Conciseness

The body is dense with tables, a decision tree, and compact code, with no padding explaining concepts Claude already knows. Not 5 because the ASCII architecture diagram and the State Access / Providing Scopes sections restate detail that lives in ./references/architecture.md and sibling skills, and time-sensitive version specifics ("0.15.x", "AI SDK v7") sit in prose rather than a versioning or deprecation section.

Actionability

The Quick Start is copy-paste-ready TSX, the "Pick a Runtime" tree names the exact hook and package per backend, and commands like "npx assistant-ui@latest create" and "npx assistant-ui@latest add thread" are directly executable. Not 4: the examples cover the common cases completely rather than with minor gaps.

Workflow Clarity

The skill's single task (routing) is unambiguous: the When to Use table, the runtime decision tree, and the per-skill Related Skills list form a clear decision sequence, and "Always consult assistant-ui.com/llms.txt for the latest API" acts as an upfront checkpoint. No destructive or batch operations exist, so no validation cap applies; the simple-skill exception holds.

Progressive Disclosure

The body is a clear overview with a References section that signals the two one-level-deep files with per-file descriptions ("layers, the aui client, data flow, message model"; "every published package and when to install it"), both verified to exist, and detail is appropriately split out of the overview. Not 4: navigation and content placement are clean with no organization gaps.

Specificity

Quotes like "choosing packages, picking a runtime, or understanding the layers (elements, primitives, the aui client with AuiConfig and AuiProvider, the runtime, adapters) and the message model" name several concrete capabilities backed by a specific package/adapter/hook inventory. Not 5 because the actions are router-style and "understanding the layers" is mildly abstract rather than a fully concrete capability list.

Completeness

Explicitly answers what ("Overview and router for assistant-ui, the React library for building AI chat interfaces from composable primitives and a styled elements catalog") and when ("Use for high-level, cross-cutting, or architecture questions: choosing packages, picking a runtime...") with concrete trigger phrases, plus negative routing guidance for sibling skills. Clearly exceeds the 4 anchor, whose 'when' is less explicit.

Trigger Term Quality

Natural phrases users would say appear throughout: "building AI chat interfaces", "chat", "runtime", "elements", "primitives", plus ecosystem names (LangGraph, LangChain, Google ADK, AG-UI) that appear in real questions. Not 5 because a few common variants (chatbot, copilot, conversation UI) are absent.

Distinctiveness Conflict Risk

The niche is clear and "For a specific area route to a focused sibling instead: setup, elements, primitives, runtime, tools..." explicitly de-conflicts with the 17 sibling skills. Not 5 because, as a broad overview, it still carries minor overlap risk when a focused sibling would serve better.