AgentSkillsCN

dotnet-testing

AvnDataGenie .NET解决方案的测试模式与规范

SKILL.md
--- frontmatter
name: "dotnet-testing"
description: "Testing patterns and conventions for the AvnDataGenie .NET solution"
domain: "testing"
confidence: "high"
source: "huntress"

Context

The AvnDataGenie solution uses xUnit on .NET 10. The test project is src/AvnDataGenie.Tests/. Tests run with dotnet test src/AvnDataGenie.Tests/AvnDataGenie.Tests.csproj.

Patterns

InternalsVisibleTo

AvnDataGenie.csproj has <InternalsVisibleTo Include="AvnDataGenie.Tests" />. Use internal visibility (not public) for methods that need testing but shouldn't be part of the public API.

Test File Organization

One test class per production class or logical unit:

  • •SqlPromptBuilderTests.cs — tests for SqlPromptBuilder.CreateSystemPromptFromJson()
  • •CleanAndFormatSqlTests.cs — tests for Generator.CleanAndFormatSql()
  • •FormatSqlTests.cs — tests for Generator.FormatSql() and Generator.FormatSelectColumns()

Test Data Helpers

Use private static methods to generate test JSON inputs (e.g., MinimalSchemaJson(), RichConfigJson()). Keep test data as close to real shapes as possible — the JSON must match the deserialization models in SqlPromptBuilder.

Pure Functions First

Prioritize testing pure/static methods that have no external dependencies. In this codebase:

  • •SqlPromptBuilder.CreateSystemPromptFromJson() — public static, takes JSON strings
  • •Generator.CleanAndFormatSql() — internal static, sanitizes LLM output
  • •Generator.FormatSql() — internal static, formats SQL
  • •Generator.FormatSelectColumns() — internal static, formats SELECT columns

SchemaGenerator Requires Integration Tests

SchemaGenerator.Generator.GenerateDatabaseSchemaAsync() requires a live SQL Server connection. Unit tests cannot cover this without a database fixture.

Anti-Patterns

  • •Don't mock what you can test directly — prefer calling pure functions over mocking IChatClient.
  • •Don't test formatting by exact string match — use Assert.Contains for structural assertions since the formatting regexes produce complex output.
  • •Don't skip edge cases — LLM output is unpredictable. Test markdown fences, preambles, trailing commentary, comments, and missing semicolons.