126 lines
5.0 KiB
C#
126 lines
5.0 KiB
C#
using ClawdDotNet.Core.Audit;
|
|
using ClawdDotNet.Core.Tests.Infrastructure;
|
|
using Shouldly;
|
|
|
|
namespace ClawdDotNet.Core.Tests.Audit;
|
|
|
|
/// <summary>
|
|
/// A3 an der Engine: Jeder Tool-Aufruf wird protokolliert, jeder Lauf bekommt einen Beleg.
|
|
/// Entscheidend ist die Provenienz — die Engine stempelt Agent, Modell und Herkunft aus
|
|
/// ihrem eigenen Wissen, nicht aus dem Tool-Ergebnis.
|
|
/// </summary>
|
|
public sealed class EngineAuditTests
|
|
{
|
|
[Fact]
|
|
public async Task Jeder_Tool_Aufruf_landet_im_Audit_Log()
|
|
{
|
|
var audit = new InMemoryAuditRepository();
|
|
var fixture = new EngineFixture(audit).WithTool(FakeTool.Returning("ok"));
|
|
var agent = fixture.AddAgent("agent-a", "TestTool");
|
|
|
|
// Zwei Tool-Aufrufe, dann Text.
|
|
fixture.Client
|
|
.RespondsWithToolCall("TestTool")
|
|
.RespondsWithToolCall("TestTool")
|
|
.RespondsWithText("fertig");
|
|
|
|
await fixture.Engine.RunAsync(agent, "los", "test-instance", default);
|
|
|
|
var entries = audit.Entries.ToList();
|
|
entries.Count.ShouldBe(2);
|
|
entries.ShouldAllBe(e => e.Tool == "TestTool");
|
|
entries.ShouldAllBe(e => e.Status == AuditStatus.Ok);
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Die_Herkunft_wird_von_der_Engine_gestempelt()
|
|
{
|
|
var audit = new InMemoryAuditRepository();
|
|
var fixture = new EngineFixture(audit).WithTool(FakeTool.Returning("ok"));
|
|
var agent = fixture.AddAgent("agent-a", "TestTool");
|
|
fixture.Client.RespondsWithToolCall("TestTool").RespondsWithText("fertig");
|
|
|
|
await fixture.Engine.RunAsync(agent, "los", "test-instance", default, source: "task", taskId: "t-7");
|
|
|
|
var entry = audit.Entries.Single();
|
|
entry.AgentId.ShouldBe("agent-a");
|
|
entry.Model.ShouldBe("test/model");
|
|
entry.Source.ShouldBe("task");
|
|
entry.RunId.ShouldNotBeNullOrWhiteSpace();
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Ein_verweigerter_Aufruf_wird_als_denied_protokolliert()
|
|
{
|
|
var audit = new InMemoryAuditRepository();
|
|
// Tool ist registriert, aber dem Agenten NICHT zugewiesen → PermissionGate greift.
|
|
var fixture = new EngineFixture(audit).WithTool(FakeTool.Returning("ok"));
|
|
var agent = fixture.AddAgent("agent-a" /* kein Tool zugewiesen */);
|
|
fixture.Client.RespondsWithToolCall("TestTool").RespondsWithText("fertig");
|
|
|
|
await fixture.Engine.RunAsync(agent, "los", "test-instance", default);
|
|
|
|
audit.Entries.Single().Status.ShouldBe(AuditStatus.Denied);
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Ein_Tool_Fehler_wird_als_error_protokolliert()
|
|
{
|
|
var audit = new InMemoryAuditRepository();
|
|
var fixture = new EngineFixture(audit)
|
|
.WithTool(FakeTool.Throwing(new InvalidOperationException("kaputt")));
|
|
var agent = fixture.AddAgent("agent-a", "TestTool");
|
|
fixture.Client.RespondsWithToolCall("TestTool").RespondsWithText("fertig");
|
|
|
|
await fixture.Engine.RunAsync(agent, "los", "test-instance", default);
|
|
|
|
var entry = audit.Entries.Single();
|
|
entry.Status.ShouldBe(AuditStatus.Error);
|
|
entry.Summary.ShouldContain("kaputt");
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Ein_Lauf_bekommt_einen_Receipt_mit_Task_Verknuepfung()
|
|
{
|
|
var audit = new InMemoryAuditRepository();
|
|
var fixture = new EngineFixture(audit).WithTool(FakeTool.Returning("ok"));
|
|
var agent = fixture.AddAgent("agent-a", "TestTool");
|
|
fixture.Client.RespondsWithToolCall("TestTool").RespondsWithText("erledigt");
|
|
|
|
await fixture.Engine.RunAsync(agent, "los", "test-instance", default, source: "task", taskId: "t-42");
|
|
|
|
var receipt = audit.Receipts.Single();
|
|
receipt.TaskId.ShouldBe("t-42");
|
|
receipt.Source.ShouldBe("task");
|
|
receipt.Status.ShouldBe("Completed");
|
|
receipt.ResultRef.ShouldBe("erledigt");
|
|
receipt.StepCount.ShouldBeGreaterThan(0);
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Receipt_und_Audit_teilen_dieselbe_RunId()
|
|
{
|
|
var audit = new InMemoryAuditRepository();
|
|
var fixture = new EngineFixture(audit).WithTool(FakeTool.Returning("ok"));
|
|
var agent = fixture.AddAgent("agent-a", "TestTool");
|
|
fixture.Client.RespondsWithToolCall("TestTool").RespondsWithText("fertig");
|
|
|
|
await fixture.Engine.RunAsync(agent, "los", "test-instance", default);
|
|
|
|
var runId = audit.Receipts.Single().RunId;
|
|
audit.Entries.Single().RunId.ShouldBe(runId, "alle Aufrufe eines Laufs tragen dieselbe RunId");
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Ohne_Audit_Repository_laeuft_alles_normal_weiter()
|
|
{
|
|
// Gegenprobe: A3 ist optional — die Engine darf ohne Audit-Repo nicht anders laufen.
|
|
var fixture = new EngineFixture().WithTool(FakeTool.Returning("ok"));
|
|
var agent = fixture.AddAgent("agent-a", "TestTool");
|
|
fixture.Client.RespondsWithToolCall("TestTool").RespondsWithText("fertig");
|
|
|
|
var result = await fixture.Engine.RunAsync(agent, "los", "test-instance", default);
|
|
result.Status.ToString().ShouldBe("Completed");
|
|
}
|
|
}
|