Conciseness

The body is dense with genuinely non-obvious, environment-specific facts (AllToolsets vs. Unreal MCP plugin split, serial game-thread execution, tool-search mode, label vs. name, lux/Kelvin calibration) and never explains concepts Claude already knows. It is not score 5 because a few rules are stated redundantly across sections (the one-call-at-a-time rule appears in Operating Loop #2, #3, and again in Pitfalls; "save often" advice repeats) and a few sentences could be trimmed. It is not score 3 because there is no padded or unnecessary explanation block.

Actionability

Highly concrete: exact menu paths ("Edit > Plugins", "Editor Preferences > General > Model Context Protocol"), copy-paste commands (`hermes mcp install unreal-engine`, `ModelContextProtocol.StartServer`, `ModelContextProtocol.RefreshTools`), exact tool names (`mcp_unreal_engine_list_toolsets`), config keys and paths (`mcp_servers.unreal-engine.timeout` in `~/.hermes/config.yaml`). It is not score 5 because no inline example of a complete `call_tool` invocation with an argument payload is shown — that shape is deferred to `references/recipes.md`. It is not score 3 because everything given is directly executable, not pseudocode or high-level hints.

Workflow Clarity

The 7-step Operating Loop and the 5-step "From Plain English to a Scene" plan are explicitly sequenced, with validation woven in ("Read every result", read-back after property writes, re-query + screenshot at each milestone) plus a dedicated Verification Checklist and concrete error-recovery branches (missing tools = wrong start order; hung call = modal dialog; timeout = raise config timeout). This matches the anchor-5 example's validate → fix → retry structure with a checklist for the complex process; batch/destructive operations are explicitly guarded by save-before-and-after rules.

Progressive Disclosure

SKILL.md is a genuine overview (~240 lines of rules and loop) with a Reference Files table describing each of the five reference files, plus targeted inline pointers at the exact point of need ("references/scene-craft.md has the numbers", "references/advanced-workflows.md" for batching). All five referenced files exist, are one level deep, and their contents match the table's descriptions. It is not score 4 because there is no organization gap — navigation is easy and every split is appropriate.

Specificity

"Automate Unreal Engine editor scenes, actors, and renders." names the domain and the objects worked on (scenes, actors, renders) but hangs them all on a single generic verb ("Automate"), so it lists 1-2 concrete actions rather than several specific ones. It is not score 4 because no distinct operations (spawn/move actors, set lighting, import assets, capture screenshots) are enumerated; it is not score 2 because the domain is named and the objects are concrete, not minimal.

Completeness

The "what" is clear (automate editor scenes, actors, and renders) but there is no "Use when..." clause or any equivalent trigger guidance, which caps completeness at 3 per the rubric. It is not score 4 because "when" is entirely absent rather than merely under-specified; it is not score 2 because the "what" is concrete and specific, not vague.

Trigger Term Quality

Core natural terms are present ("Unreal Engine", "scenes", "actors", "renders"), but common variations and synonyms users actually say are missing: "UE5"/"UE", "level", "lighting", "cinematic", "screenshot". It is not score 4 because the keyword set stops at the domain nouns without the synonym spread of the anchor-4 example; it is not score 2 because the terms given are natural and relevant, not generic.

Distinctiveness Conflict Risk

"Unreal Engine editor" is a clear niche with distinct triggers; no other common skill would compete for "Automate Unreal Engine editor scenes, actors, and renders." It is not score 4 because there is no meaningful overlap risk with closely related skills — the domain naming is unambiguous.