Oh My Hermes bolts a CTO team onto the Hermes Agent
An opinionated workflow layer for building, shipping, and operating apps with Hermes Agent
At a glance
- What is it?
- Oh My Hermes is a workflow layer for the Hermes Agent that bundles 36 skills and seven role agents behind one CTO loop. This write-up covers the install path, how work routes to Claude Code or Codex, and what a v0.0.1 release implies.
- Who is it for?
- Oh My Hermes fits a founder or small team already running a Hermes bot who want idea-to-production automated and accept agent automation on their own repos and accounts. Do not point it at a codebase you cannot roll back, and read the approval list first, since only production release, rollback, public posting, licensed media, payment, and destructive account changes pause for you.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 54 days ago.
- What is it written in?
- Mainly Shell, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 17, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What the layer adds to a plain Hermes bot
Hermes, the agent runtime from Nous Research, gives you a bot you message on Telegram, Slack, Discord, or WhatsApp. Out of the box, what it does depends on how you configure skills, crons, and memory yourself. Oh My Hermes is an opinionated layer over that runtime, and the README reaches for the Oh My Zsh comparison: install it once, and the runtime becomes genuinely useful for real software projects. Concretely it ships 36 skills and seven role agents organized around one CTO loop, backed by Hermes Kanban, memory, cron, and completion evidence. The README insists it is not a chatbot wrapper and not a prompt pack; these are skills Hermes loads and runs autonomously on your VPS or laptop, wherever the bot lives. The alternative is running Hermes bare, where you define each skill and schedule by hand. The difference in approach is that this layer arrives with the hierarchy, the delivery integrations, and the recurring jobs already wired, at the cost of adopting its opinions about how products get built.
The CTO loop and its seven agents
Work follows one arc the README states as Understand, Design, Build, Check, Ship, Learn, and it can start from an idea, customer feedback, production logs, analytics, or an issue. A founder starts the loop; a CTO agent owns lifecycle, roadmap, delegation, and decisions, and coordinates six more. Product covers briefs, priorities, positioning, SEO, and content. Designer covers UX, visual verification, and launch assets. Builder produces working product increments. Reviewer handles journeys, visual and accessibility checks, and PR review. Security owns release risk plus daily and weekly assessment. Ops handles deploy, health, deduplicated logs, and incidents. Underneath, the loop leans on Hermes Kanban, memory, cron, and completion evidence. Hermes reads the project before asking anything, asks at most three questions with recommended defaults, and continues when you skip them. Claude Code and Codex remain optional implementation engines rather than part of the hierarchy, and Computer Use is described as a guarded shared skill for native or authenticated GUI tasks, not another agent.
Installing on top of an existing Hermes bot
Step one is not this repo. Follow the Hermes quickstart until you have a bot you can message, since the README badge requires Hermes v0.16 or newer. Step two installs the layer:
git clone https://github.com/salomondiei08/oh-my-hermes /tmp/oh-my-hermes
bash /tmp/oh-my-hermes/install.shThe README also supports curl piped to bash, with the installer self-cloning a temporary copy when repo files are not present. Step three is a message to your bot: set up the CTO loop. The bot inspects the project, asks only for missing settings that matter, and continues with defaults when you skip them. A table in the README maps the later steps: GitHub connects when useful and enables issues and PR delivery, a production URL enables health and log observation, the /goal command keeps the agent focused across long sessions, and from there it turns autonomous with product review, a daily report, and recurring security and log checks.
How implementation work is routed
The choose-engine skill routes a task to Hermes itself, Claude Code, or Codex, and the two implementation skills shape what those engines see. implement-with-claude-code scaffolds Claude Code with full context plus scope constraints, which is the pairing for building out a feature. implement-with-codex is scoped for targeted single-file fixes. The flagship ship-this-idea skill runs the whole flow from idea to brief to design to build to verify to ship. On the delivery side, deploy-to-vercel runs pre-deploy checks, deploys, and captures the URL; connect-supabase links Supabase, pushes migrations, and sets Vercel environment variables; setup-monitoring configures Sentry and Uptime Kuma; and health-check calls /api/health, validates the response, and checks Supabase and Vercel logs. The routing matters in practice. Heavy builds never occupy the bot that answers you on Telegram, and quick fixes do not pay a full scaffold tax.
GitHub delivery with a secret scan before every PR
Delivery evidence is part of the loop, not an afterthought. manage-github-issues handles triage, creation, labelling, assignment, and closing. create-github-pr opens pull requests with a secret scan before they are created, and review-github-pr verifies the product increment before approving or requesting changes, so the Reviewer agent and a human can work the same queue. auto-issue-triage runs hourly: it scores open issues, picks the top priority, and starts work on it. The secret scan deserves a note. The README states that it happens but does not document its rules or coverage, so treat it as a useful first gate rather than a replacement for scanning in CI. The same caution applies to the triage loop. An agent that starts work on the top-scored issue every hour is only as good as the scoring, and the README does not document how the scores are computed.
What runs when you are not looking
Once the CTO loop is set up, several jobs recur without a prompt: product review, a daily report, and recurring security and log assessment. onboarding is written to avoid duplicate crons, which addresses the usual failure of layering two schedulers. failure-recovery saves failed cron and agent context to dead-letter logs and alerts the founder, so a crashed job leaves a trail rather than vanishing. project-switch changes product context without mixing repos, URLs, logs, or approvals between projects, and project-status reports gateway, model, project, crons, and integrations in founder-readable form. reset-runtime backs up and clears stale Hermes runtime state without deleting credentials. For publishing, publish-with-buffer dry-runs posts first and schedules approved ones through the official Buffer CLI, and creative-production produces launch assets with licensed music evidence attached.
The approval gate and the v0.0.1 caveats
The layer draws one line around autonomy: only irreversible actions require your approval, and the README lists them as production release, rollback, public posting, licensed media, payment, and destructive account changes. Everything else proceeds on defaults, which is the appeal and the risk. Where the project stands is worth stating plainly. The only tagged release is v0.0.1. The repository metadata does not list a license even though the README badge links to one, so verify the license file situation before redistributing it. And the layer sits on a platform it does not control. generate-with-seedance runs approved paid video shots through the official Volcengine Ark API, which means real spend behind one of the skills; the approval list covers payment, but a monthly invoice is still a thing to watch. For a tool whose core promise is acting on your behalf, the single release and the thin license story are the two facts to weigh first.
Editorial conclusion
Oh My Hermes fits a founder or small team already running a Hermes bot who want idea-to-production automated and accept agent automation on their own repos and accounts. Do not point it at a codebase you cannot roll back, and read the approval list first, since only production release, rollback, public posting, licensed media, payment, and destructive account changes pause for you. Verify Hermes v0.16 or newer is running, run the install, and watch one full CTO loop cycle before enabling the hourly triage or the autonomous daily report.
Frequently asked questions
Is Oh My Hermes free to use?
The README badge states an MIT license, and no pricing is documented for the layer itself. Note that generate-with-seedance runs paid video shots through the official Volcengine Ark API, which is separate spending.
Do I have to connect GitHub before using it?
No. The README says GitHub and a production URL can be connected later and do not block product discovery and building.
What happens if a scheduled job fails?
The failure-recovery skill saves failed cron and agent context to dead-letter logs and alerts the founder.
Which Hermes version is required?
The README badge lists Hermes v0.16 or newer, installed through the Hermes quickstart before the layer itself.
Community notes