Connect Ideas

When to use

The user wants serendipitous structural insight — a pair of notes that share content but no wikilink edge. Useful during a slow-thinking session or as part of a weekly review.

Procedure

1. Call vault_info.
2. Call find_hubs(top_k=10) to identify central notes — these are the strongest candidates for a missed connection.
3. For each hub, call find_related(seed=hub_path, top_k=8). Filter the results: keep only notes that are NOT already wikilinked to the hub (use find_backlinks(hub_path) to determine this).
4. Score each candidate pair by find_related similarity × log(candidate's wikilink count). High-quality candidates are dense-similar AND not already isolated.
5. Pick the single highest-scoring pair. If no candidate clears a similarity threshold of 0.55, tell the user "I didn't find a strong enough pair this round" and stop.
6. Call read_note on both notes. Skim. Identify the specific overlap.
7. Surface to the user IN CHAT:

• Note A path + 1-line summary
• Note B path + 1-line summary
• Why they should connect (2–3 sentences, pointing at the actual textual overlap)
• The proposed wikilink insertion: which note gets the link, near which phrase, with what alias if any.

1. If the user confirms, call link_notes(from_path=..., to_path=..., context_phrase=..., confirm=true).

Output format

Found a pair:
  A: [[Note A]] — {one-line summary}
  B: [[Note B]] — {one-line summary}

Rationale: {2–3 sentences citing the overlap}

Proposed: Insert `[[Note B]]` into [[Note A]] near "{context phrase}".

Anti-patterns

• Don't propose more than one pair per invocation. Decision fatigue kills the workflow.
• Don't link two notes that are already 1-hop connected via a third note unless the connection is genuinely value-add. Surface the existing path instead.
• Don't propose a link that would make Note A's text grammatically nonsensical. Re-read the context phrase before suggesting.
• Don't insert the link without confirm=True from the user (unless --unattended).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: bettyguo
• Source: bettyguo/obsidian_mcp
• License: MIT

Install and usage instructions live in the source repository linked above.