Why a distribution
This page explains what kind of thing omh is — a distribution — and why that shape was chosen. Read it before judging omh as a framework, or before proposing something for its base set.
In short
- omh is a distribution, not a framework, an abstraction layer or a marketplace: most of what it does, something else already does, and its job is to choose and wire. → A distribution, not an abstraction layer
- An app store cannot pick winners; a distribution has to. omh’s product is subtraction, and its metric is decisions removed. → What omh is not
- No tool is both curated and harness-neutral. That is the gap omh is in: your curated setup, anywhere. → Competitive position
- The curation half is not earned yet. Today the verified, defensible half of omh is isolation. → The honest weak spot
- Being a distribution is a commitment: a shortlist that is earned, expires, and can be escaped. → What follows
A distribution, not an abstraction layer
Section titled “A distribution, not an abstraction layer”The single most useful thing to understand about omh: it is a distribution. Not a framework, not an abstraction layer, not a marketplace.
Judged as an abstraction layer, it looks thin — most of what it does, something else already does. That is the wrong axis. Debian didn’t write the kernel. Homebrew didn’t write the software. oh-my-zsh didn’t write zsh, and its genius was never the plugin catalog — it was that installing it gave you a good shell immediately.
“It already exists” is the precondition for a distribution, not a refutation of one. Every part below is somebody else’s work; omh’s job is the last column.
| Part | Who does it | omh’s job |
|---|---|---|
| Isolation | sbx, Docker |
choose one, wire it |
| Rules & skills portability | AGENTS.md, SKILL.md standards | place them correctly |
| Code graph | codebase-memory-mcp et al |
pick one, index on init |
| Model routing | LiteLLM, OpenRouter | one env var |
| Curation | Claude marketplace, community | subtract from it |
What omh is not
Section titled “What omh is not”Anthropic’s marketplace ships 200+ curated plugins into an ecosystem of 23,600+ skills and 12,700+ MCP servers.
That is an app store, and an app store is structurally incapable of being opinionated. The moment it picks winners, every excluded publisher becomes a business problem.
| App store | Distribution | |
|---|---|---|
| Optimizes for | catalog size | working defaults |
| Success metric | items listed | decisions removed |
| Can it pick winners? | no | yes — that is the product |
23,600 skills is not an opinion. It is the problem statement. And it grows every month, which means the case for a distribution strengthens over time rather than eroding.
The product is subtraction. A dozen MCP servers, not 12,700. The metric is decisions-to-productive, and the target is zero.
Corollary: “we support everything” is an anti-feature. Mechanisms may be broad; the UX must not expose that breadth as choice.
Competitive position
Section titled “Competitive position”| curated | isolated | harness-neutral | |
|---|---|---|---|
| Anthropic marketplace | ✓✓ | ✗ | ✗ |
Docker sbx |
✗ | ✓✓ | ✓ |
| Sculptor | ✗ | ✓✓ | ~ |
| Conductor, Nimbalyst | ✗ | ✓ | ~ |
| Plexus, agent-rules-sync | ✗ | ✗ | ✓ |
| omh | — | ✓ | ✓✓ |
No tool is both curated and harness-neutral. That is the gap.
It also sharpens the pitch. omh is not “a curated setup” — Anthropic will always beat us at that on their own harness. omh is your curated setup, anywhere. Isolation we don’t build; curation we inherit and port; provenance so it stays debuggable.
A sobering note: Vibe Kanban’s company shut down in April 2026. This category has already produced a casualty.
The honest weak spot
Section titled “The honest weak spot”The em-dash in that table is not a typo. omh’s curation column is currently unearned — the base set is two MCP servers, six hooks and three rules sections, justified by argument rather than measurement.
The name promises curation. The verified, defensible half of the product today is isolation.
No benchmark is coming to fix that, and pretending otherwise would be its own kind of dishonesty — see measure the cost, argue the benefit. What closes the gap is the ordinary work of a distribution: more entries that earn their place, each with stated criteria, a measured cost, and a one-line way to remove it.
Stating this here rather than discovering it in an issue thread is the point of writing it down.
What follows from being a distribution
Section titled “What follows from being a distribution”- The shortlist must be earned. Every base-set entry states what it costs (measured, in bytes and seconds), what it buys (argued, in a sentence), what was considered instead, and how to remove it. An entry that cannot fill in all four is taste pretending to be curation.
- The shortlist expires. A distribution’s real work is re-choosing quarterly
as the catalog churns, so the base set is versioned (
2026.08) and re-cutting it marks every carried-over measurement for re-affirmation. - Curation is a recurring commitment, which is the honest reason a solo distribution is hard. See risks.
- The opinion must be escapable. A distribution whose opinion cannot be
overridden is not trustworthy — hence
omh eject, and hence runtime backends staying plural even when one is clearly better.
- Trust
How the curation stays stated, visible and reversible — and why costs are measured but benefits argued. - The base set
What omh ships today, each entry with its cost and how to remove it. - Risks
What is still weak, the curation commitment among it.