Pharmaclaw Literature Agent

Search and retrieve biomedical and scientific literature from PubMed and Semantic Scholar with structured summaries, citation tracking, and open-access PDFs.

Security Scan

The skill's code, instructions, and resource requests are consistent with a literature-search agent that queries public APIs (PubMed, Semantic Scholar, bioRxiv, ClinicalTrials); it does not request secrets or install arbitrary code from unknown hosts.

Purpose & Capability

Name/description (PubMed, Semantic Scholar, bioRxiv, ClinicalTrials) match the code: all network calls target public literature APIs (eutils.ncbi.nlm.nih.gov, api.semanticscholar.org, api.biorxiv.org, clinicaltrials.gov). No unrelated credentials, binaries, or config paths are requested.

Instruction Scope

SKILL.md and the scripts limit actions to constructing queries and calling public APIs, parsing returned JSON/XML, deduplicating/enriching results, and producing summaries/recommendations. Scripts do not read arbitrary local files or environment variables beyond normal module imports and do not post data to unexpected external endpoints.

Install Mechanism

This is instruction-plus-scripts only; there is no install spec that downloads or executes code from an untrusted URL. The only runtime dependency is the requests library (standard third-party package). No archive downloads or unusual install behaviors are present.

Credentials

The skill declares no required environment variables or secrets, and the code does not attempt to read credentials or unrelated config. Network calls use only public API endpoints appropriate for the stated functionality.

Persistence & Privilege

The skill does not set always: true and does not attempt to persist configuration or modify other skills. Autonomous invocation is allowed by default (platform normal) but the skill itself does not request elevated persistence or cross-skill config changes.

Assessment

This skill appears coherent and only queries public literature APIs. Before installing, be aware: (1) it makes outbound HTTP requests to public endpoints (PubMed/NCBI, Semantic Scholar, bioRxiv, ClinicalTrials) so run it in an environment where outbound network traffic is acceptable; (2) it depends on the Python requests package (scripts print an error if requests is not installed); (3) API rate limits and terms of service apply — heavy/automated use may require API keys or throttling; (4) novelty scoring and TLDRs are heuristic and may surface unreviewed preprints (exercise caution when using findings for critical decisions). If you need stricter controls, review the scripts locally and run them in an isolated environment or add rate-limiting and logging controls before integrating into automated workflows.

Literature Agent v1.0.0

Overview

Dual-source literature search combining PubMed (biomedical focus) and Semantic Scholar (broader CS/ML/AI coverage). Deduplicates across sources, enriches with citation metrics and TLDR summaries.

Key capabilities:

• PubMed search with MeSH terms, abstracts, publication types
• Semantic Scholar search with citation counts, influential citations, TLDR
• Paper lookup by DOI or PMID
• Citation tracking (who cited this paper?)
• Related paper discovery (what did this paper reference?)
• Automatic query construction from compound/target/disease inputs
• Cross-source deduplication and enrichment

Quick Start

# Search by topic
python scripts/pubmed_search.py --query "KRAS G12C inhibitor" --max-results 5

# Search Semantic Scholar (includes ML/AI papers)
python scripts/semantic_scholar.py --query "graph neural network drug discovery"

# Full chain: compound + disease context
python scripts/chain_entry.py --input-json '{"compound": "sotorasib", "disease": "lung cancer"}'

# Look up a specific paper and find who cited it
python scripts/semantic_scholar.py --paper-id "DOI:10.1038/s41586-021-03819-2" --citations

# Recent papers only (last 3 years)
python scripts/pubmed_search.py --query "organometallic catalyst drug synthesis" --years 3

Scripts

scripts/pubmed_search.py

PubMed via NCBI E-utilities (public, no key required, rate limit: 3 req/sec).

--query <text>          Required. Search query
--max-results <N>       1-50 (default: 10)
--sort <type>           relevance | date (default: relevance)
--years <N>             Limit to last N years

Returns: PMID, title, authors, journal, year, DOI, abstract, MeSH terms, keywords, publication types.

scripts/semantic_scholar.py

Semantic Scholar API (public, no key required, rate limit: 100 req/5 min).

--query <text>          Search query
--paper-id <id>         Paper ID (DOI:xxx, PMID:xxx, ArXiv:xxx)
--related               Get references of a paper (requires --paper-id)
--citations             Get papers citing a paper (requires --paper-id)
--max-results <N>       1-50 (default: 10)
--year-range <range>    e.g., "2020-2026" or "2023-"

Returns: title, authors, year, abstract, TLDR, citation count, influential citations, DOI, ArXiv ID, open-access PDF URL.

scripts/chain_entry.py

Standard PharmaClaw chain interface. Searches both PubMed and Semantic Scholar, deduplicates, and sorts by citation impact.

Input keys: query, compound/name, target, disease, mechanism, reaction, topic, doi, pmid, max_results, years, context

Automatic query building: {"compound": "aspirin", "disease": "colorectal cancer"} → searches "aspirin colorectal cancer"

Chaining

From	Input	To
Chemistry Query	Compound name/SMILES	Literature → find published studies
Catalyst Design	Reaction type	Literature → find catalyst optimization papers
Literature	Key findings	Pharmacology → validate claims
Literature	Synthesis references	Chemistry Query → retrosynthesis
Literature	Patent mentions	IP Expansion → FTO analysis