cyc2002tommy/deep-research-agent

Deep Research Agent (also called Deep Science Writer) is a skill for AI coding agents such as Hermes that automates the process of conducting a scientific literature review and producing a written academic report. Rather than summarizing abstracts, the system is designed to download and read full paper texts before drawing conclusions, with the goal of reducing AI-generated statements that are not actually supported by the source papers. The pipeline runs in seven phases. It starts by building a search plan and pausing for the user to approve it before proceeding. It then queries academic databases (Scopus via an MCP server, OpenAlex, and optionally Exa for neural search) to find relevant papers, applies quality filters that restrict results to Q1 and Q2 journals and explicitly exclude MDPI publications, and downloads full texts for a subset of the most relevant results. After extraction, it drafts a structured article with APA 7th edition citations. An anti-hallucination phase strips AI-style phrasing and pings every generated DOI to confirm the links are real, deleting any citation whose DOI returns a 404. An internal peer-review step then rewrites the draft until it meets academic tone standards. Finally, Python scripts generate charts and compile everything into a formatted Microsoft Word document. The skill also handles knowledge management. After the report is compiled, it saves a summary to an Obsidian vault and uploads each cited reference individually to Google NotebookLM for audio overview and cross-referencing. Setup requires a Hermes Agent (or compatible runner), Python 3.10, Node.js, a free Elsevier Scopus API key, and a Google account authenticated with the NotebookLM MCP server. The output directory defaults to a specific Windows path in the configuration, so users on other systems need to adjust the path settings. The project is MIT licensed.