The HyperAgent learning loop
Skyflo can learn knowledge from completed work, suggest reusable Skills and evaluate changes to route scoring. These mechanisms operate within fixed controls for authority, credentials, spending and stop conditions.
A learned change is a proposal until it passes its admission requirements. Having the mechanism does not prove that Skyflo has improved its routing on your work.
Inspect the record
Open Memory › Improvements to see proposed or adopted changes, their evidence and who judged them. Skill records show their content and scope. Route-scoring changes record the policy they replace and the evaluation that allowed adoption.
An empty record means no improvements have been recorded there. It is not a reason to manufacture a change. The routing loop can retain its existing policy when the evidence does not support a better one.
Revert a change
Choose Revert on a promoted improvement and read the confirmation. A Skill stops applying, or an earlier version resumes. Route scoring returns to its applicable previous policy or built-in scoring. The historical record stays marked as reverted.
Reverting an improvement affects future work. It does not undo file edits, deployments or other effects from past missions.
Boundaries that do not learn away
Learned Skills and route scoring cannot grant credentials, raise spending limits, waive approvals or alter the fixed root controls. A more capable route still runs within the authority given to the mission.