Generate personalized literature reviews from a one-line topic
Suggested workflow: Use a custom agent to generate personalized literature reviews from a one-line topic. The agent coordinates APIs (OpenAlex, Semantic Scholar, CrossRef, Unpaywall, arXiv, PubMed) to retrieve, deduplicate, and structure data, then generates reviews in Markdown, Word, or PDF formats following citation standards. Task: The author developed a custom 'literature review' Skill for an AI Agent to generate personalized reviews from a single-line topic. The workflow involves defining search boundaries (keywords, timeliness, domain routing), retrieving data via six academic APIs (OpenAlex, Semantic Scholar, CrossRef, Unpaywall, arXiv, PubMed), and deduplicating results using DOI and title similarity thresholds. The agent structures data into JSONL and generates outputs in Markdown, Word, or PDF formats following GB/T 7714 citation standards. Human involvement includes initial topic definition and final review of the generated content.; tools: custom-built or unnamed agent. Human role: The original post does not say how a person stayed involved.; limitations: The tool relies on the availability and rate limits of external academic APIs. It requires a compatible AI Agent framework to execute the Skill. The source notes that the online version (Feishu Docs) is less effective than the local version. Users must manually verify citations, as the system prioritizes real database records but may still miss non-indexed literature. The post is behind a paywall, limiting full access to the Skill code..
Tools
custom-built or unnamed agent
Published on Agent Work Atlas:
Overview
Suggested workflow: Use a custom agent to generate personalized literature reviews from a one-line topic. The agent coordinates APIs (OpenAlex, Semantic Scholar, CrossRef, Unpaywall, arXiv, PubMed) to retrieve, deduplicate, and structure data, then generates reviews in Markdown, Word, or PDF formats following citation standards.
Workflow
The author developed a custom 'literature review' Skill for an AI Agent to generate personalized reviews from a single-line topic. The workflow involves defining search boundaries (keywords, timeliness, domain routing), retrieving data via six academic APIs (OpenAlex, Semantic Scholar, CrossRef, Unpaywall, arXiv, PubMed), and deduplicating results using DOI and title similarity thresholds. The agent structures data into JSONL and generates outputs in Markdown, Word, or PDF formats following GB/T 7714 citation standards. Human involvement includes initial topic definition and final review of the generated content.
Reported result
The author reports that the tool has been iterated and promoted within their team for over six months, with students expressing satisfaction. However, the source does not provide quantitative metrics on time saved or error reduction for this specific tool. It cites external research showing that standard AI models without such tools have high hallucination rates (e.g., only 26.5% of AI-generated references are fully correct). Actual performance metrics for this specific Skill are not disclosed.
Human role
The original post does not say how a person stayed involved.
Limitations
The tool relies on the availability and rate limits of external academic APIs. It requires a compatible AI Agent framework to execute the Skill. The source notes that the online version (Feishu Docs) is less effective than the local version. Users must manually verify citations, as the system prioritizes real database records but may still miss non-indexed literature. The post is behind a paywall, limiting full access to the Skill code.
Try it
Act as a research assistant. I need to generate a literature review on [replaceable topic, e.g., 'LLM agents in healthcare']. 1. Define search boundaries: Identify core keywords, synonyms, and time sensitivity (e.g., last 2 years for AI topics). 2. Retrieve data: Query OpenAlex, Semantic Scholar, and arXiv/PubMed via API. Use CrossRef for DOI validation and Unpaywall for OA links. 3. Deduplicate: Merge results using exact DOI match and title similarity >0.85. 4. Structure: Save raw data in JSONL format. 5. Generate: Write a review in Markdown following GB/T 7714 citation style. First, perform a dry run with 5 sample papers and show me the JSONL output and citation format for confirmation before generating the full review. If APIs are unavailable, list the required endpoints and ask for mock data.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.