vault-prep
/vault-prep
Przygotowuje vault Obsidiana do vault2kg: zamienia frontmatter na linki i relacje bez udziału LLM.
- Autor
- felixvew ↗
- Repozytorium
- vault2kg ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 12.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 4
- Skrypty
- 2
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 2 skrypty w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: scripts/apply_plan.py, scripts/profile_vault.py
Co robi i kiedy
Przygotowuje vault Obsidiana do vault2kg: zamienia frontmatter na linki i relacje bez udziału LLM.
Wyzwalacze z opisu
Oryginalny opis (EN)
Optional preparation of an Obsidian vault for vault2kg. Rewrites its frontmatter so chosen facts become certain relationships without an LLM; turns property values that name things (a team, a system, a risk level) into [[links]], splits combined values, adds note IDs as aliases and creates hub notes with definitions the user gives. Use when the user asks to rewrite a vault's frontmatter into links for vault2kg, or needs a fact complete in the graph that the vault guide's structure doesn't deliver. For general advice on how to structure notes, use VAULT_GUIDE.md instead.
Jak działa
- Rules
- Steps
- Example
- Model
Podgląd SKILL.md
---
name: vault-prep
description: Optional preparation of an Obsidian vault for vault2kg. Rewrites its frontmatter so chosen facts become certain relationships without an LLM; turns property values that name things (a team, a system, a risk level) into [[links]], splits combined values, adds note IDs as aliases and creates hub notes with definitions the user gives. Use when the user asks to rewrite a vault's frontmatter into links for vault2kg, or needs a fact complete in the graph that the vault guide's structure doesn't deliver. For general advice on how to structure notes, use VAULT_GUIDE.md instead.
---
# Vault prep for vault2kg (optional)
**Start with the vault guide** (`guides/VAULT_GUIDE.md` in the vault2kg repository). For most vaults the better fix is in the notes: a `**Label:** value` line in the text for each fact that should become a relationship. That's how a real 88-note vault got its services linked in 82 of 88 notes, with no rewrite. If the notes come from a converter, change the converter, not the vault.
This skill is the other way. vault2kg turns a property into a relationship only when its value is a link (`team: "[[Imaging]]"` becomes a `TEAM` edge for every note that has it). This skill rewrites the frontmatter so chosen facts are links: certain, without an LLM, codes like `R3` included. The vault stays a normal Obsidian vault, and the links also appear in Obsidian's graph view and backlinks.
**The trade-off.** Linked frontmatter values reach the extraction prompt as known names. A small model (4B) then mostly repeats those names and misses new ones in the text: on the demo vault, MediaPipe and ViTPose were missed after the rewrite and found without it. So recommend this skill when the user runs a strong model, or needs a particular fact (such as a risk code) complete. Tell the user about the trade-off before planning.
## Rules
- **Only frontmatter changes.** Never edit, move or delete note text. Body text is the user's; vault2kg guarantees it reaches the graph unchanged.
- **Nothing is written without a backup:** a clean git repository, or `--backup DIR`. `apply_plan.py` refuses otherwise. Never work around that.
- **Plan first, then ask.** Show the user the plan and wait for a clear yes before `--write`.
- **Never invent definitions.** Hub note text comes from the user or from the vault's own source documents. Without a definition, no hub note: an unresolved link works fine in Obsidian and in vault2kg.
- **Use the scripts** for every change, not hand edits: they are repeatable and safe to run twice.
## Steps
1. **Profile.** Run `python scripts/profile_vault.py <vault>`. It lists every property with its number of notes, distinct values, values shared by several notes, a suggestion and examples. Read a few notes too, to see what the values mean.
2. **Draft the plan** (a JSON file, all parts optional; see `example-plan.json`):
| Part | For | Typical properties |
|---|---|---|
| `link` | values that name a thing several notes share | team, service, department, system, tool, client, risk level, status |
| `split` | one text holding several values: `"A AND B"`, `"A; B"` | data sources, systems, owners |
| `rename` | spelling variants of one value, fixed before linking | `"Imgaging": "Imaging"` |
| `alias` | an ID per note that the text uses to refer to it ("see 2.1.1") | item_id, ticket, code |
| `hubs` | a note for a link target, with a definition the user gives | `"R3": "Risk level 3: …"` |
Leave as properties: numbers and scores, dates, free text, values that differ in every note, and values (nearly) every note shares, such as a source file. Those last ones would link everything to everything. In Neo4j, properties stay filterable as `fm_*`.
**Linked properties stop being plain properties.** In Neo4j, filter them through the edge instead:
`MATCH (n:Note)-[:RISK]->(r {name: 'R3'}) RETURN n.name` rather than `WHERE n.fm_risk = 'R3'`.
3. **Show the plan** as a table: property, action, number of notes, example before and after. Ask about anything unclear, such as which values are spelling variants and what a code like `R3` means if a hub note is wanted.
4. **Dry run.** `python scripts/apply_plan.py <vault> plan.json` prints what would change and one example note, and writes nothing. Show the user the result.
5. **Write**, after a clear yes. In a git repository with nothing uncommitted: `--write`. Otherwise: `--write --backup <new folder>`.
6. **Check.** Run apply_plan again: it should report 0 notes to change. Then build the graph (`python src/vault2kg.py <vault>`, or `--no-llm` for a quick structural check) and confirm the new edges exist for every note that has the property. With an LLM, also compare the entities from the note text with a run on the backup: if names from the text went missing, the model is too small for the rewritten vault (see the trade-off above).
7. **Report** what changed, where the backup is, and how to undo it: `git checkout -- .` in git, or copy the backup back.
## Example
Before: `team: "Imaging"`, `systems: "Lab server AND GPU node"`, `item_id: "1.1.2"`. After:
```yaml
team: "[[Imaging]]"
systems:
- "[[Lab server]]"
- "[[GPU node]]"
aliases:
- "1.1.2"
```
vault2kg then builds `TEAM`, `SYSTEMS` edges for every note, and "see 1.1.2" in any note resolves to this one.
## Model
This step decides what the graph will contain, and it is run once. Use the strongest model available for it.
Pliki w paczce (4)
- SKILL.md 5 KB
- example-plan.json 1 KB
- scripts/apply_plan.py 11 KB
- scripts/profile_vault.py 5 KB
Wersje
W bazie nie ma innych skilli o nazwie vault-prep.
Migawka bazy z 13.11.2026 · wersja paczki fe3d69ec45c9 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100obsidian-vault
Opisuje konwencje vaulta Obsidiana: strukturę, nazewnictwo plików, frontmatter, tagi, zadania i linkowanie. - 90note
Porządkuje pliki w vaultcie Obsidian w ~/Notes: strukturę PARA, nazwy plików, frontmatter, szablony i tablice kanban. - 100obsidian
Obsługuje vault Obsidiana: czyta notatki po nazwie, przeszukuje zawartość, tworzy i dopisuje notatki, sprawdza backlinki i pracuje z notatkami dziennymi. - 100obsidian
Obsługuje funkcje specyficzne dla Obsidiana: wikilinki, backlinki, canvas, notatki dzienne i właściwości frontmatter, gdy użytkownik wprost wskazuje Obsidiana lub jego możliwości vaulta. - 100obsidian
Pomaga przeszukiwać, tworzyć, edytować i przenosić notatki w vaultcie Obsidiana, zarządzać frontmatter YAML, budować strukturę Zettelkasten i synchronizować vault przez Obsidian Headless. - 95obsidian-vault-builder
Porządkuje rozrzucone pliki markdown w vault Obsidian: dzieli zbyt długie pliki, dodaje frontmatter i naprawia linki.