memory-management

# Memory Management Retrieve first. Write only what should persist. ## Core model - `MEMORY.md` — curated long-term memory - `memory/YYYY-MM-DD.md` — daily notes and running context - `memory_search` — find likely snippets - `memory_get` — read exact lines needed ## Retrieval When user asks about prior work, decisions, dates, people, preferences, or todos: 1. `memory_search` with a tight, specific query. 2. `memory_get` for the needed lines only. 3. Answer from retrieved text, not guesswork. 4. If results are weak or empty, say you checked. Good queries: `omar internship goal fall 2026`, `notion workspace page id` Bad queries: whole paragraphs, vague terms like `memory` ## Writing Write when user says "remember this," states a durable preference, or makes a decision that matters later. | Destination | What goes there | |---|---| | `MEMORY.md` | Standing preferences, durable facts, recurring workflows, important long-term decisions | | `memory/YYYY-MM-DD.md` | Day-to-day notes, progress, one-off events, session context | Rules: - smallest useful note - concrete facts over commentary - no secrets unless asked - do not promise memory unless written ## When not to use Skip retrieval when the answer is in the current message or task is current-turn only. Skip writing when the fact is trivial, belongs in project docs, or is temporary scratch. ## Answering rules - Cite file path/line when useful - Separate memory facts from inference - Flag stale memory - If unavailable, say so