cat:learn-from-mistakes

To diagnose context degradation in CAT orchestration, analyze conversation length, token usage, and history reliability metrics. This process uses a structured root-cause analysis workflow to identify why subagents produce incorrect results in long transcripts.

A subagent produces incorrect results when long conversations cause context degradation. By measuring token usage and verifying conversation history reliability, you can trace the root cause and implement preventive actions to stop recurrence.

The best way to perform root-cause analysis on CAT orchestration mistakes is applying a structured RCA framework with A/B testing. It enforces history verification, documents mistakes, and collects context metrics to guide preventive actions.

Yes, you need token metrics and get-history access to analyze CAT mistakes. These dependencies are required to measure conversation length, verify history reliability, and accurately identify the root causes of context degradation.

Yes, you can use A/B testing within the RCA framework to prevent recurring errors in CAT orchestration. It compares context metrics across sessions to identify degradation patterns and guide stronger prevention strategies.

You should run a context degradation analysis on CAT sessions when transcripts are long or subagents produce incorrect results. Analyzing token usage and history reliability at this stage prevents further context loss.