| name | update_vault_index |
| description | Keep vault_index.md synchronized with current vault structure |
📇 Update Vault Index Skill
Automatically sync .claude/vault_index.md with your actual MOCs/, Projects/, and Reading/ directories.
PURPOSE
The vault index is the entry point to your research graph (per CLAUDE.md rules). It must stay current to reflect:
- All MOCs (research areas)
- All projects categorized by status (Seed/Idea/Active/Writing/Finished)
- Quick access to recent papers
WORKFLOW
0. Read last-run timestamp
Read .claude/vault_maintanance.md and note the update_vault_index: timestamp.
Default: incremental scan — only process files modified after the update_vault_index: timestamp.
Use find Projects/ MOCs/ -newer <reference_file> to identify changed files.
To use a reference file: touch -t YYYYMMDDHHMI /tmp/vault_index_ref && find ... -newer /tmp/vault_index_ref
- If no files were modified: skip the scan and report "Index already up to date."
- Only run a full rescan if the user explicitly requests it or no timestamp exists yet.
- Always do a full scan of
Projects/ for status categorization even in incremental mode (status may have changed without file modification time changing).
1. Scan vault structure
Scan directories:
MOCs/ → extract all MOC file namesProjects/ → extract all project file names + status fieldReading/ → optional: list recent papers (last 5-10)
Extract metadata:
- For each project: read YAML frontmatter to get
status field
- For each MOC: extract 1-line description if present
2. Categorize projects
Group projects by status (maintain this order):
- Seed: status =
seed (early-stage ideas generated by Claude)
- Idea: status =
idea (refined projects approved by human)
- Active: status =
active (actively working on it)
- Writing: status =
writing (writing paper and finishing project)
- Finished: status =
finished (completed)
Sort alphabetically within each group.
3. Build index sections
Structure (preserve exactly):
# Vault Index
## Research Areas (MOCs)
- [[MOC Name]] - brief description (1 line)
- ...
## Seed Projects
- [[Project Name]] - brief description (1 line)
- ...
## Idea Projects
- [[Project Name]] - brief description (1 line)
- ...
## Active Projects
- [[Project Name]] - brief description (1 line)
- ...
## Writing Projects
- [[Project Name]] - brief description (1 line)
- ...
## Finished Projects
- [[Project Name]] - brief description (1 line)
- ...
Get descriptions from:
- MOC files: first non-heading paragraph or subtitle
- Project files: "Seed Idea" or "Current Working Idea" field
- Keep descriptions to 1 line max
4. Generate minimal descriptions
Rules:
- Extract from actual file content (don't invent)
- Max 1 line (70 chars)
- Use present tense
- Keep factual, avoid marketing language
Example:
- [[Persona Fingerprinting to Prevent Emergent Misalignment]] -
Preserve identity fingerprint during finetuning to prevent misalignment
5. Update vault_index.md
- Replace entire file (sections only, no other changes)
- Preserve markdown formatting
- Verify all wiki links use exact file names
- Verify no broken links
6. Update vault_maintanance.md
After writing vault_index.md:
- Run
date via Bash to get the current system time
- Read
.claude/vault_maintanance.md
- Update
update_vault_index: line to current date and time in format DD-MM-YYYY HH:MM
- Reset
papers_since_vault_index: to 0
- Reset
projects_since_vault_index: to 0
- Write the updated file
Example update:
update_vault_index: 10-02-2026 14:35
7. Verify output
✅ All MOCs present with descriptions
✅ All seed projects listed (status: seed)
✅ All idea projects listed (status: idea)
✅ All active projects listed (status: active)
✅ All writing projects listed (status: writing)
✅ All finished projects listed (status: finished)
✅ Projects in correct order (Seed → Idea → Active → Writing → Finished)
✅ Alphabetical within each section
✅ Wiki links match exact file names in [[brackets]]
✅ No orphan files referenced
TOKEN EFFICIENCY
Optimizations:
- Read only YAML frontmatter + 1-2 lines per file
- Use grep to extract status field:
status:\s*
- Single pass through each directory
- No deep content analysis
- Use file listing instead of reading metadata separately
Fast execution:
- ~50-100 tokens for full vault with 20+ projects
OUTPUT RULES
- Minimal, structural updates only
- One-line descriptions max
- Bullet format only
- No prose explanation needed in output
- Example: "Index updated: 4 MOCs, 5 seed, 0 idea, 1 active, 0 writing, 0 finished"
SELF-CHECK
✅ Did I read all MOCs, projects from actual directories?
✅ Did I extract status field from each project file?
✅ Did I categorize correctly (Seed/Idea/Active/Writing/Finished)?
✅ Did I maintain correct section order (Seed → Idea → Active → Writing → Finished)?
✅ Did I use exact file names in [[wiki links]]?
✅ Did I get descriptions from actual file content?
✅ Did I keep descriptions to 1 line max?
✅ Did I verify no broken links in output?
✅ Did I preserve vault_index.md structure exactly?
✅ Did I update vault_maintanance.md with current timestamp?
