Skill UI

SAELens provides a framework for training and analyzing Sparse Autoencoders (SAEs). SAEs decompose the dense, often polysemantic activations of large language models into sparse, monosemantic features. Use this when you need to discover the discrete, interpretable concepts a model has learned, study feature superposition, or analyze specific safety-relevant behaviors (like bias or deception) within deep neural networks.

Implements Speak-specific privacy controls to classify PII, manage audio consent/retention, log learning data, and ensure GDPR/CCPA export/delete workflows when integrating Speak language learning services.

Deploy Speak-powered language learning apps to Vercel, Fly.io, or Cloud Run, including platform-specific secrets setup, health-check endpoints, and environment configuration for production-ready audio infrastructure.

Guides configuring Speak across development, staging, and production by outlining environment-specific configs, secret management, feature flags, and health checks for reliable multi-environment deployments.

A specialist tool designed to reverse-engineer and extract structured specifications from existing, undocumented, or legacy codebases. It maps code dependencies, identifies undocumented business logic, generates API documentation, and helps create comprehensive architecture documentation by adhering to established formats like EARS. Ideal for onboarding new team members or documenting crucial system components.

Verifies that blockchain code precisely implements documented specifications by comparing logic, invariants, flows, and assumptions against whitepapers or design documents, flagging undocumented behavior, mismatches, and compliance gaps during audit engagements.

Guides authors through writing clear specs.md files that define goals, scope, and testing settings so the Sprint plugin's agents deliver focused features without drift.

Speeds up LLM inference by combining speculative decoding with Medusa multiple heads and lookahead techniques so chatbots, code assistants, and other real-time workloads can run with 1.5-3.6× lower latency while preserving target-model quality.

Automate Spotify workflows through the Composio MCP to manage playlists, search the catalog, control playback, retrieve track/album data, and inspect authenticated user profiles once SPOTIFY_* tools are enabled via https://rube.app/mcp.

Guides teams through the Sprint plugin’s convergent diffusion model, covering phases from loading specs to finalization with agent orchestration, parallel implementation, testing, and review so users understand when and how to run sprints.

This skill specializes in optimizing complex SQL queries, designing efficient database schemas, and diagnosing performance bottlenecks across various SQL dialects (PostgreSQL, MySQL, SQL Server, etc.). Use it when dealing with slow queries, complex data aggregation, advanced joins, or when needing to implement best practices like CTEs, window functions, and covering indexes. It provides deep analysis of execution plans (EXPLAIN/ANALYZE) to ensure peak data retrieval performance.

Guides how to craft detailed prompts for Google Stitch so it generates high-quality web and mobile UI designs, covering specificity, visual style, multi-screen structure, platform targeting, and iterative refinement.