Content
88%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.
A strong, dense conventions skill: nearly every line encodes project-specific knowledge Claude could not guess (source-gen serializer context, JsonDocument camelCase assertions, FixtureLoader compatibility pattern, one test per event shape), and the anti-patterns section directly counters the failure modes named in the intro. The only soft spots are the code-less later patterns and everything living inline in a single file.
Suggestions
Add a short executable snippet to Pattern 4 or 5 (e.g. a ToolCallBuilder feed/flush/assert or an Assert.Collection example) so all high-frequency patterns are copy-paste ready, not just Patterns 1–3.
Consider moving the per-project pattern details (e.g. Protobuf/Server/Formatting specifics) into references/ files (one per project) and keeping a one-line pointer per pattern in SKILL.md, which would tighten progressive disclosure.
State where FixtureLoader lives (project/namespace) so a new test file can resolve it without searching the repo.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every section carries non-obvious, project-specific knowledge: the table of the five test projects, naming conventions ("{TypeUnderTest}Test", "{Category}CompatibilityTest"), the rule to serialize via "AGUIJsonSerializerContext.Default.{Type}", and the rationale "This is what catches [JsonPropertyName] bugs". No basic-concept padding (e.g. no explanation of what xunit or JSON is); assumes Claude's competence throughout. | 5 / 5 |
Actionability | Patterns 1–3 are copy-paste-ready executable C# with concrete assertions ("Assert.Equal(\"RUN_STARTED\", doc.RootElement.GetProperty(\"type\")...)", "FixtureLoader.LoadFixture(\"run-events.json\")"), and the run command "dotnet test tests/AGUI.Abstractions.UnitTests/" is given. Patterns 4–6 (client, server, SSE formatter) are concrete at the API level ("Assert.Collection", "MediaType", "CanRead(contentType)") but prose-only without code snippets — a minor gap versus the level-5 anchor. | 4 / 5 |
Workflow Clarity | This is a single-task conventions skill (write a unit test for a type), and the single action is unambiguous: the project table maps a type to its test project, the Conventions section fixes file placement ("directly under the project root"), class naming, and fact/theory style, and "Run one project: dotnet test tests/AGUI.Abstractions.UnitTests/" gives the verification step. Under the simple-skill guidance, an unambiguous single action scores 5; the work is itself validation, so no destructive/batch cap applies. | 5 / 5 |
Progressive Disclosure | The body is well organized (project table, conventions, six labeled patterns, anti-patterns section) with clear section navigation and no nested references; the one external pointer ("sdks/dotnet/AGENTS.md") is clearly signaled. However, at ~120 lines with per-project detail (Patterns 3–6) all inlined in SKILL.md and no bundle files to offload to, it sits just below the level-5 'appropriately split' anchor — good structure with minor organization gaps. | 4 / 5 |
Total | 18 / 20 Passed |