Show HN: Shackle – Deterministic runtime governance for AI agents (github.com)

SHACKLE has introduced a groundbreaking runtime governance layer for autonomous AI agents, implementing the SP/1.0 conformance standard that mediates agent tool calls in real-time. This system actively prevents runaway token loops, unhandled errors, and budget excesses, ensuring that agents operate within predefined limits. Unlike other proposals, SHACKLE is a functional, verifiable standard that provides a clear decision framework—allow, deny, or halt for human input—backed by a comprehensive set of testable specifications. The standard is underpinned by a conformance model that ensures every action's validity based on system capabilities, creating a deterministic environment for AI decision-making. The significance of SHACKLE lies in its potential to rectify common failures in current AI agent frameworks, such as the "Loop of Death," by enforcing strict execution policies that mitigate excessive costs and idle processing. Its innovative design features a client-side implementation that requires minimal integration effort, making it accessible for developers. By providing a certification process that emphasizes reproducibility and transparency, SHACKLE addresses a crucial demand from enterprises and regulators for proof of operational integrity in autonomous systems, ensuring they only perform authorized actions consistently. This governance layer represents a vital advancement towards safer, more accountable AI operations across various frameworks like CrewAI, AutoGen, and LangGraph.