memory-uselisted

# memory-use — the consumption contract This skill answers ONE question: **how should an agent interact with the MemoryVault?** The kit ships 8 MCP tools and 8 per-type playbooks. The descriptions on each MCP tool already explain "when to call it." This skill is the **orchestration**: when to call which tool, in what order, with what expectations. ## The five operations (memory mode) ### 1. **Refer** — read what's already known For every question the user asks about their work, **call `memory_ask` FIRST**. Then answer using only what came back. ``` memory_ask(question="<user's question, lightly cleaned>", k=5, context="<recent conversation distilled to 1-3 sentences>") ``` **Always pass `context` when you have it.** It's not used for retrieval — it's persisted into the gap memory if the query comes back thin. The next /memory-refresh queue drain reads the context to inform its deep-dive into native MCPs, so the agent doing the fill has more than just the bare query string to work with. Without context, that downstream agent is guessing at what you (the asker) actually wanted. What to put in `context`: - What the user is trying to accomplish in this session (the goal, not the literal exchange) - Any entities/people/projects they've mentioned that aren't in the query string itself - Time references the user implied but didn't say ("this week", "since the all-hands") Keep it to 500-1500 chars. Not a transcript paste — a distilled summary. **Response sh