test(worker): add opt-in real-claude smoke test

Spawns the actual claude binary and asserts exit code 0, a session id,
non-empty result, and output tokens > 0 (plan-verification §1.0 step 3).
Inert unless CLAUDE_AUTHENTICATED=1, since it needs an authenticated CLI.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
mika kuns
2026-06-04 11:36:51 +02:00
parent 3756b81817
commit 773811d060
2 changed files with 63 additions and 3 deletions

View File

@@ -148,9 +148,9 @@ Voraussetzung: Gitea-Release unter `git.kuns.dev/releases/ClaudeDo` mit `ClaudeD
- `Hub/PlanningHubTests.cs`, `AgentSettingsHubTests.cs` decken Hub-Methoden via Fakes ab.
- **Optional:** echter Roundtrip mit `WebApplicationFactory<Program>` + `HubConnectionBuilder` — niedriger Mehrwert.
### 5.3 Smoke-Test gegen echten `claude`
- `tests/.../Runner/ClaudeProcessSmokeTest.cs` existiert nicht. Real-CLI-Test als `[Fact(Skip=...)]`, nur lokal bei `CLAUDE_AUTHENTICATED=1`.
- **Aufwand:** klein.
### 5.3 Smoke-Test gegen echten `claude`
- `tests/ClaudeDo.Worker.Tests/Runner/ClaudeProcessSmokeTest.cs` spawnt das echte `claude`-Binary und prüft `ExitCode==0`, `SessionId`, non-empty `ResultMarkdown`, `TokensOut > 0` (deckt §1.0 Step 3 ab).
- Inert ohne Opt-in (`CLAUDE_AUTHENTICATED=1`), da xUnit 2.5 kein dynamisches `Assert.Skip` hat (gleiche Konvention wie die `GitAvailable`-Guards). Lokal: `CLAUDE_AUTHENTICATED=1 dotnet test --filter FullyQualifiedName~ClaudeProcessSmokeTest`; Binary via `CLAUDEDO_CLAUDE_BIN` überschreibbar.
### 5.4 ExternalMcpService-Tests ✅
- Service exponiert **18 Tools** (war 11): `ListTaskLists`, `ListTasks`, `GetTask`, `AddTask`, `UpdateTask`, `UpdateTaskStatus`, `ReviewTask`, `RunTaskNow`, `CancelTask`, `DeleteTask`, `GetTaskStatusValues`, `GetTaskWorktree`, `GetTaskDiff`, `MergeTask`, `ListWorktrees`, `CleanupTaskWorktree`, `GetDailyPrepCandidates`, `SetMyDay`.

View File

@@ -0,0 +1,60 @@
using ClaudeDo.Worker.Config;
using ClaudeDo.Worker.Runner;
using Microsoft.Extensions.Logging.Abstractions;
namespace ClaudeDo.Worker.Tests.Runner;
// Real-CLI smoke test (plan-verification §1.0 step 3). Spawns the actual `claude`
// binary and needs an authenticated session, so it is inert unless the developer
// opts in with CLAUDE_AUTHENTICATED=1. Run locally with:
// CLAUDE_AUTHENTICATED=1 dotnet test tests/ClaudeDo.Worker.Tests \
// --filter FullyQualifiedName~ClaudeProcessSmokeTest
// Override the binary with CLAUDEDO_CLAUDE_BIN if `claude` is not on PATH.
public class ClaudeProcessSmokeTest
{
private static bool Enabled =>
Environment.GetEnvironmentVariable("CLAUDE_AUTHENTICATED") == "1";
[Fact]
public async Task RunAsync_PingPrompt_ReturnsResultSessionAndTokens()
{
if (!Enabled)
{
Assert.True(true, "CLAUDE_AUTHENTICATED != 1 — skipping real-CLI smoke test");
return;
}
var cfg = new WorkerConfig
{
ClaudeBin = Environment.GetEnvironmentVariable("CLAUDEDO_CLAUDE_BIN") ?? "claude",
};
var sut = new ClaudeProcess(cfg, NullLogger<ClaudeProcess>.Instance);
var args = new ClaudeArgsBuilder().Build(new ClaudeRunConfig(
Model: null, SystemPrompt: null, AgentPath: null, ResumeSessionId: null,
MaxTurns: 3, PermissionMode: "auto"));
var workDir = Path.Combine(Path.GetTempPath(), $"claudedo_smoke_{Guid.NewGuid():N}");
Directory.CreateDirectory(workDir);
var lines = new List<string>();
try
{
using var cts = new CancellationTokenSource(TimeSpan.FromMinutes(2));
var result = await sut.RunAsync(
args,
"Reply with exactly the word: pong",
workDir,
line => { lock (lines) { lines.Add(line); } return Task.CompletedTask; },
cts.Token);
Assert.Equal(0, result.ExitCode);
Assert.False(string.IsNullOrWhiteSpace(result.SessionId), "session_id should be populated");
Assert.False(string.IsNullOrWhiteSpace(result.ResultMarkdown), "result should be non-empty");
Assert.True(result.TokensOut > 0, "output_tokens should be > 0");
}
finally
{
try { Directory.Delete(workDir, true); } catch { }
}
}
}