Oh My Hermes: 36 Skills, 7 Agents, One Kanban Board
Oh My Hermes is a community skill collection that layers a complete software engineering organization onto Hermes Agent. It ships with 36 skills, 7 role-specific agent profiles, and 13 shell scripts that handle setup, orchestration, and monitoring. Install once, message your Hermes bot "set up the CTO loop," and it inspects your project, asks up to three questions, and begins operating.
The project crossed 695 stars in two months since its May 8, 2026 creation. It is the first community-built skill pack to give Hermes a structured multi-agent workflow comparable to what teams build with crewAI or AutoGen - but running natively inside Hermes, without external orchestration dependencies.
Architecture: CTO-led Kanban with role specialization
The seven agent profiles map to a standard product team. Each has a dedicated role definition file in agents/ that Hermes loads at startup:
| Agent | Role | Definition size |
|---|---|---|
| CTO | Product lifecycle orchestration | 3,729 bytes |
| Product Manager | Positioning, requirements, growth | 2,315 bytes |
| Designer | UX, visuals, design verification | 2,520 bytes |
| Developer | Implementation, PRs, code generation | 1,740 bytes |
| QA | Review, testing, verification | 1,761 bytes |
| Security | Release gate, risk assessment | 2,063 bytes |
| Ops | Deployments, monitoring, incident response | 1,877 bytes |
| Total | 7 profiles | 16,005 bytes |
The CTO agent is the coordinator. It decomposes incoming work - whether from a GitHub issue, a chat message, a product idea, or a cron trigger - into tasks on a Kanban board. Each task routes to the appropriate specialist. The CTO tracks progress, handles dependencies, and assembles the final output.
Salomondiei08 describes it as "Oh My Zsh for Hermes" - a curated configuration layer that turns a general-purpose agent into a specialized development platform. The analogy holds: just as Oh My Zsh gives Zsh users a productive starting point with plugins and themes, Oh My Hermes gives Hermes users a productive starting point with skills and agent profiles.
Skill coverage: full product lifecycle
The 36 skills span the entire build-ship-operate cycle. They are standard Hermes .md skill files - no plugin required. The project also includes an optional v2 plugin for hook-based role injection, atomic state management, and evidence gathering, but the skills work standalone.
Major categories:
Requirements and planning: clarify-requirements reads existing context before asking questions. product-brief writes a compact source of truth with acceptance criteria. kanban-task creates and updates cards on the Hermes Kanban board at every stage.
Implementation: choose-engine routes tasks to Hermes, Claude Code, or Codex based on complexity. implement-with-claude-code scaffolds full context and scope constraints. implement-with-codex targets single-file fixes. The Developer agent uses these to pick the right tool for each subtask.
Review and quality: review-github-pr verifies the product increment against the brief, then approves or requests changes. security-review runs a tool-backed release gate with daily and weekly assessments. await-merge-approval presents the founder with four options: YES, NO, CLOSE, or LATER.
Deployment and operations: deploy-to-vercel runs pre-deploy checks, deploys, and captures the URL. health-check validates /api/health, Supabase connectivity, and Vercel logs. observe-logs deduplicates runtime errors and escalates only actionable changes. backup-hermes-data tarballs the Hermes state directory to S3, Dropbox, or local storage.
Product and marketing: product-marketing handles positioning, copy, SEO, launch strategy, and content scheduling. creative-production generates product assets and HyperFrames launch videos. publish-with-buffer dry-runs and schedules posts through the Buffer CLI.
Meta-operations: create-skill is a meta-skill that generates new skill files in the correct format. failure-recovery saves failed cron context to dead-letter logs and alerts the founder. cto-status-report delivers a daily morning summary of what is in progress, done, and blocked.
Setup: three commands to a running team
The scripts/setup-cto.sh script (15,379 bytes, the largest file in the repo) handles profile creation, Kanban initialization, and cron scheduling. It asks at most three questions, with sensible defaults for everything else.
git clone https://github.com/Salomondiei08/oh-my-hermes.git
cd oh-my-hermes
./scripts/setup-cto.sh
After setup, message your Hermes bot: "set up the CTO loop." The agent inspects the project directory, reads existing configuration, and begins operating. The onboarding skill handles inference of existing setup to avoid duplicate crons or conflicting profiles.
The project runs on Hermes Agent v0.16 and above. The scripts assume a standard Hermes installation with ~/.hermes/ as the state directory.
Comparison with Hermes-native Kanban
Hermes Agent introduced its own Kanban board in v0.17, with SQLite-backed task tracking, dispatcher coordination, and multi-profile isolation. The v0.18 release hardened the database layer with WAL concurrency fixes and busy-timeout PRAGMA settings. That infrastructure is what Oh My Hermes runs on top of - it does not replace Hermes's Kanban system. It provides the skills, agent profiles, and workflow conventions that make the Kanban system productive for a specific use case: building and shipping software.
The relationship is similar to how a framework sits on top of a language. Hermes provides the primitives (memory, skills, Kanban, cron, multi-profile dispatch). Oh My Hermes provides the conventions (CTO-led orchestration, role specialization, complete pipeline skills).
What the project does not do
The skills are opinionated toward a specific stack: Vercel for hosting, Supabase for data, Buffer for social, Slack for notifications. This is not a general-purpose agent toolbox - it is a workflow layer for a specific kind of project. If you deploy to Railway or use PostHog instead of Sentry, you will need to adapt the relevant skills.
The plugin layer (hook-based role injection, evidence gathering) is optional and requires pyyaml. The core skills have zero Python dependencies.
The project does not run a "team" of simultaneously executing agents in parallel. Each agent profile is a separate Hermes session. The CTO coordinates by creating Kanban cards and dispatching tasks to the appropriate profile. This is sequential orchestration through a shared task board, not concurrent multi-agent negotiation. The distinction matters for latency - a full ship-this-idea pipeline runs end-to-end through sequential stages, not parallel workstreams.
[^1]: Salomondiei08. "Oh My Hermes." GitHub. May 2026.
[^2]: mrru5s3ll. "Oh My Hermes transforms Hermes Agent into a full software development platform." X. July 13, 2026.
[^3]: Julian Goldie. "I barely write code any more. My agents write it." X. July 13, 2026.
[^4]: Nous Research. "Hermes Agent - the personal AI agent that runs on your machine." GitHub.