init

Initialize a Trove

A Trove is a personal, file-based knowledge index that grows as you use Claude Code. It is just plain files on disk (Markdown for prose, JSON for structured data), so it is portable, greppable, diffable, and yours. This skill creates it.

Steps

1. Pick a scope (if the user did not say in the argument):

   • User scope ~/.claude/trove/ (default): one trove that follows you across every project.
   • Project scope ./.claude/trove/: a trove that lives in and travels with this repository (commit it to share with a team).

   If unclear, default to user scope and mention you did.

2. Scaffold the structure. Create the directory, an entries/ subfolder, and an INDEX.md:

   TROVE="$HOME/.claude/trove"          # or "$PWD/.claude/trove" for project scope
   mkdir -p "$TROVE/entries"
   

   Then write INDEX.md (only if it does not already exist) with this exact starter:

   # Trove Index
   
   > Your personal Claude Code trove. One line per entry, newest first.
   > Loaded into context at the start of every session by the trove plugin.
   > Capture new entries with /trove:remember and search with /trove:recall.
   
   <!-- entries below -->
   

3. Confirm to the user with the resolved path and the three commands they will use: /trove:remember to capture, /trove:recall to search, and that it auto-loads each session.

The entry format (so every skill agrees)

Each fact is its own file in entries/, in one of two formats. The remember skill chooses by the shape of the content: Markdown for prose, JSON for structured data.

Markdown entries/<slug>.md for prose: decisions, gotchas, conventions, narrative notes.

---
title: <Human readable title>
slug: <kebab-case-slug>
type: decision | gotcha | preference | reference | project | snippet
created: <YYYY-MM-DD>
tags: [tag1, tag2]
---

<One or two short paragraphs: the durable thing worth keeping.>

**Why it matters:** <when this applies / why future-you will care>
**Related:** [[other-slug]]

JSON entries/<slug>.json for structured context: mappings, lists of records, config, schemas.

{
  "title": "<Human readable title>",
  "slug": "<kebab-case-slug>",
  "type": "reference | snippet | project | decision | gotcha | preference",
  "created": "<YYYY-MM-DD>",
  "tags": ["tag1", "tag2"],
  "summary": "<one-line hook for the index>",
  "data": { "...": "the structured payload, any shape" }
}

Either way, add the matching one-line entry to INDEX.md. The file extension tells readers (and the session hook) which format it is:

- [<Title>](entries/<slug>.md) - <one-line hook>
- [<Title>](entries/<slug>.json) - <one-line hook>

Do not overbuild. Keep entries simple, atomic, and honest.