lamm-mit Literature Survey Workflow Summary
This document describes research-collect, a CLI skill for conducting systematic literature surveys. Here are the key components:
Purpose
Automate paper discovery, filtering, and organization to support downstream research skills.
Five-Phase Workflow
Phase 1 (Prep): Generate 4-8 search terms and establish directory structure.
Phase 2 (Search Loop): For each term:
- Query arXiv with
arxiv_search - Score results (1-5 scale); retain ≥4
- Download relevant papers
- Store metadata in
papers/_meta/{id}.json
Phase 3 (Code References):
- Identify top 5 papers (score ≥4)
- Search GitHub using paper titles + keywords
- Clone 3-5 repositories to
repos/ - Document selections in
prepare_res.md
Phase 4 (Organization): Cluster papers into 3-6 research directions; organize into papers/{direction}/ folders based on metadata analysis.
Phase 5 (Reporting): Generate survey/report.md with summary, directions, top papers, and recommended reading order.
Key Principles
- Incremental processing prevents context bloat
- Metadata-driven classification using JSON files
- File structure directly mirrors categorization
Available Tools
arxiv_search: Query papersarxiv_download: Retrieve documentsgithub_search: Find reference repositories