What problem does it solve? Agents repeat the same mistakes across sessions because they lack a persistent, structured memory of past execution gaps and success patterns. This Skill enforces a disciplined read-before-work and update-after-work loop so each agent accumulates conditional, actionable improvements over time. ## Core Features & Use Cases - Pre-task memory loading: Reads the agent-specific markdown file under .ept/self-improvement/ before any planning, tool use, or reply, applying matching Condition/Action entries. - Post-task review and update: Converts observed execution gaps or reinforced success patterns into improvement entries, then adds, strengthens, replaces, or skips them based on similarity checks. - Deduplication and strengthening rules: Prevents vague or duplicate lessons by requiring conditional, enforceable Do/Don't actions and rewriting weak entries into precise ones. - Use Case: A QA agent repeatedly misses edge-case checks. After a task, the skill stores an entry like "When reviewing test plans, do verify boundary inputs explicitly", which the agent loads and applies in every future task. ## Quick Start Use the self-improvement skill to load my agent memory before starting this task and update it with lessons learned when finished.
