Tune SOUL.md

SOUL.md is the highest-leverage thing you control. A generic SOUL.md makes a generic agent, and time spent here compounds with every conversation.

01
Give it a persona and a hard scope "You are a financial research assistant focused on Indonesian banking and telecommunications. You always pull live data from Sectors before answering about a stock. You do not speculate about future prices or make buy/sell recommendations."
02
Define output format explicitly Want tables for comparisons? Prices in IDR? Say so once in SOUL.md, and stop repeating the same request every session.
03
Set action guardrails For anything that sends a message or writes a file: "Before sending any message or writing any file, confirm with the user unless the action is part of a pre-approved scheduled task."
04
Iterate in small steps Change one instruction at a time, run the same test prompt before and after. Keep notes on what improved, don't rewrite everything at once.

Review Your Skill Library

After a few sessions, Hermes will have auto-generated skills, not all of them good. Regular curation keeps the library clean and the agent fast.

List and search skills, from within a chat
/skills
/skills search <term>

Browse published skills in the Hermes Hub. With skills.write_approval on, review pending writes:

Review staged skill writes
/skills pending
/skills diff <id>
/skills approve <id>
/skills reject <id>

Each skill is plain markdown in ~/.hermes/skills/, open, edit, or delete it like any other file.

What to Look For

  • Overly specific skills, written for one exact prompt, not reusable. Inflate the library without adding value.
  • Skills based on bad reasoning, sometimes generalized from a poor first attempt. Delete and let Hermes rebuild from better inputs.
  • Duplicates, two skills doing the same thing, written at different times. Keep the better one.
A leaner library runs faster. Cull monthly once the agent's been running a while.

Deployment Options

The same three install paths from Module 03, revisited now that you know what you need.

Terminal

Runs while your machine is on, good for experimentation, zero infrastructure cost. You don't need a VPS to get value from Hermes.

Desktop App

Same core, config, sessions, skills, memory, with a polished UI. Already on terminal? Launch it with hermes desktop.

VPS + Docker

Always-on, independent of any single machine, the path for workflows that need to run while your laptop's closed. Upgrading is pulling a new image, no lost config, skills, or memory.

Hermes vs OpenClaw: When to Use Which

Both tools work. Switching from OpenClaw? hermes claw migrate imports persona, memory, skills, model config, and tokens automatically, previewing and backing up first. A few things like HEARTBEAT.md need manual redoing.

OpenClaw

Broad channel coverage (Discord, Telegram, WhatsApp, Slack, all at once) and a large skill library on ClawHub, installable in one command. Better fit when breadth and quick deployment matter more than an agent that rewrites its own procedures.

★ Hermes

A personal daily workflow that repeats and improves over time. Memory and learning built in, no add-ons. Safer defaults with command approval. Trade some flexibility for an agent that gets meaningfully better the more you use it.

Community & Next Steps

Where to go after this workshop ends.

01
Hermes docs, hermes-agent.nousresearch.com Install, MCP, skills, delegation, voice mode, and cron, in more depth than this workshop has time for.
02
GitHub, github.com/NousResearch/hermes-agent Issues, discussions, the source. Fixes show up here first.
03
Community hub or Discord Check the docs site for the current link before sharing one with a class.
04
Sectors API Docs, sectors.app The full API reference for designing custom scheduled workflows.

You've Finished the Course

You have a working Hermes agent connected to live Indonesian market data and scheduled workflows. The rest is iteration.

← Back to Course Home Hermes Docs