Official Wardstone SDK for LLM security, prompt injection detection, content moderation, and AI guardrails
io8l5b
Sandboxing system for secure tool execution (Seatbelt/Landlock) - magical containment
Task decomposition and plan mode for AI agents - planning the attack
Hierarchical agent orchestration system - the scheming group
Checkpoint and undo system for AI agent sessions - safe burrow to return to
MCP (Model Context Protocol) connection manager - sneaking connections to the outside
Tool registry and execution system for AI agents - goblin loot and treasures
Protocol types for agent communication - signals between goblins