SAGE

Overview

SAGE is a reinforcement-learning framework in which agents accumulate and reuse skills across sequential rollouts. It makes skill acquisition part of an ongoing training loop rather than an occasional manual packaging step.

Why it matters

It matters because it pushes skill-library self-improvement past prompt tinkering and into a persistent learning regime. The promotion of new skills becomes part of the loop itself.

Distinctive trait

Its distinctive trait is cross-rollout skill accumulation: learned capabilities persist and shape future episodes instead of being relearned from scratch each time.

Relationships

Read SAGE with SkillX, memento-skills, and self-evolving-workflows. It also provides a useful RL-flavored contrast to the more artifact-driven library systems such as SkillFoundry.