Oh My Hermes
Hermes-native orchestration

Oh My Hermes

A stronger operating layer for the Hermes you already use.

OMH turns plain requests into a clear capability, a reviewed next step, and an honest evidence boundary. It strengthens planning, research, creation, coding handoff, operations, and project memory without replacing Hermes or hiding who did the work.

install
curl -fsSL https://raw.githubusercontent.com/rlaope/oh-my-hermes/main/install.sh | sh omh setup
Local contracts Executor neutral Prepared is not observed
Oh My Hermes illustrated brand artwork
OMH request-to-handoff ready
Hermes message I want to safely add a feature to this repo.
route
request-to-handoff
family
delegate coding and ship
owner
selected executor
next
present safe plan
readyRequest classified
preparedHandoff contract
not observedRuntime evidence

Prepared is not observed. Execution, review, CI, and merge need matching runtime evidence.

82installable workflow skills
6capability families
8locally routed languages
0network calls in core routing

Capability system

Start with the job, not the command.

OMH groups a large skill catalog into six human-readable families. Hermes can recommend the smallest useful lane while keeping every trigger, quality bar, provider requirement, and claim boundary inspectable.

Route with context

Normal language in. A bounded first move out.

English, Korean, Japanese, Chinese, Spanish, French, German, and Hindi operator requests can be classified locally. OMH returns a family, skill, owner, next action, and what still is not evidence.

  • Deep interview, planning, durable goals, and research
  • Media, data, web, paper, and best-practice routes
  • Guarded intent matching to reduce common-name overroutes
route previewlocal
requestReview recent agent memory papersEN
familyLearn and gathermatched
skillweb-researchsource-backed
nextprepare research briefnot run

Create with quality gates

Files, visuals, reports, and interfaces get the right production lane.

OMH distinguishes source analysis from generation, and a prepared visual or document brief from an observed artifact. Frontend, accessibility, PDF, slide, report, and image requests carry format-specific review expectations.

  • Frontend publishing plus visual and accessibility QA
  • Image, report, deck, document, and deliverable packages
  • External image, video, browser, and connector providers stay explicit
artifact contractprepared
sourcerelease notesreviewed
formatvisual summary cardplanned
providerexternal image providerrequired
verificationrendered artifact inspectionnot observed

Coding handoff

Use the coding environment you trust, with more context.

Codex, Claude Code, Hermes runtime, and generic executors are first-class owners. OMH prepares scope, non-goals, acceptance criteria, worktree guidance, available local skills, review expectations, and evidence gates before dispatch is claimed.

  • Executor readiness and skill-aware prompt preparation
  • Dynamic typed workflows across models, tools, wrappers, and agents
  • Builder, verifier, reviewer, docs, and PR responsibility lanes
coding_runtime_handoff/v1selected owner
scopeaccepted plan
workspaceworktree recommended
executorCodex or Claude Code
proofawait runtime observation

A prepared handoff can improve the run. It cannot prove that the run started.

Operate, learn, remember

Turn work into inspectable operating context.

OMH can prepare service-quality analysis from an external metric provider, review project-local memory, connect a wiki provider, inspect sessions, and turn weak routes into learning candidates. Provider access and outcome improvement remain separate observed claims.

  • Service quality, release, reliability, and incident review
  • Reviewed project memory and external knowledge-provider boundaries
  • Workflow learning, missed-route regression, and session recovery
provider-neutral contextcontract
metricsexternal_metric_provideradapter-ready
knowledgeexternal knowledge providerreview first
memory.omh/memory approved recordslocal
learningtrace, eval, patch proposalhuman review

Capability impact

Measure each claim at its own boundary.

OMH reports routing precision separately from guidance depth, host availability, external providers, artifact verification, and comparative outcomes. It does not hide those different claims in one score.

capability impactagent / maintainer evaluation
route selectiondeterministic in-repo corporachecked
guidance depthskill quality bars and handoff policypartial
host + providerrequires loaded runtime or provider evidenceunproven
outcome qualityrequires blinded paired evaluationnot run
No aggregate capability score.

Accuracy, availability, verification, and comparative quality are different questions.

Read the impact contract

Go deeper

Reference pages when the lane needs detail.

Start with the docs map, then open a focused workflow or architecture guide.

Install

Three commands for people. The rest belongs to the agent layer.

Set up OMH, keep it current, and check its health. After that, ask Hermes for the work in normal language.

# Install OMH once curl -fsSL https://raw.githubusercontent.com/rlaope/oh-my-hermes/main/install.sh | sh # Human OMH commands omh setup omh update omh doctor # Optional Hermes-native skill tap hermes skills tap add rlaope/oh-my-hermes hermes skills install rlaope/oh-my-hermes/skills/oh-my-hermes --yes