Skip to content

rlaope/oh-my-hermes / skills/omh-routing

omh-routing

Verified

[omh] Choosing among OMH skills for a request: router guidance for using oh-my-hermes workflow skills inside Hermes Agent.

Oh My Hermes Router

Reasoning demand: light

Use this skill when the user mentions oh-my-hermes or a workflow keyword such as deep-interview, ultraperf, ralplan, loop, product-docs, research, research-department, source-finder, paper-learning, long-document-reading, data-analysis, command-operator, connector-operator, live-info-operator, external-connector-readiness, physical-device-readiness, content-operator, media-input-operator, feedback-triage, finance-analysis, people-ops, legal-compliance-review, support-operations, curriculum-design, localization-review, sales-development, product-brief, materials-package, img-summary, design-quality-gate, frontend, accessibility-audit, visual-qa, browser-operator, workspace-file-operator, automation-blueprint, harness-session-inventory, agent-debug, failure-signal-audit, instinct-ledger, skill-scout, skill-health, workflow-learning, codebase-onboarding, codegraph-refresh, codebase-uml, context-budget-review, run-efficiency, provider-profile-posture, decision-recall, security-safety-review, application-threat-model, live-incident-response, code-review, build-failure-triage, ultrawork, ultraqa, doctor.

Routing Contract

This is best-effort Hermes prompt guidance. It does not override Hermes core routing and it does not claim exact runtime parity with another agent framework.

Normal users should talk to Hermes Agent or invoke installed Hermes skills. Do not ask chat users to run omh; it is bootstrap and backend infrastructure.

Why This Exists

oh-my-hermes exists to keep Hermes chat routing conservative: it maps plain requests to the right workflow, explains evidence boundaries, and avoids making every keyword look like hidden implementation.

Do Not Use When

  • The user already invoked a more specific installed skill and its routing signals are unambiguous.
  • The message is ordinary chat, status acknowledgement, or a question that does not need workflow routing.
  • The wrapper wants to claim execution, review, CI, or merge evidence that no observed artifact provides.

Examples

Good example:

  • Prompt: Use OMH request-to-handoff for: safely add a feature to this repo.
  • Expected behavior: Classify the request, name the retained Hermes lane or prepared coding handoff, and expose the observed/prepared evidence boundary.
  • Why: The user asks for OMH-shaped routing without naming a narrow workflow, so the router should choose the safest next surface.

Bad example:

  • Prompt: omh
  • Expected behavior: Show the workflow picker or ask what the user wants to do next; do not infer a coding workflow.
  • Why: A bare product name is a picker or clarification signal, not implementation evidence.

Completion Checklist

  • The selected workflow, confidence reason, evidence boundary, and user-facing next action are named.
  • Low-confidence or conflicting signals return a picker or clarification instead of forced routing.
  • Catalog answers are rendered without shell approval when wrapper metadata is sufficient.

Recovery Notes

  • If a skill is not installed, keep the intent; hand back the install gap as unresolved.
  • If executor readiness is unknown, keep routing; hand back the check as not_observed.

OMH Awareness Primer (Compact)

OMH is Hermes-native workflow guidance, not a hidden executor or core patch. Hermes should retain routing, web/source research, deep interview, planning, status, and evidence narration. Coding-heavy work becomes an explicit prepared_not_observed handoff to the selected executor/runtime profile until observed.

Compact lane map:

  • Intent -> plan: deep-interview, ralplan, plan, dag-ulw, loop.
  • Research and company ops: research, source-finder, research-department, paper-learning, feedback-triage, strategy-brief, meeting-brief.
  • Retained knowledge: wiki.
  • Materials and visual summaries: design-quality-gate, frontend, accessibility-audit, visual-qa, materials-package, img-summary, report-package, deliverable-package.
  • Operations and evidence gates: workspace-audit, production-audit, verification-gate, agent-evaluation, rules-distill, agent-ops-review, harness-session-inventory, ops-observability-card, instinct-ledger, workflow-learning.
  • Coding handoff and review: idea-to-deploy, code-review, ultrawork, ultraqa.

OMH Orchestration Posture

Treat OMH as the operating layer above individual Hermes-native skills. For a workflow-shaped request, first frame the problem, success criteria, constraints, risks, and evidence needed; then select the smallest OMH workflow and harness that can coordinate the work. Hermes-native skills, tools, and subagents are capabilities used inside that OMH-selected workflow, not competing top-level owners.

  • On an unfamiliar or first-use pattern, briefly recommend the OMH-led route: explain that OMH can structure the problem, select the needed skills, and keep evidence boundaries clear.
  • After repeated accepted local patterns for the same user and workflow, continue OMH-led exploration, problem framing, skill composition, and prepared planning automatically. Keep the current workflow, next action, and prepared-versus-observed boundary visible.
  • Never let that autonomy bypass existing confirmation gates for destructive changes, credentials, external writes, deployment, executor dispatch, or starting a follow-on workflow engine (dag-ulw, ultrawork — including its coordinated-scope, single-owner-persistence, delivery-boundary, and durable-checkpoint capabilities — loop, ultraqa) from another skill's output: an accepted plan or clarified brief is planning evidence, not permission — recommend the engine that fits the work's shape and wait for the user's explicit go-ahead. Do not claim that a native skill, subagent, review, CI, or merge ran unless matching observation exists.
  • If a native Hermes capability is relevant, present it as an optional subordinate capability under the selected OMH workflow. OMH policy remains responsible for selecting and governing the workflow.

Priority Rules

  1. Exact or near-exact OMH maintenance commands (omh update, omh setup, omh doctor, omh uninstall, omh install, omh list, and Korean forms such as omh 업데이트해줘, omh 닥터 돌려줘, omh 삭제해줘, omh 셋업해줘) route as operator_maintenance_command. Run the requested command, report observed output, and avoid repo mutation unless asked. omh uninstall: --dry-run, then confirm.
  2. Explicit slash skill invocation wins when it is not one of those maintenance commands.
  3. Explicit workflow keywords route to the matching adapted skill when installed.
  4. Broad planning requests route to ralplan or plan before implementation.
  5. Persistence or finish-until-done requests route to ultrawork's single-owner-persistence capability only after scope is concrete.
  6. Unknown or conflicting signals stay in this router and ask one concise clarification question.

Direct Picker Aliases

If the user has only typed ./, /, ./o, or /om, show a command preview with exactly one top-level suggestion: omh. Selecting it should insert ./omh or /omh and then open the workflow picker.

For messenger-native setup, wrappers can call omh chat native-command --source discord, --source slack, or --source telegram. When plain-message autocomplete is not available, render the returned omh_command_fallback_card/v1 as an Open omh button/card before opening the picker.

If the user types ./omh, /omh, ./skills, or /skills without a task, show a compact workflow picker instead of creating a plan. Keep real skill names unchanged and keep chat_response.state.skill_picker.options as the flat-list fallback.

Choosing a skill is routing intent, not plan acceptance, dispatch, execution, or verification evidence. Do not make the user approve omh list just to see the catalog.

Install And CLI Boundary

Hermes-native install paths should converge on the same skill-visible state:

  • hermes skills tap add rlaope/oh-my-hermes, then hermes skills install rlaope/oh-my-hermes/skills/omh-routing --yes installs this tap-compatible skill pack directly when Hermes supports taps.
  • omh setup installs generated managed skills and registers their directory through skills.external_dirs when a local bootstrap or repair path is preferred.

Use compact human summaries for normal omh setup, omh doctor, omh update, omh uninstall, omh install, and omh list operator flows.

Wrapper Backend Summary

omh chat route, omh_interact, omh_recommend, omh coding delegate, omh memory ..., and omh hermes plan are adapter/backend surfaces, not normal chat UX. This is a deterministic wrapper-side decision layer; it does not patch Hermes core or require platform network access from omh.

When a wrapper prepares coding work, check executor_readiness/v1 for Codex, Claude Code, Hermes, or oh-my runtime profiles before first dispatch. A readiness probe is not dispatch, implementation, verification, review, CI, merge-readiness, or merge evidence.

Runtime Evidence

Record only what is observed. A task card, route, plan, coding_delegation.json, or prepared_coding_delegation run envelope proves preparation, not execution. Executor-choice, prompt-only, and runtime handoffs do not create lifecycle runtime runs.

Hermes Compatibility

  • Use Hermes-native tools, file operations, and subagent/delegation features when available.
  • Do not require runtime tools, role prompts, or overlays Hermes Agent does not expose.
  • Translate runtime-specific mechanisms to Hermes-native artifacts:
    • goal tools -> .omh/goals/ ledgers, goal status cards, or explicit checklists with named next actions,
    • question renderers -> one concise question in the current Hermes interface,
    • native subagents -> Hermes delegation when available, otherwise sequential lanes,
    • shell bridge commands -> optional bridge mode only.
  • Record observed delegation results when exposed. If unavailable, record not_available or not_observed.
  • Reply in the user's own words and the host's own voice: its SOUL.md persona owns reply language, tone, speech level, and sentence endings, progress updates included (where it sets no language, use the one the user wrote in), and OMH shapes structure and content only. When the requested output is delivered, put any open points inside the result and end with no question or offer; only when the person asked to choose or decide and that choice is still theirs, end with one specific question about it, not a list, a generic offer, or a footer; the shared rail's Reply Language And Host Voice and Turn Ending sections hold the full rule.

Progressive Disclosure References

Load these only when exact detail matters:

  • references/operator-maintenance.md for short omh maintenance command semantics.
  • references/catalog-index.md for the full-catalog shortlist (every skill name plus one-line description); shortlist there first, then confirm with omh recommend --json --limit 3 (authoritative next action and evidence boundary); never paste full catalog dumps into chat.
  • references/workflow-registry.md for representative workflow triggers and role registry; load the specific workflow skill for the full trigger list.
  • references/harness-registry.md for representative harnesses and priority.
  • references/wrapper-routing.md for backend/plugin/chat/coding delegation contracts.
  • references/coding-handoff-progress-reporting.md for active progress cadence, background executor watchdogs, PR head/merge verification, and memory/context collision pitfalls.
  • references/evidence-boundaries.md for prepared-vs-observed, target topology, memory, and compatibility rules.
  • references/structural-code-search.md for ast-grep structural code search patterns and the grep fallback.
  • references/human-voice.md before writing prose a person reads: AI-writing tells to remove, Korean habits, reply shape.
  • references/skill-common-rail.md for the shared rail: the full Reply Language And Host Voice and Turn Ending rules this body's reply line names.

Recovery

  • If exact route detail matters, load references/workflow-registry.md plus the specific workflow skill before answering.
  • If harness behavior matters, load references/harness-registry.md.
  • If wrapper/backend behavior matters, load references/wrapper-routing.md.
  • If delegated coding work is running or being reported, load references/coding-handoff-progress-reporting.md.
  • If maintenance command behavior matters, load references/operator-maintenance.md.
  • If evidence or target topology is disputed, load references/evidence-boundaries.md.
  • If the search target is a syntactic shape rather than a string, load references/structural-code-search.md.
  • If the right skill was not loaded, call skills_list or skill_view.
  • If a slash command exists, use the explicit slash skill such as /ulw-work.
  • If a skill name collides, keep the OMH-selected policy in control and present the Hermes-native skill only as an explicit recommendation; do not let a native candidate override routing.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall