Hermes Agent Kanban: multi-agent workflow walkthrough

This walkthrough explains a Hermes Agent workflow that turns reports of user problems into tool prototypes or video drafts. The speaker uses agent profiles on one machine, with Grok assigned to the X scout and GPT-5.5 assigned to the other roles in the demonstration.

The Kanban board holds work cards in a single SQLite file. A dispatcher claims ready cards and starts the assigned profile in a clean workspace. Dependencies let research tasks run in parallel before the orchestrator reviews their combined results. The speaker describes persistent task state, recovery after failures and an audit log as benefits of this design.

Scouts search X and the web, then the orchestrator removes duplicates and shelves candidates scoring below 65/100. Three researchers examine each surviving issue. Build and video proposals reach a human approval gate in Telegram before production proceeds. The gateway must remain running for that messaging path.

The demonstration produces a Codex configuration reporting CLI and a video draft about Claude Code subagents and MCP tools. The speaker also shows recovery from slides saved in temporary workspaces. These outputs still need review: the CLI reports configuration details without modifying them, and it does not establish what the VS Code extension enforces. A generalized open source workflow template accompanies the video.