reports the issues it finds, with --svelte-version to target 4 or 5 and --async for async Svelte mode.
The workflow the skill prescribes is what makes it valuable: look up documentation when uncertain, run the autofixer while debugging, and always run the autofixer once more before finalising any component. That last step turns the skill into a validation gate instead of another prompt full of advice, which matters because Svelte 5's runes changed enough that models trained on older code confidently write syntax that no longer works.
One practical detail: when passing runes such as $state through the terminal you must escape the $ as \$ to stop the shell substituting it as a variable.
Ships in the official Claude Code plugin and in the Codex CLI, GitHub Copilot CLI and OpenCode plugins, or can be dropped into .claude/skills manually from the repository releases. MIT licensed.
by Svelte
The Svelte team's own guidance on writing modern Svelte 5 — when to reach for each rune, why effects are an escape hatch, and the reactivity traps models fall into.
by Senzing
Drives the Senzing MCP server through the real entity-resolution jobs: map a messy CSV to Senzing's format, scaffold SDK code in five languages, decode error codes, migrate V3 to V4.
by Anthropic
Guidance for distinctive, intentional UI design — aesthetic direction, typography, and choices that avoid looking like templated defaults.
by Google Labs (Stitch)
Google's official Stitch skill for distilling a project into a DESIGN.md — a semantic design system agents can read before they generate anything.