v0.5.0 — startup briefs, glance-view recall, and a Recently reminder

Startup memory now carries a lightweight "what is this" layer so the assistant can see a note without loading its body, and reminds you what you worked on recently.

Highlights

• Brief on every save — each memory leaf stores a deterministic one-line brief in frontmatter (first ≥3-word heading → first sentence; daily notes use the first captured item's title). No LLM call, injection-safe. Older leaves get a brief computed on the fly at read time.
• sections glance view — search_memory / recall_lessons accept sections:["frontmatter"] to return brief + type + status/progress + tags + priority with no body — a cheap way to scan hits without spending context. Omit it for the usual excerpted body (byte-identical to before).
• 🧠 "Recently — last N days" reminder at session start — dated bullets with clickable file:// links, context-budget bounded — plus a tighter plan list (top 1–2, in-progress first). New settings: recall.recentActivityDays (default 3, 0 disables) and recall.planContextMax (default 2).
• Sharper capture bar in the distiller prompt so low-signal notes are skipped (the zero-atom skip is unchanged).

Upgrade

Backward-compatible — no migration, no config removed, no runbook. From your project root:

git -C .llm-wiki-memory/src fetch origin && git -C .llm-wiki-memory/src merge --ff-only origin/main
( cd .llm-wiki-memory/src && npm install --no-audit --no-fund )
./.llm-wiki-memory/src/bootstrap.sh

Then reconnect the MCP server (or restart your session) so the running server loads the new code. Existing notes without a brief get one computed on the fly.

Full test suite: 1050 passing.