Continuous Learning Agent

A meta-skill that enables AI agents to learn from experience and improve over time by separating journaled memory from policy changes that alter future behavior.

Core Concept

Traditional agents reset completely between sessions. This skill treats memory and learning as related but distinct operations:

• Journal / memory records what happened, what was tried, and what evidence exists.
• Learning / policy changes what the agent will do next time for a recognizable event class.

Do not call a session log, decision journal, or context note "learning" unless it produces a policy delta, threshold revision, banned move, or acquired pattern that changes future behavior.

Learning Mechanisms

Every learning loop has two layers:

1. Journal layer: episodic or semantic records used for auditability.
2. Policy layer: compact behavioral deltas used to improve performance on future tasks.