Impasse Detector

Detect agent reasoning loops by analyzing transcripts and tool outputs within a lookback window to surface stagnation. This process identifies repetitive dialogue and action patterns, outputting a structured JSON status with confidence and recommendation fields.

Analyzing transcripts for agent stagnation requires a transcript path or direct content and a lookback window. Supplying these inputs allows the tool to evaluate recent history and output a structured JSON with status, confidence, reasons, and recommended action.

Yes, you can use PowerShell scripts to monitor long-running planning sessions for impasses. By processing transcripts and tool usage patterns, the script detects unproductive states and triggers escalation or delegation to avoid wasted tokens and time.

Automated loop detection is the process of monitoring dialogue and actions to surface impasses in ongoing planning sessions. It analyzes recent history using a lookback window to determine stagnation and triggers escalation or delegation to avoid wasted tokens.

Limitations of using transcript analysis for detecting unproductive agent states include reliance on a defined lookback window and accurate transcript data. Without sufficient recent history or direct content, the tool cannot effectively surface stagnation or recommend appropriate delegation.