Agent context · second-brain audit · 2026-07-05
Obsidian second brain — vs the one you already have
Short version: you don't need to build a second brain — you already run one, and it's more agent-active than a stock Obsidian vault. The gap isn't substance; it's the view & traversal layer. Here's the honest comparison and the gaps to fill.
📚 What you already have (grounded, counted just now): a git-tracked markdown vault — VAULT-INDEX.md + 12 module notes · 20 bug patterns · 11 accepted patterns · 54 ADRs · 43 research reports · 1,183 session records — with 1,174 files using [[wikilinks]]. Plus a second vault (personal auto-memory + MEMORY.md), semantic recall (/synexar-recall over transcripts + memory), and active context injection (load-compound-knowledge feeds bug-patterns/module-notes into every task). That's a second brain — a working one.
🔬 Head-to-head
| Dimension | Obsidian | Synexar today | Verdict |
|---|
| Markdown notes vault | ✓ core | ✓ same format (.md) | TIE |
| [[wikilinks]] between notes | ✓ | ✓ 1,174 files | TIE |
| Index / map-of-content | ✓ (manual MOCs) | ✓ VAULT-INDEX + MEMORY.md | TIE |
| Structured types + "why/how" frontmatter | △ tags/props | ✓✓ feedback/project/reference/decision | YOU |
| Semantic search (meaning, not keyword) | ✗ keyword only | ✓ /synexar-recall over transcripts+memory | YOU |
| ACTIVE injection into agent context | ✗ passive human reference | ✓✓ compound-knowledge + per-session load | YOU (big) |
| Git version / PR review / audit / goal-state protection | ✗ local files | ✓✓ fully versioned | YOU |
| Backlinks — "what links here" | ✓ automatic | ✗ forward links only, no reverse index | GAP |
| Graph view | ✓ visual knowledge graph | ✗ none | GAP |
| Broken-link / dangling detection | ✓ | ✗ dangling [[links]] tolerated silently | GAP |
| Single browsable surface | ✓ one vault app | ✗ two vaults + boards + CLAUDE.md, fragmented | GAP |
🕳️ The 4 gaps to fill
1 · Backlink + link-integrity index
Build reverse links from the 1,174 [[wikilinks]] ("what references this ADR/module?") and flag dangling links. Small script; makes the existing brain traversable + trustworthy. Highest value-per-effort.
2 · Graph / browsable view
A visual of the knowledge graph. Folds directly into ③ the /runtime auto-observability dashboard — the brain-map is just another panel of it.
3 · Unify the two vaults
Runtime memory (in-repo) + personal auto-memory (~/.claude) are separate brains. Bridge/map them so there's one surface — today you (and agents) must know which vault to look in.
4 · Agent link-traversal recall
Extend /synexar-recall (or an MCP) with "give me everything linked to [[X]]" — graph-aware retrieval to complement semantic search. This is the piece that actually makes the graph help agent context.
💡 The clever move (why you may not need to build much)
Point Obsidian at your existing vault — zero migration
Your vault is already markdown + [[wikilinks]] — that's Obsidian's native format. Open synexar-runtime/memory/ (and the auto-memory dir) as Obsidian vaults and you get backlinks, graph view, and one browsable surface for FREE, today — gaps 1–3 solved for the HUMAN with no code. Nothing to convert; it just reads your files. Obsidian becomes the human lens; the agent-active layer (recall, injection) stays exactly as-is.
🧭 Recommendation: Don't migrate to Obsidian and don't rebuild the brain — finish the one you have. (a) Now, free: open the existing vaults in Obsidian as a read-only human lens → instant backlinks + graph + browse. (b) For agents (part of ③): build the backlink+integrity index + link-traversal recall so the graph helps agent context, sovereign in /runtime. Adopting Obsidian as the brain would be a downgrade — it's a human-reference tool; yours is agent-active. Use it for the eyes, keep your engine.