Loop Detection

Detects when agents get stuck repeating the same sequence of actions. Uses multiple detection methods from hash-based to semantic clustering.

Examples

• Agent calls search_tool("weather") 15 times in a row with identical queries
• Agent A asks B for clarification, B asks A, creating endless back-and-forth
• Agent paraphrases the same response 8 times using different wording
• State oscillates between two values without converging on a solution

Detection methods

Structural Matching

Detects repeated action sequences via substring matching

Hash Collision

Identifies identical state hashes indicating no progress

Semantic Clustering

Groups semantically similar messages using embeddings

Summary Whitelisting

Distinguishes recap/progress patterns from genuine loops

Calibration accuracy

From the Pisama calibration set. See detector scoreboard for the full table.