Hermes Agent setup: research skills, Zotero and Consensus

This tutorial upgrades a researcher in Hermes Agent bot mode and tests it within a team that includes an orchestrator and a librarian. The presenter adds a reusable research method, access to an existing Zotero library and academic paper discovery through Consensus MCP.

The research skill supports balanced, academic and current-practice approaches, plus a combined pass that compares academic evidence with current documentation. It uses Hermes' grounded citations skill for source tracking. The researcher prepares reports; the librarian handles admission to the LLM wiki.

For Zotero, the setup enables communication with other applications through its local API. The presenter demonstrates an agent running inside Docker accessing Zotero on the desktop. His custom skill permits reads only. An initial title search misses a relevant paper, while a broader query later retrieves more useful material.

Consensus requires an authenticated MCP connection. Its discovery skill treats summaries as leads and requires inspection of original papers before making claims. The test concerns agentic memory poisoning and passes a report to the orchestrator for review. The presenter has not checked every paper or reviewed the report in depth. He approves it despite requested provenance revisions, then reviews the librarian's proposed wiki entry. Some handoffs remain manual; scheduled research and more structured coordination are proposed next steps.