Oh My Hermes (OMH)

Multi-agent orchestration skills for Hermes Agent, inspired by oh-my-claudecode and rebuilt natively for Hermes primitives.

OMH provides composable skills for consensus planning, requirements interviewing, and verified execution — plus an optional plugin that adds hook-based role injection, atomic state management, and evidence gathering. Skills work standalone with zero dependencies.

Skill	What It Does
omh-deep-research	Multi-phase web research: decompose → parallel search → synthesize → verify citations
omh-ralplan	Consensus planning: Planner → Architect → Critic debate until agreement
omh-ralplan-driver	Dispatcher's playbook for driving an omh-ralplan run — context-package authoring (where quality is born), round dispatch, distillation, final review
omh-deep-interview	Socratic requirements interview with coverage tracking
omh-ralph	Verified execution: implement → verify → iterate until done
omh-ralph-driver	Dispatcher's playbook for driving an omh-ralph run — plan-shape, parallel batching, evidence gathering, verifier discipline, strike categorization, Step-7 final architect review, commit hygiene
omh-ralph-task	Executor's discipline for a single omh-ralph task — task-envelope contract, file-scope rigidity, stash-verify-against-HEAD for sibling-task isolation, commit-author override, structured report-back shape
omh-triage (v0.1)	Multi-role consensus triage of an issue backlog — Maintainer (code-anchored) + Skeptic (pruning); more roles coming after lived rounds. See witt3rd/oh-my-hermes#9
omh-triage-driver (v0.1)	Dispatcher's playbook for driving an omh-triage run — pre-flight backlog audit, role-pass dispatch, distillation, user sign-off gate
omh-autopilot	Full pipeline composing all three skills end-to-end

Composition (recommended pipeline for unfamiliar domains):

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

(Or fold omh-deep-research in as Phase -1 of omh-autopilot when the domain is unfamiliar; otherwise start at the interview.)

Install

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-autopilot

Or copy skills/<name>/ to ~/.hermes/skills/omh/ manually.

For the optional plugin: install plugins/omh/ to ~/.hermes/plugins/omh/ (requires Python 3.10+ and pyyaml).

For local development (live edits via symlinks), see CONTRIBUTING.md.

Getting Started

• Need background on an unfamiliar domain? → omh-deep-research
• Just need a plan? → omh-ralplan
• Driving a ralplan run yourself? → omh-ralplan-driver (load alongside omh-ralplan)
• Vague idea? → omh-deep-interview then omh-ralplan
• Have a plan, need execution? → omh-ralph
• Driving a ralph run yourself? → omh-ralph-driver (load alongside omh-ralph)
• Grooming an issue backlog? → omh-triage (load alongside omh-triage-driver if you're driving)
• End-to-end? → omh-autopilot

OMH self-seeds a .omh/ directory in the project on first use (with the plugin installed) — including a README explaining the convention and a .gitignore pre-configured for selective sharing. To scaffold up-front without running a workflow, call omh_state(action="init").

Known Gaps

• wiki/fact_store/memory persistence is not yet integrated for research artifacts produced by omh-deep-research. The confirmed report sentinel (.omh/research/{slug}-report.md with status: confirmed) is the durable interface in v1; downstream skills consume it directly. Persisting findings into fact_store or wiki is a deferred Q2 item.
• Per-call subagent tool scoping for the omh-deep-research verifier may be unavailable depending on Hermes install; the READ-ONLY contract is enforced by prose in role-research-verifier.md in that case (A5).

Cost Envelope (omh-deep-research)

A typical happy-path session is roughly 5-8 delegate_task calls (3-5 researchers + 0-1 followup + 1 synthesist + 1 verifier). With one synthesis retry, expect up to ~10-12 calls. The 3-strike retry cap bounds worst-case at ~14-16 calls before BLOCKED is surfaced.

Requirements

Hermes Agent v0.7.0+. The plugin additionally requires Python 3.10+ and pyyaml.

Documentation

• docs/concepts.md — How the four skills work
• docs/plugin.md — The v2 plugin (roles, hooks, tools)
• docs/omh-delegate.md — Hardened delegation wrapper
• docs/omc-comparison.md — Origins and design choices vs OMC
• docs/hermes-constraints.md — How OMH works around Hermes limits
• docs/gaps.md — What's not built yet
• ROADMAP.md — Versions and direction

License

MIT