Research Assistant

~2 min read

Research Assistant

You start a note on a topic with a few bullets. The agent surfaces related notes you already have, suggests links you didn’t realize were already there, drafts an outline, and writes a one-paragraph intro. The result lands as an [!ai-done] callout you can keep or delete.

Setup

Prompt

@claude This note is a research starter. Help me build it out:

  1. Use link_graph(path, depth: 2) to find related notes already in this vault.

  2. Use suggest_links to surface unlinked mentions of relevant ideas.

  3. Propose a Markdown outline (## headings only) covering the topic.

  4. Draft a one-paragraph intro.

Commit the outline + intro to the bottom of the note as an [!ai-done] callout.

Tools the agent will use

Tool Why
read_note(path) The current note’s seed content.
link_graph(path, depth) A real bidirectional neighborhood — outlinks AND backlinks.
suggest_links(path) Names of other notes mentioned in plain prose but not yet [[linked]].
search(query) Pull in notes outside the immediate graph that match key terms.
apply_edits(path, edits) Insert the outline + intro at a specific anchor in the note.
write_note(path, content, mode: "append") Or just append the callout.

Trust + safety

Example output

> [!ai-done] Research draft
> ## Outline
> ## What is MCP?
> ## Why pair it with a notes app?
> ## Open questions
>
> MCP gives agents typed tools against your `.md` files — see [[MCP]] for the protocol and
> [[Working with AI Agents]] for Lettuce's specific bindings.

Combined with the daily triage agent, you can have one agent collect ideas in Inbox/ and another one expand them into research notes.