| name | literature-review |
| description | Use when the user wants a synthesis ACROSS several papers — "write a literature review", "文献综述", "compare these papers", "synthesize what I've selected", "对比这几篇", "summarize this collection". Reads each item's text via zot and produces a themed, cross-paper review (not paper-by-paper summaries), then saves it as a standalone note. Grounded in the papers, with per-source attribution. |
| license | MIT |
| metadata | {"category":"zotero-research","primary-interface":"zot full-text + zotero_create_standalone_note"} |
| triggers | ["literature review","review","synthesize","compare papers","综述","文献综述","对比","多篇","汇总"] |
Literature review across papers
The user wants a synthesis across multiple papers — themes, agreements,
disagreements, methods, and gaps — not a stack of independent per-paper summaries. The
output is a standalone note (it spans many items, so it shouldn't hang off any single
paper).
Step 1 — establish the set
Get the papers to review:
- Selected items →
zotero_get_selected_items (the usual case).
- A collection →
zotero_get_current_collection, then enumerate its items (via
zot if needed).
If the set is empty or ambiguous, ask the user which papers/collection. Don't review a
guess.
Step 2 — read each paper (via zot, not memory)
Read each item's content through zot (full-text / PDF extraction) using its key.
The review MUST be built from the papers' actual text. If full text isn't available
for some items, work from their abstracts/notes and say which were abstract-only —
never fill gaps from memory.
If the set is large, say how many you covered and which (if any) you skipped for
length/time. Never imply full coverage you didn't do.
Step 3 — synthesize (the actual skill)
Organize by theme / question / method — not by paper. A good review:
- Groups papers by what they have in common, and contrasts where they differ.
- Names agreements, contradictions, and trends across the set.
- Compares methods/data and what each contributes.
- Ends with gaps / open questions the set leaves unanswered.
- Attributes every claim to its source(s) —
(Author, year) or the item title — so
the user can trace each point back. This doubles as a hallucination check: if you
can't attribute it, don't assert it.
A bulleted "Paper 1: … Paper 2: …" list is a FAILURE of this skill. Synthesize.
Step 4 — save it
Write the review as a standalone note:
- Preferred:
zotero_create_standalone_note (lands it in the current collection; it's
a cross-paper artifact, not a child of one item).
- If that tool isn't available, fall back to
zotero_add_note_to_current_item on the
focused item and say it was attached there instead.
Confirm in ~1 line after writing; don't re-paste the whole review into chat.
Anti-patterns (do NOT do these)
- Per-paper summaries instead of synthesis. The value is cross-paper structure.
If your output reads as N separate summaries, restructure by theme.
- Unattributed claims. Every synthesized point traces to source(s). No
attribution → don't make the claim.
- Filling gaps from memory. Only papers you read via zot this turn. Abstract-only
items must be labeled as such.
- Silent truncation. State coverage when the set is large; never imply you read
all of it if you didn't.
- Reimplementing zot. Reading/extraction is zot's job; this skill orchestrates it
and produces the synthesis.
