Oh My ClaudeCode - Setup Guide¶

Source: yeachan-heo/oh-my-claudecode via skills.sh - 36.5K combined installs across 98 indexed listings; first documented November 3, 2026 (evening sweep) GitHub: 40,550 stars, MIT, last pushed 2026-11-03, 48 SKILL.md directories Category: Agent orchestration: planning, parallel delegation, review, debugging and delivery skills for coding agents Quality Tier: 🟡 Yellow: all five sampled skills pass Socket and Snyk, but ask-codex draws a Warn from Gen Agent Trust Hub

oh-my-claudecode (OmC) is a multi-agent orchestration layer for Claude Code, described in its README as "multi-agent orchestration for Claude Code. Zero learning curve." The repo ships 48 SKILL.md directories under skills/, while skills.sh indexes 98 listings for the same source, so the catalog is wider than the tree.

The suite is built around workflow verbs rather than reference material: plan, build, verify, release, cancel. Setup is done once through the plugin marketplace, and most daily use is a single command that expands into a delegated work loop.

Installation¶

The documented path is the Claude Code plugin marketplace. Enter these one at a time, because pasting both lines at once fails:

/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode

/plugin install oh-my-claud

If you only want individual skills inside Hermes, clone the repo and copy the directories you need out of skills/ into your Hermes skills directory, then load them by directory name:

git clone https://github.com/Yeachan-Heo/oh-my-claudecode

Run the setup skill first if you want the suite configured for you rather than by hand:

cat skills/omc-setup/SKILL.md

What It Provides¶

Twenty of the 98 indexed listings clear 795 installs. The table below reports every one of them at its exact count.

Skill	Installs	Use For
ai-slop-cleaner	1,618	Stripping AI-sounding prose and filler from generated text
ralplan	1,223	Plan-first workflow before any edit
hud	1,173	Status display for a running session
deepinit	1,058	Deep project initialisation pass
autopilot	1,039	Unattended execution loop
deep-interview	1,037	Structured requirements interviewing
omc-setup	1,026	Installing and configuring the suite
ralph	1,009	Iterative build loop from a written spec
visual-verdict	931	Visual check of rendered output; not in the pack's visible tree sample
team	916	Parallel team-of-agents delegation
ultraqa	902	Extended QA sweep; not in the pack's visible tree sample
project-session-manager	894	Session state held across a project
cancel	888	Stopping a running workflow
omc-doctor	869	Diagnosing a broken install
skill	865	Creating and loading skills
release	860	Release preparation
external-context	818	Pulling outside context into the task
trace	817	Tracing execution paths
configure-notifications	809	Notification channel setup
omc-plan	795	Plan mode entry; not in the pack's visible tree sample

The remaining 78 indexed listings range from 3 to 765 installs.

Why This Matters for Hermes Agents¶

Hermes agents are good at single-turn tool calls and weak at long multi-step delivery unless something forces a plan, a check and a hand-off. OmC packages exactly that structure as prompt-level skills, so a Hermes builder can lift the workflow shape without adopting the Claude Code runtime.

The two most portable pieces are ralplan, which makes the agent write a plan before touching files, and ultraqa or verify, which forces a review pass before the agent claims success. Both map cleanly onto a Hermes loop that has a planning step and a verification step.

The ai-slop-cleaner skill is useful outside coding: it is a documented pass for removing filler and templated phrasing, which is the same problem public-facing copy has.

The team and ultrapilot skills show how the project splits work across parallel agents and merges results, which is a reasonable reference for anyone writing their own delegation prompts.

Treat the plugin install instructions as Claude Code specific. In Hermes you are copying prompt documents, not installing a runtime, and the cancel-* companions exist because those loops need an explicit stop.

Usage¶

You say	What happens
"Plan this refactor before you touch anything"	ralplan produces a written plan and holds edits until the plan is agreed
"Run the build loop until the tests pass"	ralph iterates against the spec, then reports the remaining failures
"Clean the AI tone out of this page"	ai-slop-cleaner rewrites filler and templated phrasing
"Interview me about the requirements first"	deep-interview asks the questions that change the design
"Have a team of agents review this module"	team splits the review across roles and consolidates findings
"Why is the install broken?"	omc-doctor checks configuration and reports what is missing
"Stop the running workflow"	cancel halts the active loop and returns control

Verification¶

Confirm the install location and the directory count first:

ls ~/.claude/skills | wc -l

Then read one raw skill file to confirm the content is what skills.sh indexed. This URL is built from the skills/omc-setup/SKILL.md path in the repo tree, on branch main:

curl -fsSL https://raw.githubusercontent.com/yeachan-heo/oh-my-claudecode/main/skills/omc-setup/SKILL.md | head -30

If the fetch returns frontmatter with a name and a description, the source is live and the guide's counts can be re-checked on the skills.sh page.

Security¶

skills.sh verdicts for sampled skills (verified November 3, 2026) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
ai-slop-cleaner	Pass	Pass	Pass
skillify	Pass	Pass	Pass
minimal-prose-discipline	Pass	Pass	Pass
review	Pass	Pass	Pass
ask-codex	Warn	Pass	Pass

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified November 3, 2026: 36,471 combined installs across 98 indexed listings; 40,550 GitHub stars; MIT; last pushed 2026-11-03. Counts drift over time.
The documented install path is the Claude Code plugin marketplace. No Hermes installer is described in the pack, so copying directories by hand is the practical route.
98 indexed listings against 48 SKILL.md directories means the catalog and the tree do not map one to one; some listings have no directory in the visible tree sample.
ask-codex carries a Gen Agent Trust Hub Warn, and several skills emit shell commands, so read a skill before letting it run.
Install counts are dominated by workflow verbs and a prose cleaner. They say little about the quality of the cancel-* and configure-* companions at the bottom of the list.
The republished documentation site and CLI reference are external, so version drift between the README and the skills is possible.

Pro Tips¶

Install omc-setup first and let it write the configuration; hand-editing the plugin files is the common source of a broken install.
Pair ralplan with verify. The plan alone does not stop an agent from declaring success early.
Skip the cancel-* companions until you actually run long loops, then add the one you need so the agent has a stop word.
Use ai-slop-cleaner as a final pass on any public copy generated inside the workflow, not as a first draft step.
Check the skills directory count against the 48 in this guide after an upgrade, since listings get added faster than directories.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ