Agent Evolution Patterns

Agent evolution closes the continuous-learning feedback loop: validated instincts flow from sessions into agent instructions, making agents smarter over time without touching the versioned agent files in the repo.

When to Activate

Use this skill when:

• Understanding why an agent keeps making the same mistake across sessions
• Promoting a high-confidence instinct to a specific agent
• Auditing what a reviewer agent has learned (checking its overlay)
• Rolling back a bad instinct that changed agent behaviour for the worse
• Designing a new agent that should benefit from accumulated instincts

The Problem This Solves