Skill UI

This skill provides comprehensive best practices for managing data within AI agent workflows. It covers data classification (PII, PHI), prompt-level controls, secure memory usage, and isolation techniques. It ensures that agents adhere to global regulations like GDPR, CCPA, and HIPAA when handling sensitive or regulated data.

This guide details advanced architectural patterns for building complex AI systems using multiple Language Model instances (agents). It addresses the single-agent context bottleneck by distributing work, enabling parallelism, and facilitating specialization across different subtasks. Key patterns include Supervisor/Orchestrator, Peer-to-Peer/Swarm, and Hierarchical designs, providing guidelines for context isolation and effective coordination protocols.

This protocol defines the structured communication rules for multiple specialized AI agents simulating C-suite roles (e.g., CFO, CTO, CRO). It dictates invocation syntax, response formatting, and enforces strict constraints (such as loop prevention, call depth limits, and isolation rules) to ensure coordinated, non-contradictory, and reliable cross-functional analysis for executive decision-making.

A persistent memory kit designed for Claude AI agents. It maintains deep context across multiple sessions and projects using a two-layer architecture (hot cache and knowledge wiki). Features include safety hooks, multi-project context isolation, and a /close-day command for daily synthesis, ensuring agents remember complex information and continuously learn.

This skill provides a repeatable assessment for RAG systems, testing for critical vulnerabilities in vector stores and embedding models. It specifically addresses embedding inversion, cross-tenant data leakage, and knowledge-base poisoning, aligning with OWASP LLM08:2025. Use when performing authorized security audits of retrieval layers and multi-tenant isolation.

The Neon AI Gateway provides a single API endpoint and one credential to access a diverse catalog of frontier and open-source LLMs (including OpenAI, Anthropic, Google Gemini, Meta, Alibaba, and DeepSeek). This unified layer simplifies integration by eliminating the need for separate provider accounts and managing model-specific SDK complexities. It allows developers to switch between models seamlessly by changing only the model name, while also providing crucial branch-scoped isolation for development and CI environments.

Coordinate multiple agents as a team using work items with owners, agent Kanban state, branch/worktree isolation, control pane visibility, and explicit merge gates. Use for parallel agent fan-out and merging agent output into one product.

Secure LLM-powered applications against prompt injection, data leakage, and abuse using input validation, output controls, and tenant isolation. Implements OWASP LLM Top 10 mitigations with Python and Node.js examples.

This skill covers designing multi-agent systems to overcome single-agent context limits. It includes patterns like supervisor/orchestrator, swarm, and hierarchical structures. Key concepts involve context isolation, parallel execution, token economics, and specialization for complex tasks.

Design secure multi-tenant LLM hosting platforms on shared GPU inference infrastructure, covering tenant isolation, per-tenant quotas and rate limits, billing attribution, noisy-neighbor protection, and per-tenant policy controls for vLLM on Kubernetes.

This skill manages and stores approved team context securely using RuVector isolation. It ensures tenant-local data privacy by excluding sensitive credentials and personal information. During retrieval, it preserves provenance and treats data as untrusted. It prevents memory transfer between teams, maintaining strict security boundaries for AI team operations.