A SaaS platform that monitors and prevents infinite loops in AI agents through automated protocols detecting and halting anomalous execution states.

Repository: Cogni-AI-OU/cogni-ai-agents Author: github-actions[bot] ### Issue Analysis: Agent Runaway Tool Call Loop Context: In a recent build (EA31337/ansible-role-metatrader job 73077366532), the agent entered an infinite tool call loop. The log indicates it aggressively created read requests for recursive or hallucinated file extensions, such as: molecule/resources/playbooks/verify.j2.yml.j2.yml.j2.yml.j2.yml.j2.yml Root Cause: - Lack of self-correction or arrest mechanisms when the agent generates or attempts to read recursively identical strings or paths. - AGENTS.md and FLOWS.mmd currently lack explicit guidance on identifying runaway state (such as tracking consecutive identical failures or progressively expanding string patterns). Proposed Improvements: 1. Update FLOWS.mmd to introduce an Agent_Loop_Arrest_Protocol that triggers a hard stop after N consecutive recursive failures. 2. Update AGENTS.md to enforce self-reflection rules that abort vectors with repetitive outputs yielding no progress. Impact: Implementing these circuit breakers will prevent massive token burn and timeout failures caused by unhandled agent hallucination loops.