Hermes Workflows — Development skill for Claude Code

Hermes Workflows

Multi-step automations for [Hermes Agent](https://github.com/NousResearch/hermes-agent) — drawn as a graph, run on Hermes' own primitives, with a human in the loop only where you put one.

Hermes Workflows is a dashboard plugin for Hermes Agent. You draw an automation as a graph — agent tasks, shell steps, branches, and review gates — and it runs on top of Hermes' own Kanban, Cron, and Profiles. It is not a second engine: every node compiles to a native Hermes primitive, so a workflow is something you can read, schedule, and reason about with the tools Hermes already gives you. Where a single scheduled prompt (an [Automation Blueprint](#workflows-vs-automation-blueprints)) isn't enough, a workflow is the layer above it.

Hermes Workflows now supports release-oriented runs that resume failed runs, export portable template bundles, and drive adopted board cards one at a time on a shared feature branch so cross-card context stays intact while docs and version updates happen once for the whole scope.

Why

• It runs itself. Hermes Workflows advances the run as soon as a workflow card finishes. Native Kanban lifecycle hooks move card-driven nodes forward in seconds, while the configurable self-terminating cron tick stays as the safety net and wait-node poll.
• Nothing is locked away. A workflow is a plain spec you export to YAML/JSON, re-import on a clean install, and edit by hand or in the visual editor. Data flows between nodes through the run state, not a host file path baked into the graph — so it stays portable.
• It speaks Hermes, not a private dialect. Nodes become native Kanban cards, Cron jobs, and Profile assignments; results deliver through the host's own delivery router; skills come from the host catalog. There is no parallel runtime to learn or trust.
• You see what happened. Every agent node reports live per-node telemetry (duration, tokens, tool calls, errors), pending dangerous-command approvals surface