Self-Improving Agent (Proactive Self-Reflection)

What problem does it solve?

Agents often repeat mistakes and fail to compound learning across sessions because corrections and lessons are not captured, confirmed, and applied in a structured way. This Skill provides a disciplined, local memory system and workflows so an agent can log corrections, confirm patterns, and apply lessons reliably without manual maintenance.

Core Features & Use Cases

• Tiered Local Memory: HOT/WARM/COLD storage under a local directory to control what is always loaded versus loaded on demand.
• Correction Tracking & Promotion: Automatic logging of corrections, counters to detect repeated patterns, confirmation flows after 3 occurrences, and promotion to persistent preferences.
• Operational Safety & Transparency: Explicit security boundaries, audit/export and deletion flows, and clear source citation when applying learned rules.
• Use Case: When a user repeatedly corrects formatting or workflow choices, the agent logs the corrections, prompts for confirmation after repetition, and applies the confirmed preference automatically in future sessions.

Quick Start

Before starting non-trivial work, read ~/self-improving/memory.md and any relevant domain or project files, then log any corrections to ~/self-improving/corrections.md immediately.