AgentSkillsCN

testrunner

在 .NET 项目中处理易失效的测试、崩溃测试、挂起测试、内存不足测试、栈溢出测试、测试隔离，以及测试套件稳定性问题。适用于 dotnet test 挂起、崩溃、产生 OOM 或栈溢出错误，或当用户拥有不稳定/易失效的测试套件时使用。切勿用于常规测试运行——仅在标准 dotnet test 无法满足需求时使用。

SKILL.md
--- frontmatter
name: testrunner
description: Handle flaky tests, crashing tests, hanging tests, out of memory tests, stack overflow tests, test isolation, test suite stability issues in .NET projects. Use when dotnet test hangs, crashes, produces OOM or stack overflow errors, or when the user has an unstable/flaky test suite. Do NOT use for normal test runs — only when standard dotnet test is insufficient.
argument-hint: "[filter pattern or path]"

Prerequisites

This tool requires .NET 10+ SDK (for dnx support).

Check if dnx is available:

code
dnx --help

If dnx is not available, the user needs .NET 10 SDK or later. dnx ships with the SDK — it does not need separate installation.

About Asynkron.TestRunner

Asynkron.TestRunner is an alternative .NET test runner that wraps dotnet test with added resilience:

  • •Hang detection — auto-detects stuck tests with per-test timeouts (default 20s)
  • •Test isolation — automatically isolates hanging/crashing tests by splitting the test tree into branches and running them separately
  • •History tracking — maintains pass/fail history per project, detects regressions across runs
  • •Trend visualization — generates bar charts showing test health over time

This is NOT a replacement for dotnet test. Use it only when standard tooling fails — flaky suites, hanging tests, crashes, OOM, stack overflow, or when you need isolation and regression tracking.

Running via dnx (no install needed)

dnx runs .NET tools from source without installing them, similar to npx:

code
dnx Asynkron.TestRunner [arguments]

On first run, dnx will prompt to download the package. Use --yes to skip the prompt in CI.

Common Usage

Run all tests with hang detection:

code
dnx Asynkron.TestRunner

Filter by class or namespace:

code
dnx Asynkron.TestRunner "MyTestClass"
dnx Asynkron.TestRunner "MyNamespace.Integration"

Custom dotnet test command:

code
dnx Asynkron.TestRunner -- dotnet test ./tests/MyProject

List tests without running:

code
dnx Asynkron.TestRunner list

View test history and trends:

code
dnx Asynkron.TestRunner stats

Compare last two runs for regressions:

code
dnx Asynkron.TestRunner regressions

Manual isolation (find the culprit test):

code
dnx Asynkron.TestRunner isolate

Clear all history:

code
dnx Asynkron.TestRunner clear

Timeout & Isolation Options

FlagDefaultPurpose
-t, --timeout <seconds>20 (run), 30 (isolate)Per-test timeout
--timeout 0—Disable hang detection
-p, --parallel [N]—Run N test batches concurrently
--parallel (no value)—Auto-detect and use CPU core count

How Isolation Works

When a hang is detected:

  1. •The test tree is split into branches (max 100 leaf tests per branch)
  2. •Branches are run separately in parallel
  3. •Hanging branches are drilled down sequentially
  4. •The culprit test is identified and reported

This catches tests that hang, crash the process, cause OOM, or stack overflow — scenarios where dotnet test simply dies or never returns.

History & Regression Tracking

History is stored in .testrunner/ directory, indexed by:

  • •Project (git repo root hash or current directory)
  • •Command signature (test command + filters)

Different filters and repos maintain independent histories for accurate regression detection.

When to Use This

Use Asynkron.TestRunner instead of dotnet test when:

  • •Tests hang or never complete
  • •The test process crashes (OOM, stack overflow, access violation)
  • •Tests are flaky and you need regression tracking across runs
  • •You need to isolate which specific test is causing failures
  • •You want trend visualization of test health over time

For normal, healthy test suites — just use dotnet test.

Guidelines

  • •Always try dotnet test first — only reach for testrunner when it fails
  • •If tests hang, start with the default 20s timeout before adjusting
  • •Use isolate to pinpoint the exact test causing crashes or hangs
  • •Use stats and regressions to track flaky test patterns over time
  • •Use --parallel for faster isolation on large test suites