idea-deep-dive

idea-deep-dive

Literature exploration for fuzzy research questions.

Invariant: always ask, never default

Any parameter below the user did not specify MUST be asked before doing work. Bundle related questions.

Required parameters — ASK if missing

1. Research question or hypothesis — a clear statement. If the user's framing is vague, ASK for a sharper version before searching.
2. Sources to search — Zotero library, arxiv, Semantic Scholar, general web? Multi-select.
3. Time window — all time, last N years, or explicit range.
4. Cap per source — max results (e.g., 10 / 20 / 50).

To confirm — ASK, do not assume

• Devil's advocate mode — actively hunt for contradicting evidence, or supportive-only, or balanced?
• Include user's vault — scan 3-Notes/ for existing notes touching the question? If yes, load /vaultdb.
• Include non-paper sources — blog posts, tech reports?
• Save as a brainstorm note — if yes, ASK the filename AND the folder (1-TempNotes/ for scratch, 3-Notes/ for permanent). Never default the folder.
• Follow-up hand-offs — for top 1-3 promising seeds, run /citation-trace or /paper-triage?

Workflow

1. Confirm every parameter.
2. Sharpen the question — show the user your interpretation in 1-2 sentences and ASK for confirmation before searching.
3. Search each confirmed source up to the confirmed cap.
   • Zotero: mcp__zotero__zotero_semantic_search, zotero_advanced_search.
   • arxiv: search via WebFetch on http://export.arxiv.org/api/query?search_query=....
   • Semantic Scholar: https://api.semanticscholar.org/graph/v1/paper/search?query=...&fields=title,authors,year,venue,abstract,citationCount.
4. Cluster findings into: Supporting / Contradicting / Adjacent methods / Tangential but interesting.
5. Note gaps — angles no paper seems to cover.
6. Report (and optionally save if confirmed).

Output format

## Question (as confirmed)
<Sharpened statement>

## Sources searched
<list, with confirmed caps>

## Supporting (<N>)
- <Title> — <Authors>, <Year>, <Venue>. <One-line relevance.>
...

## Contradicting (<N>)
- <Title> — ...

## Adjacent methods (<N>)
- <Title> — ...

## Tangential but interesting (<N>)
- <Title> — ...

## Gaps
<Angles or claims no paper seems to cover — written as prose, not bullets.>

## Recommended next moves
- `/paper-triage` on: <top 2-3>
- `/citation-trace` forward on: <top 1>
- `/lit-review` if the Supporting cluster crosses <threshold>

Guardrails

• Sharpen vague questions by asking, not by guessing. A fuzzy query yields a fuzzy result.
• Never fabricate titles, authors, or venues. An empty source → an empty cluster. Report honestly.
• Do not write to the vault without explicit confirmation and a confirmed folder.

$ARGUMENTS