Hermes Agent NEW Skill Bundles (Setup Guide)

Hermes Skill Bundles let you group skills into one named slash command instead of invoking them one-by-one or hoping the agent guesses the right set.

What It Solves

AI agents are probabilistic, so typing everything into one natural-language prompt gives inconsistent results — the agent often skips 2 or 3 of the 5 skills you actually need for a task. Bundles make the skill set explicit.

How A Bundle Works

A bundle is a tiny YAML file under ~/.hermes/skill-bundles/ with three required-ish fields: name (becomes the slash command, e.g. /backend-dev), skills (array of skill names like github-code-review, tdd, github-pr-workflow), and instructions (your house rules — "a mini agents.md"). description is optional. Naming your workflow research gives you /research.

Build It Three Ways

1. CLI: run /bundles create, then /bundles list, /bundles reload after manual YAML edits. 2. Hermes dashboard → Skills tab, pick and combine. No drag-and-drop yet — the creator flags this as a missing feature expected in the next Hermes update. 3. Ask the agent itself: feed it the official News Research skill bundle documentation and prompt "create a bundle for general research." The creator's Hermes produced two bundles from one request — a full general-research and a quick-research lightweight variant — showing it can propose alternatives.

Test Run

Running /research find out why Hyperliquid token is pumping hard returned a full report (the buyback engine thesis). The creator notes he bought HYPE at $32, sold at $40, and watched it hit $56–$57.

Open Questions And Bugs

Ordering is not documented as adaptive — skills may run in fixed sequence. Verify with your agent before shipping. Also spotted: switching models in Hermes (e.g. DeepSeek V4 Pro) keeps the previous model's context window — the creator's stuck on the Kimiko 2.6 window until he hard-codes the config.

Team Use

Commit the YAML files to a dotfiles repo so the whole team loads the same bundles. Keep each bundle intentionally scoped to one workflow.