memory-tree-navlisted

# Memory Tree Navigation regin's agent memory is mounted on the **approved topic graph** (`.regin/topics/topics/`, one JSON file per topic + `_meta.json`): every memory links to one or more topic nodes, and the nodes form a `parent_id` tree of ~11 top-level buckets (agent-memory, session-trace, eval-grading, rule-engines, webui, …) each fanning out to leaf topics. This skill walks that tree **coarse-to-fine** — you read node labels/blurbs and decide where to drill — instead of routing a query through embedding cosine. Use it when you'd rather navigate by structure than guess keywords, or to *complement* `recall` when a query feels familiar but semantic search misses. ## When to use this vs. plain `recall` - **Use this (tree nav)** to *orient*: "what does regin know about the trace subsystem?", "show me the eval/grading lessons", "browse what's under topic X". You get a map first, then content. - **Use `recall`** for a *specific* known question where you can phrase a tight query ("playwright stale backend"). It's one shot, semantic. - **Combine**: nav to the right subtree, then if it's thin, `recall` scoped to fill the long tail. Tree nav is precision-first; `recall` is recall-first. ## Prerequisite The three tools are served by the **`memory` MCP server** (`lib/memory/mcp_server.py`). They appear as `mcp__memory__index_root`, `…_index_expand`, `…_index_fetch`. The server is long-lived per session, so if you just added/changed these tools they only show up **after the se