| name | dr-cook:literature-reviewer |
| description | Synthesize and review academic literature. Use when summarizing multiple papers, extracting themes, creating literature review tables, or preparing systematic/narrative reviews. Triggers on: review literature, summarize papers, literature synthesis, systematic review, narrative review, 文献综述, 文献整理, 综述, 系统综述, 整理文献, 文献分析, 文献总结.
|
literature-reviewer
1. Overview
literature-reviewer synthesizes academic literature into structured reviews. It accepts uploaded papers, pasted abstracts, or PMID/DOI lists as input, and supports three review types: systematic (PRISMA-compliant), narrative (thematic), and scoping (evidence mapping). Output consists of a thematic synthesis, a key findings list, and a structured review table in a format matched to the review type and research domain. This module is the designated entry point of the Grant pipeline — it produces no upstream dependencies and writes a context_output object consumed directly by research-gap-finder and, subsequently, grant-writer. All downstream modules in the pipeline inherit domain, language, key_findings, and references from this module's output without re-asking the user.
2. Parameters
Required
| Parameter | Values | Description |
|---|
domain | tcm | bioinformatics | clinical | pharmacology | Research domain. Used to select domain-specific table columns and terminology. |
review_type | systematic | narrative | scoping | Type of review to conduct. Determines methodology, synthesis structure, and table template. |
Optional
| Parameter | Values | Description |
|---|
language | en | zh (default: en) | Output language for all generated text. |
topic | string | Explicit topic statement. Provide if the topic is not clearly derivable from the papers themselves. |
output_format | markdown | table_only | full_report (default: full_report) | Controls scope of generated output. table_only produces only the review table; full_report produces all output sections. |
Parameter collection rule
Ask for review_type first (one question, wait for the answer). Then ask for domain only if it cannot be inferred from the papers or abstracts already provided. Do not ask for optional parameters unless the user has explicitly mentioned them.
3. Workflow
Step 1 — Accept input.
Accept one or more of the following:
- User-uploaded PDF or text files (extract text from content provided).
- Pasted abstracts — one or more, each treated as a separate source.
- A list of PMIDs or DOIs. Note: PMID/DOI lists do not contain full text. If provided, inform the user that only metadata and abstracts may be available, and suggest running literature-downloader first to retrieve full texts before synthesis.
- A combination of the above types.
If no literature of any kind is provided, ask the user to supply papers, abstracts, or PMIDs/DOIs before proceeding. Do not generate placeholder content.
Step 2 — Collect missing parameters.
Apply the collection rule: ask for review_type first. If domain cannot be inferred from the provided content, ask for it next. Wait for each answer before continuing.
Step 3 — Load synthesis methodology.
Read references/synthesis-methods.md. Apply the section corresponding to the collected review_type (systematic, narrative, or scoping) to guide extraction logic, quality assessment, and synthesis structure throughout subsequent steps.
Step 4 — Extract from each source.
For each paper or abstract, extract the following fields (record as N/A if not reported or not available from abstracts only):
- Study design and sample size
- Population, organism, or cell type
- Intervention or exposure (if applicable)
- Key findings and primary outcome measures
- Methodology highlights (tools, assays, datasets, statistical methods)
- Quality or limitation notes
Step 5 — Synthesize across sources.
Group extracted information into themes. Across all sources:
- Identify recurring findings and consensus positions.
- Note explicit contradictions and unresolved debates between studies.
- Map coverage: what populations, interventions, methods, or combinations have been studied, and which are absent from the literature.
Apply synthesis guidance from references/synthesis-methods.md for the selected review_type (e.g., PICO framing for systematic; thematic grouping for narrative; evidence mapping for scoping).
Step 6 — Generate review table.
Read references/review-table-formats.md. Select the template matching the review_type, and apply domain-specific column customizations for the collected domain. For small sets (≤5 papers), use Template 1 only. For 6–20 papers with extractable outcome data under a systematic review, use Templates 1 and 2 together. For scoping reviews, use Template 3.
Step 7 — Compose output.
Produce all output sections as defined in Section 4:
- Thematic narrative synthesis: 3–6 paragraphs covering recurring themes, contradictions, and coverage gaps.
- Key findings list: 5–10 bullet points drawn directly from the synthesis. These will populate
key_findings in context_output for downstream modules.
- Review table: structured per the template selected in Step 6.
- Reference list: format all sources in Vancouver style using information extracted from the provided papers or supplied by the user. Do not fabricate bibliographic details.
Step 8 — Pipeline handoff.
After presenting all output, add this note: "context_output is ready for research-gap-finder." If this session was initiated via a pipeline command (/cook grant or /cook pipeline:), offer to automatically proceed to the next step. If this was a standalone invocation, offer: "Would you like to continue to research-gap-finder to identify research gaps in this literature?" — but wait for explicit user confirmation before proceeding.
4. Output Format
## Literature Synthesis: [Topic]
### Thematic Summary
[3–6 paragraphs covering recurring themes, contradictions, and evidence gaps]
### Key Findings
1. [Finding — specific, with gene names, percentages, or effect sizes where available]
2. [Finding]
...
### Review Table
[Structured table per review-table-formats.md — template and columns matched to review_type and domain]
### References
[Vancouver-formatted reference list — numbered sequentially]
For output_format: table_only, produce only the Review Table section. For output_format: markdown, produce all sections in the format shown above (identical to full_report in markdown rendering).
5. context_output
Reads from upstream
None. This module is the entry point of the Grant pipeline. It reads domain and language from user-provided parameters only — not from any upstream context_output.
Writes to output
{
"module": "literature-reviewer",
"domain": "<collected>",
"language": "<collected>",
"summary": "<thematic synthesis narrative, 200–300 words>",
"key_findings": ["<finding 1>", "<finding 2>", "...up to 10"],
"references": [
{ "pmid": "<if known>", "doi": "<if known>", "title": "...", "authors": "...", "year": 0 }
],
"parameters": {
"review_type":
Status values:
"success": Full text was available for all or most sources; synthesis is based on complete content."partial": One or more sources were abstracts only (from PMIDs without full-text retrieval, or pasted abstracts). Downstream modules (research-gap-finder, grant-writer) must treat key_findings and summary as potentially incomplete and should note this limitation in their own outputs.
error_message is null on success or partial; populated with a brief description if the module cannot produce any usable output (e.g., all PMIDs failed to resolve and no abstracts were provided).
Downstream readers
summary and key_findings: research-gap-finder, grant-writer, paper-writer.
references: citation-checker, paper-writer.
domain and language: all downstream modules in the pipeline.
6. References
See references/ for:
synthesis-methods.md — Review methodology by type (systematic, narrative, scoping); quality assessment tools; PRISMA flow guidancereview-table-formats.md — Structured table templates by review type and domain; column customization rules; template selection guidance
