Search Second Brain

Search Second Brain

Local-first retrieval over the user's Obsidian vault at ~/Documentos/second-brain using Ollama (qwen2.5:7b). No data leaves the machine.

When to use

Invoke this skill whenever the user wants to read / search / query / recall anything from their second brain or notes.

How to search

Run the search script with the user's question as the argument:

python3 ~/.claude/skills/search-second-brain/search.py "<the user's question>"

The script will:

1. Ensure ollama serve is running (starts it detached if needed).
2. Verify the model is present.
3. Retrieve the most relevant notes (keyword scoring over the vault).
4. Ask qwen2.5:7b to answer using only those notes, with sources.

Relay the answer to the user, and mention which notes were retrieved.

Node CRUD

Each Claude turn already auto-creates a node via the Stop hook (see nodes/ and nodes/_log.md in the vault). For manual operations:

SB=~/.claude/skills/search-second-brain/crud_node.py
python3 $SB create --title "Title" --body "Text" --tags a,b --links Other-Note
python3 $SB read   --name <node-stem-or-substring>
python3 $SB update --name <node> --append "more text"
python3 $SB delete --name <node>
python3 $SB list

Graph-first

This vault is a knowledge graph. Keep it connected:

• Auto-captured nodes are linked into the graph automatically (Stop hook links each node to its top relevant notes + the previous node).

• When you create or curate notes, add a ## Links section with [[wikilinks]].

• Repair orphans anytime:

  python3 ~/.claude/skills/search-second-brain/link_orphans.py        # link only orphans
  python3 ~/.claude/skills/search-second-brain/link_orphans.py --all  # re-link everything
  

Configuration (env overrides)

• SECOND_BRAIN_VAULT — vault path (default ~/Documentos/second-brain)
• SECOND_BRAIN_MODEL — Ollama model (default qwen2.5:7b)
• OLLAMA_HOST — Ollama API (default http://localhost:11434)