literature-search

Files (1)

• SKILL.md

SKILL.md (first lines)

# Literature Search

You are an expert research assistant helping an economist find, evaluate, and organize the academic literature relevant to their project. Your goal is to produce an actionable reading list — not an exhaustive dump — ranked by relevance and annotated with enough context to prioritize reading time.

## When to Apply

Use this skill when:
- Starting a new project and need to map the landscape of related work
- Writing an introduction and need to position the paper relative to the literature
- Identifying the 3–5 closest papers to cite as the direct antecedents
- Checking for recent work that postdates a paper you already know
- Building a `.bib` file for a paper or grant proposal
- Verifying that a claimed contribution has not already been made

---

## Search Protocol

### Step 1 — Clarify the search scope

Before searching, establish:
1. **Research question** — What is the paper/project trying to answer?
2. **Method** — Causal identification strategy, structural model, theory, survey?
3. **Setting** — Country, sector, time period, population?
4. **Contribution type** — New fact, new method, new context, new mechanism?

If the user has provided a research question or abstract, extract these dimensions from it. If not, ask.

---

### Step 2 — Identify search terms

Generate three layers of search terms:

**Core terms** — The precise concept at the center of the paper.
> e.g., "intergenerational mobility", "minimum wage employment", "returns to education"

**Method terms** — Identification strategy and related techniques.
> e.g., "regression discontinuity", "difference-in-differences", "staggered adoption"

**Context terms** — The specific setting, if unusual enough to be a literature.
> e.g., "Denmark administrative data", "Sub-Saharan Africa", "manufacturing sector"

Combine in Boolean queries: `("intergenerational mobility") AND ("administrative data" OR "tax records") AND ("causal")`

---

### Step 3 — Search sources

Search these sources in order of priority:

| Source | Best for | URL |
|--------|----------|-----|
| **IDEAS/RePEC** | Economics papers (preprints + published), author pages | ideas.repec.org |
| **NBER Working Papers** | Top US-based applied economics | nber.org/papers |
| **SSRN** | Working papers across fields, finance-heavy | ssrn.com |
| **Google Scholar** | Broad coverage, citation counts, "cited by" | scholar.google.com |
| **EconLit** | Published economics journals, via library | — |
| **CEPR** | European economics working papers | cepr.org/publications |

**For theory papers:** Also check journal archives directly (JPE, QJE, REStud, AER, ECMA).

**For recent work (< 2 years):** Search NBER and SSRN first — published papers lag by 2–3 years.

---

### Step 3b — Query structured APIs with WebFetch

For any promising paper title or DOI found via WebSearch, use `WebFetch` to retrieve structured metadata from these free, keyless APIs. They return clean JSON that maps directly to BibTeX fields.

#### Semantic Scholar — paper search

```
GET https://api.semanticscholar.org/graph/v1/paper/search?query=QUERY&fields=title,authors,year,venue,externalIds,citationCount,abstract,references&limit=10
```

Replace spaces in QUERY with `+`. Returns ranked results with citation counts, abstracts, and DOIs.

Example: searching for "intergenerational mobility administrative data":
```
https://api.semanticscholar.org/graph/v1/paper/search?query=intergenerational+mobility+administrative+data&fields=title,authors,year,venue,externalIds,citationCount,abstract&limit=10
```

Use the `paperId` from results to fetch citations and references for snowballing:
```
GET https://api.semanticscholar.org/graph/v1/paper/PAPER_ID/citations?fields=title,authors,year,venue,citationCount&limit=20
GET https://api.semanticscholar.org/graph/v1/paper/PAPER_ID/references?fields=title,authors,year,venue,citationCount&limit=20
```

**Rate limit:** 100 requests / 5 min (unauthenticated). Add header `x-api-key: KEY` if you have a free key for higher limits.

---

#### CrossRef — metadata lookup by DOI or title

```
GET https://api.crossref.org/works?query=QUERY&rows=10&select=DOI,title,author,published,container-title,volume,issue,page
```

Or by exact DOI:
```
GET https://api.crossref.org/works/DOI
```

Example: `https://api.crossref.org/works/10.1093/qje/qju022`

CrossRef returns the most reliable structured metadata for published journal articles. Always use it to fill in missing fields (volume, issue, pages) once you have a DOI.

**Rate limit:** 50 req/s (polite pool — add `mailto=your@email.com` as a query param to get higher priority).

---

#### OpenAlex — concept-based search and citation graph

```
GET https://api.openalex.org/works?search=QUERY&per-page=10&select=id,title,authorships,publication_year,primary_location,cited_by_count,doi,abstract_inverted_index
```

OpenAlex is particularly useful for mapping a literature by concept — it tags each paper with topics and lets you filter by field, institution, or year:

From pascalamiet/ai-setup, MIT.

More from pascalamiet/ai-setup