evolve

To convert sprint failure patterns into invariants, analyze sprint execution histories to identify recurring structural issues across multiple sessions. The system extracts these patterns from progress, traces, and eval reports, then generates invariants with detection methods to prevent repeat issues.

Analyzing sprint traces for recurring structural issues involves running pattern extraction on sprint progress and eval reports across multiple sessions. This surfaces structural failures and generates invariants that can be explicitly injected into targeted skills to drive self-evolution.

To stop recurring failures from repeating across multiple sprints, identify patterns that have failed independently at least three times. The system then injects an invariant into the workflow with clear detection methods to actively prevent recurrence in subsequent runs.

Generating invariants from sprint histories requires structured failure data from sprint progress, traces, and eval reports across multiple sessions. This data must support cross-feature validation to accurately identify recurring patterns and injection targets for targeted skills.

To inject generated invariants into a worker prompt for future sprints, run the evolve analysis on your latest sprint data. This generates the initial invariants which you then load directly into the worker prompt to guide and constrain future runs.

Limitations when using cross-feature validation for sprint self-evolution include the dependency on structured failure data availability. Without explicit injection targets and comprehensive eval reports across multiple sessions, the system cannot accurately identify recurring patterns or generate effective invariants.