| name | init |
| description | Use to set up (or re-profile) a vault for the vault-mind skills, e.g. "set up vault-mind", "init my vault", "vault-mind isn't finding my notes", "re-detect my vault structure". Inspects the user's Markdown vault, infers its layout and frontmatter schema, confirms with the user, and writes a vault profile that every other vault-mind skill reads. Run this once before using quiz/feynman/ask/etc. Writes only the profile file, never touches note content. |
Init (profile the vault)
Every other vault-mind skill (quiz, feynman, discuss, socratic-philosophy, ask,
connect, gaps, lint, recommend, fetch) needs to know this user's vault shape:
where notes live, where readings live, what the frontmatter keys are called, what status
and rating mean here. Vaults differ. This skill detects that once and writes it to a
profile the others load, so the rest of the kit is portable instead of hardcoded.
What this writes: only <vault>/.vault-mind/profile.md (and creates .vault-mind/).
It never reads-then-rewrites the user's note content. It may read notes to infer schema.
When to run
- First use of vault-mind in a vault (no
.vault-mind/profile.md yet).
- The user reorganized their vault (renamed folders, changed frontmatter keys).
- A skill reports it can't find the profile or the layout looks wrong.
Procedure
1. Locate the vault root
The vault root is the directory the user runs Claude from, unless they say otherwise. If
there's an .obsidian/ directory, that marks an Obsidian vault root, use it. Confirm the
root with the user if ambiguous (e.g. they're in a parent dir containing several vaults).
2. Detect the layout (don't assume; look)
Inspect the actual tree, then infer. Useful probes (from the vault root):
fd -t d -d 2 .
fd -e md . -d 2 | head -50
fd -e md . -d 3 -x sh -c 'sed -n "/^---$/,/^---$/p" "$1" | head -20' _ {} \; 2>/dev/null | sort | uniq -c | sort -rn | head -40
Infer and record:
- notes_dir: where atomic/concept notes live (often
Notes/, Zettel/, Slipbox/, or the root).
- resources_dirs: where reading material lives, with the
class each maps to
(e.g. Resources/Books→Book, Resources/Papers→Paper, Sources/→generic). May be empty
if the user keeps no reading log, several skills degrade gracefully without it.
- assets_dir: where images / fetched sources go (for
fetch).
- templates_dir: if present.
3. Detect the frontmatter schema
From the sampled frontmatter, determine the actual key names and value vocabularies this
vault uses; do not impose vault-mind's defaults:
- The note keys: tag key (
tags?), links key (related?), source key (resources?),
date key (date/created?), and whether a reviewed/review-date key exists (quiz needs
one for spacing, if absent, propose adding reviewed).
- The reading keys: title, author, a status key + its value set (e.g.
read|reading|to be read|DNF, or done|todo, or Dataview #status/... tags), and a
rating key + its scale (0–10? 1–5 stars? none?).
- The wikilink style:
[[Title]] vs [[path/Title]] vs none (plain MD links).
- Tag style: hierarchical (
a/b) or flat.
When the vault is inconsistent or sparse, ask the user rather than guessing, a wrong
schema makes every downstream skill grade/scope incorrectly.
4. Confirm, then write the profile
Show the user the inferred profile and have them correct it. Then write
<vault>/.vault-mind/profile.md using assets/profile-template.md as the structure, filled
with real values. Also create .vault-mind/ if needed. If the vault uses git, mention they
can .gitignore .vault-mind/ (it's local config) or commit it (to share the profile) ,
their call.
If the chosen note schema lacks a review-date key and the user wants spaced quiz, add a
reviewed key to the profile's note schema and tell them quiz will stamp it on demand.
5. Hand off
Tell the user the profile is written and the kit is ready, list the available skills
(/vault-mind:quiz, :feynman, :discuss, :socratic-philosophy, :ask, :connect,
:gaps, :lint, :recommend, :fetch), and point at the natural loop:
recommend → (read) → discuss / socratic-philosophy → you write the note → feynman → quiz → connect.
The kit's core rule (state it to the user)
vault-mind never writes the user's note content: that's where the learning lives. Skills
read, quiz, surface, and at most edit metadata/links (with approval). The user does the
thinking and the writing. If they want an AI that drafts notes for them, this is the wrong kit.
