Lists multiple concrete actions: applies a set of rules (CRDs separate, rendered-from-generator stays bundled, split by ownership/references/lifecycle/blast radius), routes user to the right import skill, produces a concrete Unit-slug plan. Also explicitly lists what it does NOT do with specific alternatives.

Clearly answers both 'what' (applies splitting rules and routes to the right import skill with a Unit-slug plan) and 'when' (explicit trigger phrases at the start). Also includes explicit 'do not use when' guidance with alternative skills, which strengthens the 'when' dimension further.

Excellent coverage of natural trigger phrases users would say: 'one Unit or many?', 'how should I split these resources?', 'should Deployment and Service be one Unit?', 'where do CRDs go?', 'Unit per resource or per bundle?', 'per-app or per-namespace?'. These are realistic user phrasings in the Kubernetes/ConfigHub domain.

Highly distinctive with explicit negative boundaries — clearly states not to load for authoring new YAML (use config-as-data), not for executing already-scoped imports (use import-from-* skills), and not when tools auto-split CRDs. This creates very clear boundaries against related skills.

The skill is fairly well-written and most content earns its place, but there's some redundancy — the CRD separation rationale is repeated across Rules, Anti-patterns, and Stop conditions. The anti-patterns section partially restates what the rules already cover. Some trimming is possible, but it's not egregiously verbose.

Highly actionable: provides concrete bash commands with `cub unit import` and `--where-resource` predicates, a complete worked example splitting a namespace into four Units with dry-run validation, specific slug conventions, and a clear decision flow. Copy-paste ready.

The decision flow is clearly sequenced (source → generator default vs. hand-rolled axes → slugs + predicates → hand off). Validation is explicit via `--dry-run` before committing. Stop conditions define when to halt. The preflight checklist ensures prerequisites are met before proceeding.

Well-structured with clear sections (rules, groupings table, anti-patterns, decision flow, preflight, tool boundary). References to companion skills and external docs are one-level deep and clearly signaled. Content is appropriately self-contained for a decision-helper skill while pointing to import skills for execution.