Fremde Laufzeiten: Claude Code und Befehl in der Sandbox, Egress-Proxy, Vorschlag statt Einsatz (#712)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-25 22:46:49 +02:00
parent 7d6c2c9495
commit 367ea387f6
40 changed files with 3428 additions and 29 deletions
@@ -0,0 +1,142 @@
+using ClawdDotNet.Core.Accounting;
+using ClawdDotNet.Core.Engine;
+using ClawdDotNet.Core.Mcp;
+using ClawdDotNet.Core.Runtimes;
+using ClawdDotNet.Core.Sandbox;
+using ClawdDotNet.Core.Storage;
+using ClawdDotNet.Core.Tests.Infrastructure;
+using Microsoft.Extensions.Logging.Abstractions;
+using Shouldly;
+
+namespace ClawdDotNet.Core.Tests.Runtimes;
+
+/// <summary>
+/// #712: A foreign agent in the engine — it goes to the runner instead of the model, and
+/// everything around a run (usage, chat, the refusal without a runner) stays the same.
+/// </summary>
+public sealed class ForeignRunEngineTests : IDisposable
+{
+    private readonly string _root = Path.Combine(Path.GetTempPath(), "clawd-tests", "foreign-engine-" + Guid.NewGuid().ToString("N")[..8]);
+    private readonly SqliteUsageRepository _usage;
+    private readonly EngineFixture _fixture;
+
+    public ForeignRunEngineTests()
+    {
+        _usage = new SqliteUsageRepository(new SqliteStorage(Path.Combine(_root, "state.db")));
+        _fixture = new EngineFixture(usage: _usage);
+    }
+
+    public void Dispose()
+    {
+        SqliteStorage.ReleasePooledConnections(_root);
+        try { Directory.Delete(_root, recursive: true); }
+        catch { /* cleanup is secondary */ }
+    }
+
+    private sealed class ScriptedSandbox(string output) : ISandboxRunner
+    {
+        public SandboxRequest? Request { get; private set; }
+        public bool IsAvailable => true;
+        public string? UnavailableReason => null;
+
+        public Task<SandboxResult> RunAsync(SandboxRequest request, CancellationToken ct)
+        {
+            Request = request;
+            return Task.FromResult(new SandboxResult { ExitCode = 0, StandardOutput = output });
+        }
+    }
+
+    private Core.Config.AgentConfig ForeignAgent(string kind = RuntimeKinds.Command)
+    {
+        var agent = _fixture.AddAgent("coder");
+        agent.AgentDir = Path.Combine(_root, "Agents", "Agent-coder");
+        agent.WorkspacePath = Path.Combine(agent.AgentDir, "Workspace");
+        Directory.CreateDirectory(agent.WorkspacePath);
+        agent.Runtime = new AgentRuntimeConfig { Kind = kind, Command = "/usr/bin/answer" };
+        return agent;
+    }
+
+    private ScriptedSandbox UseSandbox(string output, ClaudeCodeSettings? claude = null)
+    {
+        var sandbox = new ScriptedSandbox(output);
+        _fixture.Engine.ForeignRunner = new ForeignRunner(sandbox,
+            [new CommandRuntime(), new ClaudeCodeRuntime(() => claude ?? new ClaudeCodeSettings())],
+            new McpGrants(), null, null, NullLogger.Instance);
+        return sandbox;
+    }
+
+    [Fact]
+    public async Task A_foreign_agent_does_not_reach_the_model()
+    {
+        var agent = ForeignAgent();
+        var sandbox = UseSandbox("answer from the command");
+
+        var result = await _fixture.Engine.RunAsync(agent, "Do it.", "test-instance", CancellationToken.None);
+
+        result.Status.ShouldBe(AgentRunStatus.Completed);
+        result.FinalMessage.ShouldBe("answer from the command");
+        _fixture.Client.ReceivedRequests.ShouldBeEmpty();
+        sandbox.Request!.StandardInput!.ShouldContain("Do it.");
+    }
+
+    [Fact]
+    public async Task The_briefing_is_the_one_a_native_run_gets()
+    {
+        var agent = ForeignAgent();
+        var sandbox = UseSandbox("ok");
+
+        await _fixture.Engine.RunAsync(agent, "Do it.", "test-instance", CancellationToken.None);
+
+        // The runner removed the run's files; the briefing went to the sandbox through them.
+        sandbox.Request!.Environment["CLAWD_BRIEFING_FILE"].ShouldBe("/run/agent/briefing.md");
+    }
+
+    [Fact]
+    public async Task Without_a_runner_the_agent_does_not_fall_back_to_the_native_engine()
+    {
+        var agent = ForeignAgent();
+
+        var result = await _fixture.Engine.RunAsync(agent, "Do it.", "test-instance", CancellationToken.None);
+
+        result.Status.ShouldBe(AgentRunStatus.Failed);
+        result.FinalMessage!.ShouldContain("not set up");
+        _fixture.Client.ReceivedRequests.ShouldBeEmpty();
+    }
+
+    [Fact]
+    public async Task A_subscription_run_is_recorded_with_its_tokens_and_marked()
+    {
+        var agent = ForeignAgent(RuntimeKinds.ClaudeCode);
+        var claude = Path.Combine(_root, "claude");
+        File.WriteAllText(claude, "");
+        UseSandbox("""
+            {"type":"system","subtype":"init","model":"claude-sonnet-5"}
+            {"type":"result","subtype":"success","is_error":false,"num_turns":3,"result":"done","total_cost_usd":0.9,"usage":{"input_tokens":100,"cache_read_input_tokens":900,"output_tokens":50}}
+            """, new ClaudeCodeSettings { Executable = claude, Auth = ClaudeCodeSettings.SubscriptionAuth, OAuthToken = "oat" });
+
+        await _fixture.Engine.RunAsync(agent, "Do it.", "test-instance", CancellationToken.None);
+
+        var usage = (await _usage.GetRecentAsync(5, CancellationToken.None)).Single();
+        usage.Runtime.ShouldBe(RuntimeKinds.ClaudeCode);
+        usage.OnSubscription.ShouldBeTrue();
+        usage.CostUsd.ShouldBe(0m, "a subscription does not bill the run");
+        usage.CostIsKnown.ShouldBeTrue();
+        usage.PromptTokens.ShouldBe(1000);
+        usage.CompletionTokens.ShouldBe(50);
+        usage.Model.ShouldBe("claude-sonnet-5");
+    }
+
+    [Fact]
+    public async Task In_a_chat_each_message_is_a_run_and_both_sides_are_shown()
+    {
+        var agent = ForeignAgent();
+        UseSandbox("the answer");
+
+        await _fixture.Engine.ChatAsync(agent, "a question", "test-instance", CancellationToken.None);
+
+        _fixture.Engine.GetChatHistory("coder").Select(e => (e.Role, e.Content)).ShouldBe([
+            ("user", "a question"),
+            ("assistant", "the answer")
+        ]);
+    }
+}