Content
96%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a tightly written orchestrator skill: it gates requests on engine version and feature support, routes to exact documentation URLs via a verifiable convention, and encodes genuinely curated knowledge (docs coverage caveats, the source-of-truth ladder, undocumented widget layer) without an ounce of padding. The single material defect is that all five referenced bundle files are missing from the bundle, so the otherwise-excellent progressive-disclosure structure cannot be exercised as shipped.
Suggestions
Ship the five referenced bundle files (RULES.md, setup.md, widgets.md, platforms.md, docs-map.md) alongside SKILL.md — the body's entire routing layer depends on them, and every link is currently broken.
If any referenced file is intentionally omitted, state in SKILL.md which references are external versus bundled so the agent knows when a fetch will fail.
Consider moving the per-page coverage statistics ("26 of its 61 pages") into docs-map.md so the count that will rot fastest lives with the data it describes.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence: no explanations of what Unreal, UMG, or packaging are; every section carries curated non-obvious knowledge (the 26-of-61-pages coverage caveat, the ApiKey/BeginPlay ordering trap, the one-archive-per-engine-version silent-mismatch warning). Legacy engine versions (4.27/5.0/5.1, v1.3.0 Dec 2022) are properly quarantined in the gate table rather than scattered, so the time-sensitive version information does not pollute the main path. Matches the level-5 anchor "every token earns its place"; there is no over-explanation to trim that would justify 4. | 5 / 5 |
Actionability | Guidance is fully executable: a copy-paste bash one-liner for engine-version detection ("find . -maxdepth 2 -name \"*.uproject\" -exec sh -c 'echo == $1; grep -m1 EngineAssociation $1' ..."), an exact URL-rewriting convention with a worked example ("take the page URL, drop the trailing /, add .md"), concrete release-asset names (StreamChat-5.7.zip, StreamChat-5.8.zip), specific header paths (Plugins/StreamChat/Source/StreamChat*/Public/**.h), and named functions with signature comparisons (BanUserBP vs BanUser TOptional variants). Matches level 5 "copy-paste ready code or commands; specific examples cover the common cases". | 5 / 5 |
Workflow Clarity | The multi-step process is explicitly sequenced — Step 0 gates (engine version, feature support, C++ vs Blueprint) resolved "before any code", then Step 1 request classification, then Step 2 docs lookup — with validation checkpoints (the three gates) and error-recovery feedback loops (source-of-truth ladder with "stop at the first rung", the live-index fallback when the feature is not in docs-map.md, the no-Unreal-tab drop to rung 2, and the "at most 3 fetches" cap). This matches the level-5 anchor; the workflow is read/lookup orchestration, not destructive or batch, so the validation-cap rule does not apply. | 5 / 5 |
Progressive Disclosure | The structure itself is textbook level 5: a concise orchestrator overview, one-level-deep references, a "What this skill carries" table describing each file's role, and task-typed routing (setup.md for integrate, widgets.md for UI, platforms.md for packaging). However, none of the five referenced bundle files (RULES.md, setup.md, widgets.md, platforms.md, docs-map.md) actually exist in the bundle — no references/, scripts/, or assets/ directories are present — so the navigation the body promises cannot be verified and every link is broken as shipped. This lands at 4 ("good structure; minor organization gaps") rather than 5, and not 3 because the signaling and routing within SKILL.md itself are clear and well-organized. | 4 / 5 |
Total | 19 / 20 Passed |