Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-architected skill body: a lean decision surface with concrete commands, sequenced workflows with a CI validation gate, and clean one-level progressive disclosure into two real reference files. The only weakness is repeated inline documentation URLs that inflate the token budget.
Suggestions
Define the three Epic doc URLs once (e.g. in a short 'Sources' note) and cite them by short label in the anti-patterns and limitations tables instead of pasting the full URL on every row.
In the 'Running' section, factor the shared -ExecCmds / -unattended / -nopause flags into a single canonical block so the worked example can reference rather than re-paste the full command line.
Consider trimming the eight-row anti-patterns table to the highest-value rows, or moving the doc-citation-heavy rows into the relevant reference file, to reduce inline repetition.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and free of concept-explanation fluff, but the same full dev.epicgames.com URL is repeated inline ~10 times across the anti-patterns and limitations tables where a single defined link would do — 'could be tightened' per the level-2 anchor, not 'every token earns its place'. | 2 / 3 |
Actionability | Provides fully executable guidance: a copy-paste UnrealEditor-Cmd.exe invocation, concrete EAutomationTestFlags enums, By::Id locators, the 'state': 'Fail' report key, and a step-by-step worked example — copy-paste ready. | 3 / 3 |
Workflow Clarity | The 'How to use' six-step sequence and the five-step worked example are explicitly ordered, and the CI 'Gate' step with 'treat any state: Fail as a failed build' supplies the validation checkpoint / feedback loop the rubric rewards. | 3 / 3 |
Progressive Disclosure | SKILL.md is explicitly the 'decision surface' with two real, one-level-deep references (authoring-macros-and-apis.md, running-and-reporting.md, both present in ./references/), each clearly signaled and consulted at a defined point in the workflow. | 3 / 3 |
Total | 11 / 12 Passed |