Skip to content

Oh My Hermes (OMH) Suite — Setup Guide

Source: witt3rd/oh-my-hermes GitHub: witt3rd/oh-my-hermes (255⭐) Skills: 9 (omh-ralplan, omh-ralph, omh-deep-interview, omh-deep-research, omh-triage, omh-autopilot + 3 drivers) Category: Agent Orchestration & Multi-Agent Workflows First Seen: August 14, 2026 sweep Quality Tier: 🟡 Beta

Oh My Hermes (OMH) is a multi-agent orchestration framework for Hermes Agent, inspired by oh-my-claudecode and rebuilt natively for Hermes primitives. It provides composable skills for consensus planning (Planner → Architect → Critic debate), Socratic requirements interviewing, parallel research with citation verification, and evidence-verified execution — plus an optional plugin adding hook-based role injection and atomic state management. Skills work standalone with zero dependencies.

Note: distinct from the earlier-documented oh-my-hermes-workflow skill (publisher aradotso/hermes-skills). This is a different, larger suite from publisher witt3rd/oh-my-hermes.

Overview

Skill Installs What It Does
omh-ralplan 93 Consensus implementation planning — Planner + Architect + Critic debate until agreement (≤3 rounds)
omh-ralplan-driver 91 Dispatcher's playbook for driving an omh-ralplan run — context package, rounds, distillation
omh-ralph 91 Verified execution — one task per call, evidence required, "iron law" of proof
omh-ralph-driver 90 Dispatcher's playbook for omh-ralph — parallel batching, evidence gathering, commit hygiene
omh-ralph-task 90 Executor's discipline for one omh-ralph task — file-scope rigidity, sibling isolation
omh-deep-research 95 Parallel web research — decompose → subagents → synthesis → citation verification
omh-deep-interview 92 Socratic requirements interview — clarifies vague or ambiguous goals with coverage tracking
omh-triage 89 Multi-role consensus triage of an issue backlog (v0.1)
omh-triage-driver — Dispatcher's playbook for omh-triage runs — pre-flight audit, role-pass dispatch
omh-autopilot 96 End-to-end pipeline composing all skills: interview → plan → execute → QA → verify (v2.0.0)

Recommended composition pipeline for unfamiliar domains:

omh-deep-research → omh-deep-interview → omh-ralplan → omh-ralph

Installation

Native Hermes install (preferred):

hermes skills tap add witt3rd/oh-my-hermes
hermes skills install omh-deep-research omh-ralplan omh-ralplan-driver omh-deep-interview omh-ralph omh-ralph-driver omh-ralph-task omh-triage omh-triage-driver omh-autopilot

Or via skills.sh:

npx skills add witt3rd/oh-my-hermes

Manual alternative: copy skills/<name>/ to ~/.hermes/skills/omh/.

Optional plugin (hook-based role injection, atomic state management): copy plugins/omh/ to ~/.hermes/plugins/omh/ (requires Python 3.10+ and pyyaml).

Prerequisites

Requirement Details
Hermes Agent Recent version with hermes skills tap support
Python 3.10+ Only for the optional omh plugin (pyyaml)
Terminal toolset omh-autopilot metadata requires the terminal + omh toolsets
API keys (research) omh-deep-research uses your configured search/web tools — none extra

Key Capabilities

Multi-Role Consensus (omh-ralplan + driver)

Trigger: "plan this", "make a plan", "consensus plan". Planner, Architect, and Critic roles debate the implementation plan until consensus (max 3 rounds). The driver provides context-package authoring (where plan quality is born), round dispatch, and final review.

Verified Execution (omh-ralph + driver + task)

Trigger: "execute the plan", "implement this". One task per call, with evidence required for every claim — screenshots, logs, test output. The driver handles parallel batching and strike categorization; the task skill enforces file-scope rigidity and stash-verify-against-HEAD sibling isolation.

Deep Research (omh-deep-research)

Trigger: "research this domain". Decomposes the topic, fans out to parallel subagent searches, synthesizes findings, and verifies every citation before presenting.

Requirements Interview (omh-deep-interview)

Trigger: "what should we build", vague goals. Socratic questioning with coverage tracking until the requirements are unambiguous.

Backlog Triage (omh-triage + driver)

Trigger: "triage these issues". Maintainer (code-anchored) + Skeptic (pruning) roles reach consensus on issue priority. v0.1 — more roles coming.

End-to-End Autopilot (omh-autopilot)

Trigger: "autopilot", "build me", "handle it all". Composes all skills into a multi-session pipeline: idea → interview → plan → execute → QA → verify.

Quick Start

  1. hermes skills tap add witt3rd/oh-my-hermes && hermes skills install omh-deep-research omh-ralplan omh-ralph omh-autopilot
  2. Start with omh-deep-interview when goals are vague — it forces the user to pin requirements
  3. Run omh-ralplan to get a consensus plan before any non-trivial implementation
  4. Execute with omh-ralph — every completed task requires evidence
  5. For unfamiliar domains, prepend omh-deep-research to the pipeline

CorpusIQ Use Cases

Use Case How
Feature planning Use omh-ralplan consensus debates for CorpusIQ product specs instead of single-pass planning
Growth research Use omh-deep-research for market/competitor research with citation verification before writing content
Multi-agent task execution Use omh-ralph verified-execution discipline in CorpusIQ supervisor-agent waves (complements the existing corpusiq-supervisor-agent wave dispatch)
Requirement extraction Use omh-deep-interview on vague founder requests to pin scope before build
Backlog management Use omh-triage for issue backlog triage in corpusiq-docs and CorpusIQ repos

Limitations / Verification

  • omh-triage is v0.1 — Maintainer + Skeptic roles only, more roles planned
  • omh-autopilot is multi-session — not for single-session trivial edits
  • Skills run standalone, but full behavior (role injection, state) needs the optional omh plugin
# Verify skills installed
hermes skills list | grep omh-

# Functional test — ask Hermes in session:
#   "omh-deep-interview: interview me about a vague idea for a landing page"

Security

← Skills Catalog | Marketplace →

Powered by CorpusIQ