lit-review

lit-review

Multi-paper synthesis that matches the pattern in LLM Safety Literature Review (Applied NLP).md.

Invariant: always ask, never default

Any parameter below the user did not specify MUST be asked before doing work. Bundle related questions. Never silently choose a default.

Required parameters — ASK if missing

1. Paper set source — Zotero collection name, Zotero tag, or an explicit list of identifiers.
2. Focus question / framing — the question the review answers. "Broad survey" is acceptable only if the user explicitly picks it.
3. Note filename — in 3-Notes/.
4. topic/ tag — ASK which topic (vault requires type/leaf + topic/X + source/article).

To confirm — ASK, do not assume

• Clustering strategy — by method / by problem framing / by year / by venue. Show candidate clusters first and ASK.
• Comparison table — include one? If yes, ASK which dimensions (method, dataset, results, assumptions, ...).
• Depth per paper — one-line / one-paragraph / full mini-section. ASK.
• Gap analysis section — include? ASK.
• Vault cross-reference — wiki-link to existing vault notes on methods and concepts? Load /vaultdb and confirm.

Workflow

1. Confirm all parameters above.
2. Gather papers. Via Zotero MCP (mcp__zotero__zotero_get_collection_items, zotero_search_by_tag) or the explicit list. For each: title, authors, year, venue, abstract, citation count when available, annotations if present.
3. Propose clusters based on the set — show them to the user and let them adjust before you commit.
4. Draft per confirmed cluster. Each section opens with a framing sentence, then prose synthesis. Paper count in the heading (e.g., ### 1. Guardrail Frameworks (17 papers)).
5. Draft synthesis paragraph. Cross-cluster convergences, tensions, open problems.
6. Write to vault via /vaultdb or /obsidian-cli. Location: 3-Notes/<confirmed filename>.md.

Output structure

---
aliases:
tags:
  - type/leaf
  - topic/<confirmed>
  - source/article
related-to:
---

# <Title>

<Opening paragraph: what the review covers, why, the framing question.>

---

## Problem Framing

<Why is this review needed? What question motivates it?>

---

## Scope

<Inclusion criteria, time window, paper count, exclusions.>

---

### 1. <Cluster name> (<N> papers)

<Prose synthesis. Named papers with (First Author et al., Year, Venue) inline.>

### 2. <Cluster name> (<N> papers)

...

---

## Comparison

<If confirmed: comparison table with agreed dimensions.>

---

## Synthesis

<Cross-cluster narrative: convergences, tensions, trends.>

---

## Gaps and Open Problems

<If confirmed: what the literature has not addressed.>

---

## Recommended Further Reading

<Candidate next papers surfaced during synthesis.>

Style (match the vault)

• Prose paragraphs, not bullet dumps.
• Inline citation style: (Author et al., Year, Venue).
• Wiki-link liberally to existing concept notes (check via /vaultdb semantic search).
• Paper counts belong in cluster headings, not in bullets.

Guardrails

• Never fabricate citation counts, venues, or paper claims. Missing fields stay missing.
• Never overwrite an existing review note without explicit confirmation.

$ARGUMENTS