{"title":"State De-duplication: Breaking Execution Loops","region":"Global","category":"Reliability","description":"Preventing agents from repeating identical action-input pairs.","lastUpdated":"2026-02-24","source":"https://github.com/langchain-ai/langchain/issues/13647","steps":["Hash the last 3 'Action + Input' pairs in session state.","Compare candidate actions against the hash set before execution.","Trigger a temperature spike (0.7+) if a repeat is detected.","Mandate a 'State Reflection' turn to identify the failure cause.","Hard-terminate the chain after 2 subsequent failed unique attempts."],"url":"https://checklist.day/state-deduplication-breaking-execution-loops"}