Vault system

Routes vault meta-maintenance — the vault AS a system (its markdown corpus, knowledge graph, drift monitors, rule extraction, substrate sync). Not the content; the substrate that makes the content queryable and auditable.

Distinct from:

• Querying the vault to answer a question (use the knowledge-graph query tools)
• Bringing external data INTO the vault (use the ingest skills)
• Daily journal + coaching + pattern detection (those operate inside the substrate)

This is the SUBSTRATE layer. It maintains what every other skill uses.

Step 1. Identify the maintenance moment

Trigger	Skill / tool
Knowledge graph stale or never built for a corpus	graphify
Existing graph query (find concept / path / cluster)	your graph-query tool, or Obsidian graph view
Typed-file metadata extraction (books, meetings, people, articles, goals)	second-brain-mapping
Vault self-check (CLAUDE.md / Meta folder / skills / hooks / MCPs / journal index)	diagnose
Files edited many times in a short window (drift hotspot)	a drift-detection pass → a Drift Audit note
Recurring corruption pattern needs codification	codify it into a new rule file in your rules dir
Journals missing body / frontmatter context	backfill-journal-body-context
Vault types not yet set up (new vault or refresh)	setup-vault-types
Memory index getting long (truncation cliff)	consolidate-memory
Obsidian Bases / JSON Canvas / Markdown / CLI work	the matching obsidian:* skill

If unclear → ask ONE question. Don't guess.

Step 2. Pre-flight

1. Vault location: resolve the vault root (a VAULT_PATH / VAULT_ROOT env var, or the current directory).
2. Graph state: check whether a graph report exists and how fresh it is before answering graph questions.
3. Drift state: check the latest drift-audit output.
4. Rules state: list the rules directory.
5. Memory state: check the memory index length (keep it under the truncation cliff).
6. Git in a large vault: NEVER git add -A / . / unscoped git status on a big vault — full-tree walks are slow and lock-contending. Use explicit paths.

Step 3. Route by work-type

Knowledge graph

• Build a new graph from a corpus (code / docs / papers / images) → graphify
• Query an existing graph (search nodes / neighbors / path) → your graph-query tool
• Coverage audit (which corpus chunks were processed) → graphify's coverage audit

Second-brain mapping

• Extract structured metadata from every typed file → second-brain-mapping
• Apply wikilinks from detected entities → second-brain-mapping (wikilink phase)
• Cross-type insights (patterns no single file shows) → second-brain-mapping (insights phase)

Diagnostics + drift

• Vault self-check → diagnose
• Drift detection (files edited many times in 30d) → a drift-detection pass → a Drift Audit note

Rule extraction + memory

• Extract recurring patterns into a CLAUDE.md rule / rules file → codify manually into a new rule file
• Consolidate memory files (merge duplicates, prune the index) → consolidate-memory

Journal hygiene

• Backfill journal body / frontmatter context → backfill-journal-body-context
• Tag / taxonomy consistency check → manual against your canonical tag list

Vault setup + onboarding

• Setup vault types (frontmatter taxonomy) → setup-vault-types
• Setup the substrate on a new vault → the bootstrap flow

Obsidian tooling

• CLI (backlinks, search, unresolved, orphans, properties) → obsidian:obsidian-cli
• Bases queries (typed views) → obsidian:obsidian-bases
• JSON Canvas creation / edit → obsidian:json-canvas
• Markdown specifics (callouts, wikilinks, embeds) → obsidian:obsidian-markdown

Step 4. Post-flight (gates "done")

1. Graph rebuilt + report regenerated if graphify ran.
2. Memory index under the truncation cliff post-consolidation.
3. Drift hotspots resolved if the Drift Audit surfaced any.
4. New rules indexed in the rules directory.
5. Vault committed with explicit paths (never raw git add -A in a large vault).

Common mistakes

• git add -A in a large vault. Long walks + lock contention. Use explicit paths.
• Editing aggregator outputs directly (e.g. a "Last Session" or "Decision Log" roll-up). Edit the per-entry source files; let the aggregator rebuild.
• Hand-editing symlinked skill copies. Edit the source repo + re-sync.
• Re-reading the memory index when it is already in context.
• Routing daily journal / coaching / pattern detection here. Those are their own skills; this umbrella maintains the substrate, not the content.