learn

Instincts are atomic learned patterns extracted from accumulated tool-use observations and pipeline analytics. They guide agent behavior in future pipelines by applying confidence scoring and persona-based role tagging to improve decision making.

You create per-project instinct files by running a command to bootstrap instincts from current project observations. This establishes a dedicated instincts directory, updates confidence scores, and tracks the last_seen timestamp for each learned pattern.

Mining anti-patterns from review findings involves analyzing accumulated pipeline data to identify negative recurring behaviors. The system handles anti-pattern mining automatically alongside persona-based role tagging to tailor and refine future agent instincts.

Yes, existing pipeline observations are required to generate agent instincts. The system analyzes accumulated tool-use data, pipeline analytics, and review findings from your project to extract meaningful patterns rather than working from an empty state.

Confidence scoring for learned patterns updates dynamically as the system analyzes new pipeline observations. It tracks how frequently and recently an instinct was observed, updating both the confidence value and a last_seen timestamp per project.