Standalone Test Repro

Standalone Runtime Test Repro

Generate a source-level console application for one managed test under src/tests/. Put it under artifacts/tmp/standalone-test-repro/ and make it runnable as:

<CORE_ROOT>/corerun[.exe] <repro-output>/StandaloneTestRepro.dll

The output is an isolated repro, not a launcher that references the original test assembly. Copy and adapt the test's source and required helpers so the repro remains understandable and editable.

Scope Gate

This skill supports managed CoreCLR tests that can meaningfully execute through corerun. Before extraction, inspect the project and generated wrapper when available.

Stop and explain the limitation instead of producing a misleading repro if the test fundamentally depends on any of these:

• A native executable as the primary test process
• NativeAOT, Mono, browser, WASI, Android, iOS, tvOS, or MacCatalyst hosting
• A profiler, debugger, COM server, custom host, or externally installed tool that cannot be represented by a console app plus local files
• A coordinated multi-process harness where invoking one managed method changes the behavior under test
• Network services, credentials, privileged machine configuration, or other unavailable external infrastructure

Local native libraries, data files, child processes, and environment variables are allowed when they can be copied into the repro directory and their use is part of the behavior being reproduced.

Workflow

1. Establish the Runtime Test Environment

Read:

• src/tests/README.md
• The nearest applicable README.md files between the resolved test directory and src/tests/
• .github/instructions/tests.instructions.md
• .github/instructions/csharp.instructions.md
• Any area-specific instruction file that applies to the resolved source

Use the build-and-test skill before invoking any build or test command.

Determine the intended OS, architecture, and configuration from the supplied path or CI context. Otherwise default to the current host, x64, and Release. Prefer an existing matching CORE_ROOT. If it does not exist, generate the matching Core_Root layout using the runtime-test instructions from the build-and-test skill.

Do not modify files under src/tests/.

2. Resolve the Test

Preserve the original identifier for the final report, then load exactly one resolution sub-skill:

• For a path ending in .cmd or .sh, read and follow ../standalone-test-repro-cmd/SKILL.md.
• For a fully qualified test method name, read and follow ../standalone-test-repro-method/SKILL.md.

If the input does not clearly match either form, use the ask_user tool. The sub-skill must resolve the exact source method, owning project, and effective test invocation before continuing here.

Capture an explicitly requested test scenario separately from the test identifier. After resolving the owning project, read and follow ../test-scenario-env/SKILL.md when a scenario was supplied. Pass it the canonical scenario name, owning .csproj, target OS, architecture, configuration, runtime flavor, and output path artifacts/tmp/standalone-test-repro/<sanitized-test-name>/.env.

3. Capture the Original Test Contract

Before editing the repro, record everything that can affect behavior:

• Source files and helper types used by the test
• Project properties including Optimize, AllowUnsafeBlocks, CheckForOverflowUnderflow, DefineConstants, LangVersion, PlatformTarget, and nullable context
• RuntimeHostConfigurationOption values
• Evaluated CLRTestEnvironmentVariable names and values from the owning .csproj, including applicable imported properties and conditions
• Target-specific CLRTestBatchEnvironmentVariable and CLRTestBashEnvironmentVariable items
• Evaluated CLRTestBatchPreCommands, CLRTestBashPreCommands, CLRTestBatchPostCommands, and CLRTestBashPostCommands
• The requested scenario and its evaluated environment from src/tests/Common/testenvironment.proj, when specified
• CLRTestExecutionArguments
• Expected exit code, which defaults to 100 for runtime tests
• Working directory, input files, native libraries, and other copied assets
• Architecture, OS, configuration, and runtime flavor guards

This list is a mandatory minimum, not an exhaustive allowlist. Inspect all evaluated project metadata and the generated wrapper for additional setup, launch, cleanup, or environment behavior before declaring the contract complete.

When a local wrapper exists, treat its effective command and environment as the ground truth. Project files are the fallback when no wrapper has been generated.

4. Create the Repro

Create:

artifacts/tmp/standalone-test-repro/<sanitized-test-name>/

Use a stable, filesystem-safe name derived from the type and method or wrapper stem. If that directory already exists, inspect it first. Replace only files owned by a previous run of this skill; never recursively delete an unresolved or broad path.

The directory must contain at least:

StandaloneTestRepro.csproj
Program.cs

If a scenario was requested or the owning project defines any effective common or target-specific environment items, also create:

.env

Additional .cs files and local assets are allowed when they improve clarity or are required for faithful behavior.

Use this project as the starting point, adding only properties required by the original test:

<Project Sdk="Microsoft.NET.Sdk">
  <PropertyGroup>
    <OutputType>Exe</OutputType>
    <TargetFramework>$(NetCoreAppCurrent)</TargetFramework>
    <AssemblyName>StandaloneTestRepro</AssemblyName>
    <UseAppHost>false</UseAppHost>
    <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
    <LangVersion>preview</LangVersion>
    <Nullable>enable</Nullable>
    <Optimize>true</Optimize>
  </PropertyGroup>
</Project>

Follow these extraction rules:

• When a scenario was requested, the test-scenario-env skill owns .env generation. Do not independently reconstruct or append scenario settings.
• Write each evaluated CLRTestEnvironmentVariable and applicable CLRTestBatchEnvironmentVariable or CLRTestBashEnvironmentVariable to .env as NAME=VALUE, one variable per line. Resolve MSBuild properties and item metadata for the selected OS, architecture, configuration, and runtime flavor; do not copy unevaluated expressions such as $(SomeProperty).
• Encode an empty value as NAME=''; a bare NAME= before another entry is not safe because CoreRun's parser skips the following newline while looking for a value. Use dotenv quoting or escaping when a value contains whitespace, #, quotes, backslashes, newlines, or variable expansion syntax. Follow the syntax supported by CoreRun's dotenv parser in src/coreclr/hosts/corerun/dotenv.cpp.
• Include only environment variables defined by the test project and its applicable imports. Do not copy the ambient shell environment or CI secrets into .env.
• Merge scenario variables first, selected-target Batch/Bash environment items second, and common CLRTestEnvironmentVariable items last, matching generated wrapper order. A later value replaces an earlier value with the same name.
• If a local generated wrapper and the evaluated project disagree, use the wrapper's effective values and investigate why before continuing.
• Copy the relevant test logic, not the whole merged assembly.
• Preserve the code shape that may trigger the failure. Do not simplify control flow, types, constants, inlining attributes, optimization settings, or concurrency merely because a smaller version appears equivalent.
• Include only transitively required helper types and source files.
• Remove xUnit and runtime test-framework dependencies where practical. Replace assertions with small local checks that throw an exception containing expected and actual values.
• Prefer source-level local replacements for simple TestLibrary helpers. Do not reference the original test project or its output assembly.
• Preserve required setup and cleanup with try/finally or using; do not silently omit fixtures or disposal.
• Preserve applicable batch or Bash pre/post commands by translating their effects into explicit repro setup, launch, and cleanup steps. Stop if those effects cannot be represented faithfully by a standalone console app plus local files.
• Invoke instance methods on a correctly initialized instance.
• Await Task, Task<T>, ValueTask, and ValueTask<T> results.
• For a theory with several rows, print the row before invoking it and stop on the first failure.
• Preserve unsafe code, P/Invoke declarations, generated code, and conditional compilation required by the failing path.
• Copy required local data and native binaries into the build output using project items such as Content with CopyToOutputDirectory. Keep their relative paths consistent with the original test.
• Do not download packages, binaries, or test data from issue links. Use only trusted repository sources and already-built repository artifacts.

The console entry point should return the original success code, normally 100, and allow failures to remain visible:

internal static class Program
{
    private static int Main()
    {
        RunTest();

        return 100;
    }
}

Use static async Task<int> Main() when asynchronous invocation is required. Do not catch exceptions merely to convert them into success-shaped output. A catch is acceptable only to add test-case context before rethrowing.

5. Build

Build with the repository's bootstrapped SDK from the repository root. Use a dedicated output directory inside the repro:

.\dotnet.cmd build artifacts\tmp\standalone-test-repro\<name>\StandaloneTestRepro.csproj -c Release -o artifacts\tmp\standalone-test-repro\<name>\out

On Unix, use ./dotnet.sh and Unix path separators.

If the original test requires a different optimization or configuration, use that configuration consistently and explain it in the final report. Fix all compile errors in the repro; do not work around them by referencing the original test assembly.

6. Run with CoreRun

Run from the repro output directory so relative paths behave consistently. Apply the captured environment variables, host configuration options, and test arguments.

Windows:

Push-Location artifacts\tmp\standalone-test-repro\<name>\out
& "$env:CORE_ROOT\corerun.exe" -e ..\.env .\StandaloneTestRepro.dll
$exitCode = $LASTEXITCODE
Pop-Location

Unix:

(
  cd artifacts/tmp/standalone-test-repro/<name>/out
  "$CORE_ROOT/corerun" -e ../.env ./StandaloneTestRepro.dll
)
exit_code=$?

Include -e ../.env only when the file was generated. If RuntimeHostConfigurationOption items were present, translate them to corerun -p Name=Value arguments before -e. Pass test arguments after the managed DLL.

Success normally exits 100; a CI failure repro may intentionally crash or return another value. Record the actual exit code and meaningful output. Do not claim the CI failure reproduced unless the observed behavior matches it.

7. Validate Fidelity

Before completing:

1. Compare the repro invocation with the original wrapper or project contract.
2. Confirm the repro does not reference the original test assembly.
3. If .env was generated, compare every entry with the effective scenario, target-specific environment items, and common CLRTestEnvironmentVariable items, then confirm the corerun command uses it.
4. Confirm all required local files exist in out.
5. Run the repro with the selected corerun.
6. If practical, run the original test with the same runtime and environment and compare the relevant behavior.
7. Inspect git status and ensure only the new skill or explicitly requested repository changes are tracked; repro files under artifacts/ should remain untracked or ignored.

If the standalone app builds and runs but does not reproduce the reported failure, keep the app and state that result plainly. Include any configuration from CI that was unavailable locally.

Final Response

Lead with the result and include:

• The resolved source method and owning project
• The repro directory and built DLL
• The exact corerun command, including required environment variables and -p options
• The generated .env path and variable names, without unnecessarily echoing sensitive values
• The canonical scenario name, when one was requested
• The observed exit code and whether it matched the reported failure
• Any intentionally retained local native or data dependencies

Do not paste the full generated source unless the user asks for it.