LLM & Agents

@effect/ai integration patterns for categorical AI composition, typed error handling, and production prompt pipelines. Use when building AI applications with Effect-TS, composing LLM calls with typed errors, creating tool-augmented AI systems, implementing structured output generation, or integrating multiple AI providers (OpenAI, Anthropic) with categorical composition patterns.

Expert code reviewer focusing on quality, maintainability, performance, and best practices

Expert skill for AI model quantization and optimization. Covers 4-bit/8-bit quantization, GGUF conversion, memory optimization, and quality-performance tradeoffs for deploying LLMs in resource-constrained JARVIS environments.

Craft effective Midjourney V7 prompts for any style — photography, illustration, anime, or artistic. Provides frameworks (7-Element, F.O.C.A.L.), parameter reference (--ar, --stylize, --sref, --cref), lighting/camera terminology, and V7-specific optimization. Auto-activates when writing Midjourney prompts, discussing MJ parameters, or creating AI image prompts. Triggers: Midjourney, MJ prompt, --ar, --stylize, --sref, style reference, character reference, image generation prompt.

Invoke for complex multi-step tasks requiring intelligent planning and multi-agent coordination. Use when tasks need decomposition, dependency mapping, parallel/sequential/swarm/iterative execution strategies, or coordination of multiple specialized agents with quality gates and dynamic optimization.

Guide for implementing Claude Code hooks - automated scripts that execute at specific workflow points. Use when building hooks, understanding hook events, or troubleshooting hook configuration.

Strategic patterns for Claude-to-Gemini delegation. Covers decision criteria, execution patterns, result parsing, and error handling. Use when determining if a task should be delegated to Gemini CLI.

Connect to HACS (Human-Adjacent Coordination System) for distributed multi-agent AI coordination. Use when coordinating with other Claude instances, managing shared projects and tasks, sending messages between AI agents, or accessing institutional knowledge. Enables Claude to participate in the distributed AI coordination network at smoothcurves.nexus.

Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

Custom hook creation for preventing unwanted behaviors in Claude Code. Triggers: hook, hookify, rule, block, warn, prevent, pattern, detect, unwanted behavior, dangerous command, coding standards

Type-safe development patterns for JARVIS AI Assistant

Use when implementing flexible content schemas using EF Core JSON columns, `OwnsOne().ToJson()` patterns, or designing dynamic field storage that avoids migrations. Covers JSON column configuration, LINQ querying of JSON properties, indexing strategies, and schema evolution patterns for headless CMS architectures.

Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.

Apply R&D framework to optimize prompts and context. Use when optimizing context window usage, reducing prompt size, delegating to specialized agents, or applying systematic context management.

Activate this skill when users need help with live streaming workflows including pre-stream setup, real-time source management during streams, audio verification, scene transitions, or stream health monitoring. Triggers include requests like "help me stream", "pre-stream checklist", "manage my stream", "check audio before streaming", "switch scenes during stream", or troubleshooting streaming issues. This skill orchestrates multiple tools to guide users through complete streaming sessions from setup to teardown.

Guide for designing effective tools for AI agents. Use when creating tools for custom agent systems or any AI tool interfaces. Provides principles for tool naming, input/output design, error handling, and evaluation methodologies that maximize agent effectiveness.

Operational skill hub for LLM system architecture, evaluation, deployment, and optimization (modern production standards). Links to specialized skills for prompts, RAG, agents, and safety. Integrates recent advances: PEFT/LoRA fine-tuning, hybrid RAG handoff (see dedicated skill), vLLM 24x throughput, multi-layered security (90%+ bypass for single-layer), automated drift detection (18-second response), and CI/CD-aligned evaluation.

Guide selection of appropriate prompt level for a task. Use when choosing between simple prompts and complex workflows, applying the seven levels framework, or matching task complexity to prompt investment.

Weekly Claude Code maintenance - cleans debug logs, shell snapshots, and old todo files to free up disk space. Run this weekly to keep Claude Code lean.

Merge agent work from agent branch to task branch with validation