Skip to content
ATAI Today Brief
HomeNewsConceptsGuidesToolbox
AboutSubscribeUA
Subscribe

AI Today Brief

The daily AI-engineering brief. Built in public. EN · UA.

XTelegramLinkedInYouTubeRSS

Follow AI Today Brief on LinkedIn for daily AI-engineering updates and the weekly “5 shifts that changed how developers work” PDF.

Explore

NewsDigestsConceptsGuides

Company

SubscribeAdvertiseAbout

Legal

Editorial policyAI disclosurePrivacyTerms

© 2026 AI Today Brief. All rights reserved.

  1. Home/
  2. News/
  3. Agents & MCP/
  4. SHACKLE Releases SP-1.0 Circuit Breaker Specification to Prevent AI Agent Loops
Agents & MCP

SHACKLE Releases SP-1.0 Circuit Breaker Specification to Prevent AI Agent Loops

July 26, 2026· 5 min read
OKCurated by Oleksandr Kuzmenko, AI Product Engineer·Updated July 26, 2026·Sources cited on every story
AI-assisted · editor-reviewed·How we use AI
SHACKLE Releases SP-1.0 Circuit Breaker Specification to Prevent AI Agent Loops

SHACKLE introduced SP/1.0, an open-source specification and runtime daemon designed to act as a circuit breaker for autonomous AI agents. By executing deterministic checks against budget limits, tool repetition counters, and error amplification signals, it prevents rogue agents from burning thousands of dollars in retry loops.

Why it matters

Autonomous AI agents can easily get stuck in expensive retry loops, draining API budgets in minutes. SHACKLE provides a critical mediation layer to enforce safety limits before actions are executed.

TL;DR

  • 01SHACKLE is a runtime circuit breaker for AI agents, separating generation from execution authority.
  • 02It runs either as a sidecar daemon via gRPC/Unix sockets or as an in-process library for single-agent deployments.
  • 03Features 8 stacked check layers, including budget tracking, error amplification, and probabilistic denial.

Execution Authority vs. Generation Authority

Autonomous frameworks generate candidate action vectors, but generation should not imply automatic execution. SHACKLE runs as a sidecar daemon or an in-process library that mediates every tool invocation using a pure decide(state, call) -> Verdict evaluation loop.

Eight Stacked Guard Layers

The protocol evaluates incoming calls through eight distinct check stages:

  • Budget Tracking: Hard stops when session, agent, or org budgets drop to zero.
  • Repeat Limits: Rejection if repeat_count >= max_repeat_calls for identical parameters.
  • Error Amplification: Automatically decrements the allowed repetition threshold by 1 when HTTP error patterns (401, 403, 500, timeout, rate limit) are detected in parameters.
  • Probabilistic Denial: Introduces jitter when budget falls below 20% to prevent adversarial boundary probing.

Cryptographic Audit Trails

Every execution verdict produces an Ed25519-signed, append-only log entry. Communication uses Protocol Buffers envelopes signed with SHA-256 HMACs to enforce state integrity between client shims and the central daemon.

#SHACKLE#gRPC#Protocol Buffers#Redis
ShareShare on XShare on LinkedIn
← Previous storyPublicly Shared Claude Conversations Expose Embedded API Keys and Secrets to Search Engines

Related stories

  • Agents & MCPVending-Bench Exposes Autonomous Agent Collusion and Price Fixing in Business Simulations
  • Agents & MCPStructuring Claude Subagent Instructions for Minimal Context Overhead
  • Agents & MCPOfficial Model Context Protocol C# SDK v2.0 Releases with Stateless HTTP Transport
  • Agents & MCPNVIDIA Open-Sources NOOA: Python Object-Oriented Framework for AI Agents

Email digest

Get the morning AI brief

One email a day — the stories that matter for engineers, founders and tech leads. Human-edited, with links to primary sources.

  • ✓120+ sources scanned daily
  • ✓Edited by a human
  • ✓1 email per day
  • ✓EN + UA

By subscribing you agree to the privacy policy.