Domain Loadout Router

OMK routes incoming tasks to a domain capability profile ("inherited document") before dispatch. Each profile is a curated bundle of skills, MCP servers, hooks, a tool gate, an authority, and a detailed English routing prompt — all selected from the live capability inventory. The router is deterministic, I/O-free, and explainable.

Auto-generated from src/core/domain-loadouts.ts + src/core/domain-router.ts. Regenerate with node --import tsx scripts/gen-domain-docs.mjs.

How routing works

1. Signal extraction. The task text (plus optional path hints and upstream tags) is lowercased and scored against every domain's triggers.
2. Weighted multi-signal scoring.
   • keyword — literal phrase occurrences (counted, capped at 3) × weight.
   • regex — intent cluster tested once against the task text × weight.
   • extension — file suffix on any path hint × weight.
   • path — path fragment contained in any path hint × weight.
3. Ranking. Domains are sorted best-first; ties break by registry order (deterministic).
4. Confidence.
   • top score ≥ 8 → confident
   • 4 ≤ top score < 8 → tentative
   • top score < 4 (or zero signals) → fallback to general
5. Ambiguity. When the runner-up is within 2 of a tentative leader, the result is flagged ambiguous (the leader still wins; the caller can ask for clarification).

Thresholds: STRONG_THRESHOLD = 8, WEAK_THRESHOLD = 4, AMBIGUITY_MARGIN = 2.

Domains (13 + 1 fallback)

• frontend-ui — Frontend & UI
• visual-qa — Visual QA & Website Cloning
• korean-document — Korean Document (HWP/HWPX)
• backend-api — Backend & API
• data-science — Data Science & Analysis
• security-audit — Security Audit
• devops-infra — DevOps & Infrastructure
• research — Research & Investigation
• mobile — Mobile (iOS / Android / KMP)
• docs-writing — Docs & Technical Writing
• qa-testing — QA & Testing
• grok-harness — Grok xAI Harness
• ai-agent-ops — AI Agent Engineering & Ops
• general — General (fallback)

Worked examples

task	path hints	routed to	confidence	reason
build a responsive login form with tailwind	src/app/page.tsx	frontend-ui	confident	Frontend & UI selected (20)
scan for xss and sql injection vulnerabilities	—	security-audit	confident	Security Audit selected (19)
do a literature review on RLHF, cite arxiv	—	research	confident	Research & Investigation selected (17)
write a dockerfile and deploy to vercel	Dockerfile	devops-infra	confident	DevOps & Infrastructure selected (19)
fix the failing playwright e2e tests	tests/login.test.ts	qa-testing	confident	QA & Testing selected (28)
add a postgres migration for the users table	—	backend-api	confident	Backend & API selected (11)
train a classifier on the dataset, plot results	notebooks/model.ipynb	data-science	confident	Data Science & Analysis selected (22)
hello there	—	general	fallback	no domain signals detected

Composition with role loadouts

A domain profile is a LoadoutProfile, so it composes with the existing role-based system (BUILTIN_LOADOUTS: inspect / plan / code / test / review / security / package-maintainer). The domain gates which skills/MCP/hooks are active; the role sets authority/tools/commands. Use domainLoadoutProfiles() to get plain profiles consumable by applyLoadoutProfile().

API

import { routeDomain } from "./core/domain-router.ts";
const result = routeDomain({ task: "build a login form", paths: ["page.tsx"] });
// result.primary   -> DomainProfile
// result.confidence -> "confident" | "tentative" | "fallback"
// result.scores    -> ranked DomainScore[] with matchedSignals
// result.ambiguous -> boolean

Adding a domain

1. Add a new entry to DOMAIN_PROFILES in src/core/domain-loadouts.ts (id, label, authority, tools, curated skills/mcp/hooks, triggers, routingPrompt).
2. The router and docs pick it up automatically — no other code changes.
3. Run npm run check (biome + tsgo + tests) and regenerate docs.