{
  "markdown": "# Oh My Hermes (OMH)\n\nMulti-agent orchestration skills for [Hermes Agent](https://github.com/NousResearch/hermes-agent),\ninspired by [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode)\nand rebuilt natively for Hermes primitives.\n\nOMH provides composable skills for consensus planning, requirements\ninterviewing, and verified execution — plus an optional plugin that adds\nhook-based role injection, atomic state management, and evidence gathering.\nSkills work standalone with zero dependencies.\n\n| Skill | What It Does |\n|-------|--------------|\n| **omh-deep-research** | Multi-phase web research: decompose → parallel search → synthesize → verify citations |\n| **omh-ralplan** | Consensus planning: Planner → Architect → Critic debate until agreement |\n| **omh-ralplan-driver** | Dispatcher's playbook for driving an `omh-ralplan` run — context-package authoring (where quality is born), round dispatch, distillation, final review |\n| **omh-deep-interview** | Socratic requirements interview with coverage tracking |\n| **omh-ralph** | Verified execution: implement → verify → iterate until done |\n| **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 |\n| **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 |\n| **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](https://github.com/witt3rd/oh-my-hermes/issues/9) |\n| **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 |\n| **omh-autopilot** | Full pipeline composing all three skills end-to-end |\n\nComposition (recommended pipeline for unfamiliar domains):\n\n```\nomh-deep-research → omh-deep-interview → omh-ralplan → omh-ralph\n```\n\n(Or fold `omh-deep-research` in as Phase -1 of `omh-autopilot` when the\ndomain is unfamiliar; otherwise start at the interview.)\n\n## Install\n\n```bash\nhermes skills tap add witt3rd/oh-my-hermes\nhermes skills install omh-deep-research omh-ralplan omh-ralplan-driver omh-deep-interview omh-ralph omh-ralph-driver omh-ralph-task omh-autopilot\n```\n\nOr copy `skills/<name>/` to `~/.hermes/skills/omh/` manually.\n\nFor the optional plugin: install `plugins/omh/` to `~/.hermes/plugins/omh/`\n(requires Python 3.10+ and `pyyaml`).\n\nFor local development (live edits via symlinks), see [CONTRIBUTING.md](CONTRIBUTING.md).\n\n## Getting Started\n\n- **Need background on an unfamiliar domain?** → `omh-deep-research`\n- **Just need a plan?** → `omh-ralplan`\n- **Driving a ralplan run yourself?** → `omh-ralplan-driver` (load alongside `omh-ralplan`)\n- **Vague idea?** → `omh-deep-interview` then `omh-ralplan`\n- **Have a plan, need execution?** → `omh-ralph`\n- **Driving a ralph run yourself?** → `omh-ralph-driver` (load alongside `omh-ralph`)\n- **Grooming an issue backlog?** → `omh-triage` (load alongside `omh-triage-driver` if you're driving)\n- **End-to-end?** → `omh-autopilot`\n\nOMH self-seeds a `.omh/` directory in the project on first use (with the\nplugin installed) — including a README explaining the convention and a\n`.gitignore` pre-configured for selective sharing. To scaffold up-front\nwithout running a workflow, call `omh_state(action=\"init\")`.\n\n## Known Gaps\n\n- **wiki/fact_store/memory persistence** is not yet integrated for\n  research artifacts produced by `omh-deep-research`. The confirmed\n  report sentinel (`.omh/research/{slug}-report.md` with `status:\n  confirmed`) is the durable interface in v1; downstream skills consume\n  it directly. Persisting findings into `fact_store` or wiki is a\n  deferred Q2 item.\n- **Per-call subagent tool scoping** for the `omh-deep-research`\n  verifier may be unavailable depending on Hermes install; the\n  READ-ONLY contract is enforced by prose in `role-research-verifier.md`\n  in that case (A5).\n\n## Cost Envelope (omh-deep-research)\n\nA typical happy-path session is roughly **5-8 `delegate_task` calls**\n(3-5 researchers + 0-1 followup + 1 synthesist + 1 verifier). With one\nsynthesis retry, expect **up to ~10-12 calls**. The 3-strike retry cap\nbounds worst-case at ~14-16 calls before BLOCKED is surfaced.\n\n## Requirements\n\nHermes Agent v0.7.0+. The plugin additionally requires Python 3.10+ and\n`pyyaml`.\n\n## Documentation\n\n- [`docs/concepts.md`](docs/concepts.md) — How the four skills work\n- [`docs/plugin.md`](docs/plugin.md) — The v2 plugin (roles, hooks, tools)\n- [`docs/omh-delegate.md`](docs/omh-delegate.md) — Hardened delegation wrapper\n- [`docs/omc-comparison.md`](docs/omc-comparison.md) — Origins and design choices vs OMC\n- [`docs/hermes-constraints.md`](docs/hermes-constraints.md) — How OMH works around Hermes limits\n- [`docs/gaps.md`](docs/gaps.md) — What's not built yet\n- [`ROADMAP.md`](ROADMAP.md) — Versions and direction\n\n## License\n\nMIT\n",
  "bytes": 5205,
  "sha": "4807d8b803164c9d31d7bd8a705775ff5076d27535ff59508038c35fb7f3f164",
  "repo_slug": "witt3rd/oh-my-hermes",
  "fonte": "repo",
  "truncated": false,
  "api": "https://api.agentalog.com/api/listings/skl_witt3rd_oh_my_hermes_omh_ralplan_driver_db757fe8/readme"
}