Subagent-Driven Literature Review
Use parallel subagents for large-scale paper screening and deep dive analysis
Profile →npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill subagent-driven-review --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
Subagent-Driven Literature Review instructs a main agent to split a large list of papers into batches, dispatch multiple subagents in parallel to screen or deeply analyze each batch, collect and consolidate the results, and generate a retrieval and review output (e.g., papers-reviewed.json and SUMMARY.md) with quality checks and batch tracking.
How it works
- Load a list of papers and decide batch size (typically 15-25 papers per subagent).
- Create TodoWrite entries for each batch (e.g., Batch 1: PMIDs 1-20, Batch 2: PMIDs 21-40, etc.).
- Dispatch subagents in PARALLEL within a single message using multiple Task calls to screen papers or perform deep dives.
- Each subagent fetches abstracts or full texts as specified, applies a scoring rubric (0-10) based on domain-relevant criteria (keywords, data types, methods), and returns structured JSON with per-paper scores and statuses.
- Main agent collects results, validates format, consolidates into a unified list, and merges into papers-reviewed.json with batch identifiers.
- Generate a SUMMARY.md from consolidated data and perform a quality review to ensure consistent scoring and coverage.
- Optionally perform deep dives on highly relevant papers in batches or sequentially, returning structured findings per paper.
When to use it
Use for large literature searches (50+ papers), parallelizable screening where papers are independent, deep-dive extractions on multiple papers, and citation-network exploration when the main context is getting full or time-pressured.
What it can touch
- Prompts and prompts templates for subagents
- Task tool calls to dispatch subagents in parallel
- JSON-formatted results from subagents (screened_papers, summary data)
- Files to update: papers-reviewed.json and SUMMARY.md (as consolidated outputs)
Caveats
- Do not update papers-reviewed.json from subagents directly; consolidation is required by the main agent.
- Ensure parallel dispatch uses a single message with multiple Task calls.
- Maintain consistent rubrics across subagents to avoid inconsistent scoring.
- Monitor for rate-limits when using external data sources and incorporate appropriate delays.
<!-- ╔══════════════════════════════════════════════════════════════╗ ║ 本文件为开源 Skill 原始文档，收录仅供学习与研究参考 ║ ║ CoPaper.AI 收集整理 | https://copaper.ai ║ ╚══════════════════════════════════════════════════════════════╝ 来源仓库: https://github.com/kthorn/research-superpower 项目名称: research-superpower 开源协议: MIT License 收录日期: 2026-04-02 声明: 本文件版权归原作者所有。此处收录旨在为社会科学实证研究者 提供 AI Agent Skills 的集中参考。如有侵权，请联系删除。 --> # S
What does the Subagent-Driven Literature Review skill do?
Use parallel subagents for large-scale paper screening and deep dive analysis
How do I install it?
Run `npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill subagent-driven-review --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From brycewang-stanford/Auto-Empirical-Research-Skills, a repository with 3,244 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.