/read

Setup

1. Read ${CLAUDE_PLUGIN_ROOT}/config.json → extract vault.path.
2. Read notes.yaml from the current working directory → extract folder. If missing, tell the user to run /notes:init first and stop.
3. Resolve full notes directory: {vault.path}/{folder}/

Load notes via subagent

Today's date is ${CURRENT_DATE} (YYYY-MM-DD). Derive the date string from that.

Launch a Task with subagent_type: Explore and description: "Read session notes". Pass it a prompt containing:

• The resolved notes directory path
• Today's date string
• The $ARGUMENTS hint (may be empty)

The subagent prompt should instruct it to:

1. If a hint is provided:

   • Glob {notesDir}/*{hint}*.md to match by topic/slug.
   • Also glob {notesDir}/{today}-*.md for today's files.
   • Combine results, deduplicate.

2. If no hint:

   • Glob {notesDir}/{today}-*.md for today's files.

3. If no files found in steps above:

   • Glob {notesDir}/*.md for all notes. Sort by filename descending (most recent first). Take the 3 most recent.

4. Read all matched files.

5. Return a structured summary with this format for each file:

   ### {filename}
   **Title:** ...
   **Summary:** (from the Summary section)
   **Key decisions:** ...
   **Open questions:** ...
   **Next steps:** ...
   **Context:** (from Context for future sessions)
   

   Omit any subsection that wasn't present in the note. Be concise but preserve specific details (file paths, function names, concrete decisions).

After subagent returns

Do NOT repeat the summary back to the user — it's already in your context and repeating it wastes tokens. Instead:

1. List which files were read (filenames only).
2. Say you're caught up on the context.

The user can then ask about specific details (open questions, decisions, next steps, etc.) and you answer from what's already loaded.

If no notes were found at all, tell the user and suggest running /notes:save to create one.