memory-asklisted

# memory-ask You have access to the user's MemoryVault. Whenever the user asks anything that references their work — a person they know, a customer they have, a project they're on, a decision they've made — query the vault FIRST, then answer based on what comes back. ## When retrieval misses — the gap auto-logs If `memory_ask` returns no useful results (top scores are low, or the results aren't related to the question), the MCP server **already logs a feedback memory** with `tags: [coverage-gap, retrieval-thin]` into the vault (see `log_retrieval_gap.py`). The response includes a `gap_logged: <mem_id>` field when this happens. You don't have to do anything to capture the gap — it's captured. But if you have new context that could enrich it, call `memory_update` on the gap memory: - Replace the templated body with a narrative about what you actually know (or what the user actually wanted) - Suggest where the missing content might be (which Slack channel, which person, which doc) - Set `enriched: true` ## Enriching stub gap memories when you encounter them If your retrieval returns a `mem_GAP_*.md` memory with `enriched: false` in its frontmatter, that's a stub waiting to become a real description. Read its `## Evidence` section — the kit pre-gathered the entity's context. Combine with your current session context. Then `memory_update` the gap with a grounded narrative. This is consumption-side enrichment: the gap was *captured* programmatically, but you (the consumi