Self-Improving Agent (Proactive Self-Reflection)

This Skill turns user corrections and self-reflection into persistent, tiered memory stored under ~/self-improving/, ensuring future responses improve over time across sessions.

Tiered agent memory optimizes workflows by storing always-loaded HOT preferences, load-on-demand WARM project patterns, and archived COLD insights, compounding performance improvements for multi-step execution.

Self-reflection for quality assurance works by evaluating completed tasks against the original user intent, identifying areas for improvement, and logging those insights into durable memory for future use.

Yes, the agent memory storage requires a local file-based system under the ~/self-improving/ directory, automatically logging corrections with promotion rules after repeated confirmation counts.

Strict security boundaries prevent storing sensitive credentials or third-party personal data, ensuring only non-sensitive workflow patterns, preferences, and self-reflection insights are retained in memory.

Yes, transparent memory operations support exporting items, forgetting specific corrections, and reporting memory stats with source tracking expectations to manage what the agent retains.