Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› jel-literature-synthesis
jel-literature-synthesis
GitHub用于系统性地搜集、阅读并综合大量经济学文献，以构建JEL综述的证据基础。通过种子集、引文雪球及JEL代码扫描确保覆盖全面，利用证据矩阵对比研究方法与发现，评估可信度并识别争议，避免引用遗漏。
Trigger Scenarios
需要系统性阅读领域文献时
担心遗漏重要贡献或阅读缺乏系统性时
拥有大量文献但无法比较或调和结果时
预判审稿人可能质疑文献遗漏时
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jel-literature-synthesis -g -y
SKILL.md
Frontmatter
{
"name": "jel-literature-synthesis",
"description": "Use when systematically gathering, reading, and synthesizing a large body of economic research for a Journal of Economic Literature (JEL) survey — coverage discipline and avoiding citation gaps. Builds the evidence corpus and synthesis notes; it does not impose the analytical spine (jel-organizing-framework) or write prose."
}
Literature Synthesis (jel-literature-synthesis)
When to trigger
- The proposal was encouraged and it is time to read the field thoroughly
- Reading feels unsystematic; you fear missing important contributions
- You have many papers but no way to weigh, compare, or reconcile their findings
- A referee at JEL is likely to ask "why did you omit X / the Y literature?"
Coverage discipline: read the field, not a convenience sample
A JEL survey's credibility rests on the reader's belief that you read everything that matters. Build coverage systematically rather than from memory:
- Seed set. Start from the proposal's key references and the field's canonical pieces.
- Forward + backward snowball. Backward: every paper's reference list. Forward: who cites the seeds (Google Scholar / Scopus citation counts). Iterate until new searches stop turning up unseen important work (saturation).
- Keyword + JEL-code sweep. Search EconLit / RePEc / NBER / SSRN by topic keywords and by the relevant JEL classification codes (see
jel-classification-system) to catch work indexed under adjacent labels. - Working papers + adjacent fields. A current survey must include recent working papers (the frontier) and the bordering literatures the field draws on, or referees will flag staleness and siloing.
- Saturation log. Record when each search stops yielding new must-cite work — this is your evidence of comprehensiveness for
jel-comprehensiveness-and-balance.
From reading to synthesis (not summary)
Summarizing is restating each paper; synthesizing is making the papers talk to each other. Maintain an evidence matrix as you read:
| Column | What to capture |
|---|---|
| Study | author–year, the result you will cite it for |
| Question/estimand | exactly what it measures (so non-comparable studies are not pooled) |
| Method/data | design and sample (so you can weight credibility) |
| Finding | direction + magnitude (for who-found-what tables) |
| Quality/credibility | identification strength, sample, robustness — your appraisal, since JEL does not run new estimates |
| Tension | which other studies it agrees/conflicts with, and why |
This matrix is the raw material for the organizing framework, the summary tables, and the even-handed treatment of controversies. Crucially, you appraise primary studies (you are the field's referee-of-record); you do not re-estimate them.
Checklist
- Seed set drawn from canonical pieces + proposal references
- Forward and backward snowball iterated to saturation
- EconLit / RePEc / NBER / SSRN swept by keyword and JEL code
- Recent working papers and adjacent literatures included (no staleness, no silo)
- Saturation log records where searches stopped yielding new must-cites
- Evidence matrix captures estimand + method + finding + your credibility appraisal + tensions
- Non-comparable studies are flagged as such (not pooled into a false consensus)
- No major author/school whose omission a referee could name
Anti-patterns
- Citing from memory or from a personal reading list (predictable coverage gaps)
- Summarizing paper-by-paper instead of synthesizing across papers
- Pooling studies that estimate different objects into one "the literature finds…" claim
- Ignoring working papers (looks stale) or ignoring adjacent fields (looks siloed)
- Re-running or "correcting" primary studies — a JEL survey appraises, it does not re-estimate
- Leaning on review databases without the JEL-code sweep that catches adjacently-indexed work
Output format
【Seed set】<canonical + proposal references>
【Snowball status】backward/forward iterated to saturation? Y/N
【Databases swept】EconLit / RePEc / NBER / SSRN — by keyword + JEL code? Y/N
【Frontier + adjacency】recent WPs and bordering literatures included? Y/N
【Saturation evidence】<where searches stopped yielding new must-cites>
【Evidence matrix】rows ready with estimand / method / finding / appraisal / tension? Y/N
【Coverage risks】<any author/school an omission referee could name>
【Next step】→ jel-organizing-framework (impose the analytical spine on the matrix)
Version History
- 1839142 Current 2026-07-05 13:34
