| name | research-lr-ra |
| description | Literature review research assistant workflow for Codex. Use when the user asks Codex to do LR, literature review, literature mapping, paper discovery, research gap finding, Zotero-backed reading, Google Scholar/gs skill searching, Chrome MCP browsing, user-mediated Zotero intake, OneFind local-library synthesis, or to turn a topic, research question, course assignment, advisor suggestion, or rough idea into a structured review plan and reading map. |
Research LR RA
Core Principle
Connect the whole literature-review chain: search, save, read, organize, and synthesize. Do not treat search results as the final answer. Aim to turn a rough topic into a managed local literature base plus a research map the user can act on.
Workflow
-
Clarify the topic lightly.
- Accept rough inputs such as an advisor's phrase, assignment prompt, early research interest, or formal research question.
- Identify likely disciplines, keywords, synonyms, core constructs, methods, populations, and time range.
- Avoid over-asking. If the user has not specified scope, choose a reasonable initial scope and state it.
-
Search the literature.
- Prefer the user's available scholar-search tools, especially gs skill with Chrome MCP for Google Scholar.
- Search for seminal papers, recent papers, review papers, high-citation papers, methods papers, and papers frequently co-cited with core works.
- Iterate keywords across theory terms, empirical setting terms, method terms, and adjacent literatures.
- Track search terms and why each query was used.
-
Prepare promising papers for saving.
- Use Zotero MCP or the local API only for read-only lookup, export, and verification. Prepare DOI/BibTeX/RIS or an import queue and ask the user to perform Zotero imports or Desktop writes.
- Preserve metadata: title, authors, year, venue, DOI/URL, abstract, tags, and notes when available.
- Prefer a durable pending-import record before deep synthesis so the user does not lose useful findings.
-
Read and synthesize the local library.
- Use OneFind or equivalent local-library search once papers are in Zotero.
- Treat Zotero as the user's research knowledge base, not just citation storage.
- Compare papers by question, theory/mechanism, data, method, identification, findings, limitations, and relation to the user's topic.
-
Produce actionable LR outputs.
- Literature map: 3-6 main streams with concise descriptions.
- Priority reading list: which papers to read first and why.
- Classification table: seminal/recent/review/theory/method/empirical/mechanism papers.
- Research gap memo: crowded areas, open spaces, plausible contribution angles.
- Next-search plan: missing keywords, venues, authors, datasets, or adjacent fields to search next.
Quality Checks
- Distinguish what was found in sources from inference.
- Do not overstate novelty from a shallow search.
- Prefer papers with stable bibliographic metadata over vague web snippets.
- Flag access gaps, missing PDFs, uncertain metadata, and likely duplicate records.
- Do not launch, control, or write through Zotero Desktop; keep every Zotero mutation user-mediated.
- When tools are unavailable, still run the same conceptual workflow manually and clearly say which tool-backed steps could not be executed.
Output Shape
For exploratory LR, default to:
Topic scope
Search strategy
Core papers
Literature streams
Priority reading order
Potential gaps
Next actions
For a more mature project, default to:
Research question
Position in literature
Mechanisms and methods
What is already crowded
What remains open
Candidate contribution
Reading and evidence plan