semantic-scholar-lookup

---

name: semantic-scholar-lookup description: Use when user needs to find papers from the open web (not local Zotero library) to support a writing argument, fill a citation gap, or discover literature in a field. Triggers include: "find papers on X", "I need references for this claim", "what's been published about Y", Zotero returns empty results, or topic is outside local library. allowed-tools: [Bash] compatibility: Semantic Scholar API key optional (recommended for stable throughput)

Semantic Scholar Lookup

Search the open Semantic Scholar corpus and return citation-ready evidence packs for scientific writing. Use when Zotero is unavailable or insufficient.

When to Use

• Finding papers on a topic not in the local Zotero library
• Supporting a specific argument or claim with external citations
• Discovering recent literature in a field
• Zotero semantic search returns empty or irrelevant results
• User explicitly asks to search Semantic Scholar or the web for papers

Do NOT use when the user wants citations from their own curated library — use zotero-research-lookup instead.

API Key Setup

The script works anonymously but an API key gives stable 1 req/s throughput. Before running, check if the user has configured their API key:

1. Look for a .env file in the project directory or parent directories
2. Check for SEMANTIC_SCHOLAR_API_KEY=<key> in the .env file
3. If not found, inform the user they can:
   • Create a .env file with SEMANTIC_SCHOLAR_API_KEY=your-api-key-here
   • Or set the environment variable: export SEMANTIC_SCHOLAR_API_KEY=your-api-key-here
   • Get a free API key from: https://www.semanticscholar.org/product/api#api-key-form

The script automatically detects the .env file and warns if the key is missing (anonymous mode proceeds with rate limits).

Deterministic Workflow

Step 1: Translate argument/topic to search query

If input is a natural-language argument or claim, extract 3–5 focused English keywords before searching. See Argument-to-Query section below.

Step 2: Search papers

# Use python3 (macOS/Linux default)
python3 scripts/semantic_scholar_lookup.py "<query>" --year-from <YYYY>

# Or use uv to run in isolated environment
uv run python scripts/semantic_scholar_lookup.py "<query>" --year-from <YYYY>

Default returns top 12 papers ranked by relevance + recency + citation impact + venue quality.

Step 3: Expand with recommendations (optional)

Omit --no-recommendations (default ON) to let the script fetch related papers from top seed results. Disable for speed:

python scripts/semantic_scholar_lookup.py "<query>" --no-recommendations --limit 10

Step 4: Return evidence pack

Report results using the Output Format below. Always include traceability status per citation.

Argument-to-Query Translation

When the user provides a claim or section topic rather than keywords, decompose it before searching:

Input type	Example	Query to use
Broad claim	"gut microbiome affects mental health"	gut microbiome depression mechanism
Section topic	"research background on AlGaN UV detectors"	AlGaN UV photodetector performance
Hypothesis	"sleep deprivation causes insulin resistance"	sleep deprivation insulin resistance causal
Comparison argument	"mRNA vaccines vs traditional for cancer"	mRNA cancer vaccine clinical trial efficacy

Rules:

• Use 3–5 English keywords (Semantic Scholar indexes English abstracts globally)
• Include mechanism/outcome words when the claim is causal
• Add year range when recency matters: --year-from 2020
• Run 2–3 query variants if first result set is weak

Output Format

## Query
<query used>

## Top Evidence
### 1) <title>
- paperId: <id>
- authors: <names>
- year: <year>
- venue: <venue>
- citationCount: <n>
- traceable: true/false
- claim: <tldr or abstract excerpt>
- url: <semanticscholar.org/paper/...>

## Synthesis
<1–3 sentences connecting top papers to the original argument>

## Limitations
- <e.g., N papers lack abstracts>
- <e.g., N papers lack DOI>

## Rate Limit Status
- total_requests: <n>
- retry_429_count: <n>

Guardrails

• All API calls throttled at 0.8 req/s via built-in GlobalRateLimiter
• 429 → respect Retry-After; fallback exponential backoff 2s→4s→8s
• Prefer /paper/batch over per-paper detail calls
• Exclude non-traceable papers by default (no paperId, DOI, or URL)

Script Reference

# Basic search with year range (use python3 on macOS/Linux)
python3 scripts/semantic_scholar_lookup.py "QUERY" --year-from 2019 --year-to 2026

# JSON output to file
python3 scripts/semantic_scholar_lookup.py "QUERY" --json -o evidence.json

# Fast mode, no recommendation expansion
python3 scripts/semantic_scholar_lookup.py "QUERY" --no-recommendations --limit 10

# Using uv (recommended for isolated environment)
uv run python scripts/semantic_scholar_lookup.py "QUERY" --year-from 2020

Run python scripts/semantic_scholar_lookup.py --help for all flags.

Failure Handling

• Search returns empty: rephrase query with broader or different keywords, try 1–2 variants
• All results non-traceable: add --non-strict-traceability and flag in output
• 429 exhausted after retries: report rate limit hit, suggest retry in 60s
• No API key set: script proceeds anonymously and warns on stderr; direct user to API Key Setup section above

References

• API endpoint details: references/api_usage.md