memory-discipline

What problem does it solve? Agent memory only pays off when reads happen before work starts and writes happen at decision points; without a disciplined loop, agents rediscover known facts and lose the reasoning behind settled decisions. ## Core Features & Use Cases - Recall Before Work: Run a project-scoped memory_smart_search as the first tool call on any nontrivial task to surface prior decisions and constraints. - Save at Decision Points: Persist each settled decision with its reason, 2-5 specific concepts, and real file paths the moment it resolves, rather than batch-saving at session end. - Corrections Become Lessons: Hand off user corrections to the lesson skill so they carry confidence and resurface before similar work. - Use Case: Before modifying an auth refresh flow, search memory for prior auth decisions; when you settle on cursor pagination over offset, save the choice with the reason that offset scans broke past 100k rows. ## Quick Start Ask the agent to search project memory for the current task topic before reading any code, then save each decision with its reason as it settles.