using System.Diagnostics; using System.Text.Json; using ClawdDotNet.Core.Api.Models; using ClawdDotNet.Core.Engine; using ClawdDotNet.Core.Tests.Infrastructure; namespace ClawdDotNet.Core.Tests.Engine; /// /// The hardening of the harness against silent loss: #547 to #551. /// /// What connects these five findings is not the place in the code, but the /// failure mode — none of them produces an error message. The run finishes, /// the result looks plausible, and only the quality is worse. Exactly /// that's why each needs a test: by hand such a thing is not noticed. /// public class HarnessHardeningTests { // ─────────────────────────── #547 ─────────────────────────── [Fact] public async Task The_text_accompanying_the_tool_call_stays_in_the_history() { var fixture = new EngineFixture().WithTool(FakeTool.Returning("Ergebnis")); var agent = fixture.AddAgent("a1", "TestTool"); fixture.Client .RespondsWithToolCallAndContent("TestTool", "Ich sehe zuerst in der Konfiguration nach.") .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); // The second request carries the history the first one produced. var history = fixture.Client.ReceivedRequests[1].Messages; var withCall = history.Single(m => m.Role == "assistant" && m.ToolCalls is { Count: > 0 }); Assert.Equal("Ich sehe zuerst in der Konfiguration nach.", withCall.Content); } [Fact] public async Task Without_accompanying_text_the_message_stays_without_content() { // Counter-check: an empty text must not go along as "". Some providers // reject an assistant message with empty content and tool_calls. var fixture = new EngineFixture().WithTool(FakeTool.Returning("Ergebnis")); var agent = fixture.AddAgent("a1", "TestTool"); fixture.Client .RespondsWithToolCall("TestTool") .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); var withCall = fixture.Client.ReceivedRequests[1].Messages .Single(m => m.Role == "assistant" && m.ToolCalls is { Count: > 0 }); Assert.Null(withCall.Content); } // ─────────────────────────── #548 ─────────────────────────── [Theory] [InlineData("length")] [InlineData("max_tokens")] [InlineData("LENGTH")] public void An_abort_at_the_output_limit_is_recognised(string reason) => Assert.True(AgentEngine.IsTruncated(reason)); [Theory] [InlineData("stop")] [InlineData("tool_calls")] [InlineData("end_turn")] [InlineData(null)] public void A_regular_end_does_not_count_as_an_abort(string? reason) => Assert.False(AgentEngine.IsTruncated(reason)); [Fact] public async Task A_truncated_answer_is_not_reported_as_done() { var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); fixture.Client.RespondsWithTruncatedText("Der Bericht beginnt mit"); var result = await fixture.Engine.RunAsync(agent, "Bericht", "inst", CancellationToken.None); Assert.Equal(AgentRunStatus.Truncated, result.Status); // The partial result goes along anyway — throwing it away would be worse than // passing it on marked. Assert.Equal("Der Bericht beginnt mit", result.FinalMessage); } [Fact] public async Task A_complete_answer_stays_Completed() { var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); fixture.Client.RespondsWithText("Fertiger Bericht."); var result = await fixture.Engine.RunAsync(agent, "Bericht", "inst", CancellationToken.None); Assert.Equal(AgentRunStatus.Completed, result.Status); } // ─────────────────────────── #549 ─────────────────────────── [Fact] public async Task The_reasoning_goes_back_unchanged() { const string details = """ [{"type":"reasoning.text","text":"Erst die Konfiguration prüfen.","id":"r1","format":"anthropic-claude-v1","index":0,"signature":"sha256:abc"}] """; var fixture = new EngineFixture().WithTool(FakeTool.Returning("Ergebnis")); var agent = fixture.AddAgent("a1", "TestTool"); fixture.Client .RespondsWithToolCallAndContent("TestTool", "Ich prüfe.", reasoningDetailsJson: details) .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); var withCall = fixture.Client.ReceivedRequests[1].Messages .Single(m => m.Role == "assistant" && m.ToolCalls is { Count: > 0 }); Assert.NotNull(withCall.ReasoningDetails); // Signature and order must be preserved — by them the provider recognizes // its own block. var block = withCall.ReasoningDetails!.Value[0]; Assert.Equal("reasoning.text", block.GetProperty("type").GetString()); Assert.Equal("sha256:abc", block.GetProperty("signature").GetString()); } [Fact] public void The_reasoning_survives_serialisation_and_the_way_back() { // The history is written to disk and loaded again at startup. // If the reasoning does not survive that, it is gone after a restart. const string details = """ [{"type":"reasoning.encrypted","data":"AAAA","id":"r1","index":0}] """; var original = ChatMessage.AssistantWithToolCalls( [new ToolCall { Id = "c1", Type = "function", Function = new ToolCallFunction { Name = "T", Arguments = "{}" } }], content: "Text", reasoningDetails: JsonDocument.Parse(details).RootElement.Clone()); var json = JsonSerializer.Serialize(original); var back = JsonSerializer.Deserialize(json)!; Assert.Equal("Text", back.Content); Assert.NotNull(back.ReasoningDetails); Assert.Equal("reasoning.encrypted", back.ReasoningDetails!.Value[0].GetProperty("type").GetString()); Assert.Equal("AAAA", back.ReasoningDetails!.Value[0].GetProperty("data").GetString()); } [Fact] public void Without_reasoning_there_is_no_empty_field_in_the_request() { var without = ChatMessage.Assistant("Nur Text"); var json = JsonSerializer.Serialize(without); Assert.DoesNotContain("reasoning", json); } // ─────────────────────────── #550 ─────────────────────────── [Fact] public async Task The_request_asks_for_a_provider_that_supports_everything() { var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); fixture.Client.RespondsWithText("ok"); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); var request = fixture.Client.ReceivedRequests[0]; Assert.True(request.Provider?.RequireParameters); Assert.True(request.Usage?.Include); } [Fact] public async Task The_provider_binding_can_be_switched_off_per_agent() { var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); agent.RequireProviderParameters = false; fixture.Client.RespondsWithText("ok"); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); Assert.Null(fixture.Client.ReceivedRequests[0].Provider); } [Fact] public async Task Without_reasoning_depth_no_reasoning_field_is_sent() { var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); // the default is "off" fixture.Client.RespondsWithText("ok"); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); Assert.Null(fixture.Client.ReceivedRequests[0].Reasoning); } [Theory] [InlineData("high", "high", null)] [InlineData("MAX", "max", null)] [InlineData("4096", null, 4096)] public async Task The_reasoning_depth_is_translated_correctly(string setting, string? effort, int? tokens) { var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); agent.ReasoningEffort = setting; fixture.Client.RespondsWithText("ok"); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); var reasoning = fixture.Client.ReceivedRequests[0].Reasoning; Assert.NotNull(reasoning); Assert.Equal(effort, reasoning!.Effort); Assert.Equal(tokens, reasoning.MaxTokens); } [Fact] public async Task The_reported_amount_is_taken_instead_of_estimated() { var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); fixture.Client.RespondsWithText("ok", new Usage { PromptTokens = 100, CompletionTokens = 20, TotalTokens = 120, Cost = 0.0042m }); var result = await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); Assert.Equal(0.0042m, result.ReportedCostUsd); } [Fact] public async Task Without_reported_amount_the_field_stays_empty() { // Important: not 0. A run booked as free would be worse than // no information at all, because it silently falsifies the budget calculation. var fixture = new EngineFixture(); var agent = fixture.AddAgent("a1"); fixture.Client.RespondsWithText("ok"); var result = await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); Assert.Null(result.ReportedCostUsd); } // ─────────────────────────── #551 ─────────────────────────── [Fact] public async Task Approved_tools_really_run_concurrently() { var tool = FakeTool.SlowParallelSafe(TimeSpan.FromMilliseconds(300)); var fixture = new EngineFixture().WithTool(tool); var agent = fixture.AddAgent("a1", "ParallelTool"); fixture.Client .RespondsWithToolCalls( ("ParallelTool", "{}", "c1"), ("ParallelTool", "{}", "c2"), ("ParallelTool", "{}", "c3")) .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); // The reliable proof: calls overlapped. // // Here there was additionally a wall-clock check (below 750 ms, serially it would be 900). // It failed on 2026-09-24 with 927 ms when the test projects ran in parallel — // the time measures the load of the machine, not the behavior of the engine. The // peak of simultaneous calls proves the same without wobbling. Assert.True(tool.MaxConcurrent > 1, $"Erwartet: überlappende Aufrufe. Höchststand war {tool.MaxConcurrent}."); } [Fact] public async Task Non_approved_tools_stay_serial() { var tool = FakeTool.SlowSerial(TimeSpan.FromMilliseconds(80)); var fixture = new EngineFixture().WithTool(tool); var agent = fixture.AddAgent("a1", "SerialTool"); fixture.Client .RespondsWithToolCalls( ("SerialTool", "{}", "c1"), ("SerialTool", "{}", "c2"), ("SerialTool", "{}", "c3")) .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); Assert.Equal(1, tool.MaxConcurrent); } [Fact] public async Task The_order_of_results_follows_the_calls() { // The slowest call comes first. If the engine sorted by completion, // it would be at the end — and every tool_call_id would be assigned wrongly. var tool = new DelayedMarkerTool(); var fixture = new EngineFixture().WithTool(tool); var agent = fixture.AddAgent("a1", "StaffelTool"); fixture.Client .RespondsWithToolCalls( ("StaffelTool", """{"delay":250,"marke":"erste"}""", "c1"), ("StaffelTool", """{"delay":10,"marke":"zweite"}""", "c2"), ("StaffelTool", """{"delay":120,"marke":"dritte"}""", "c3")) .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); var answers = fixture.Client.ReceivedRequests[1].Messages .Where(m => m.Role == "tool") .ToList(); Assert.Equal(["c1", "c2", "c3"], answers.Select(m => m.ToolCallId)); Assert.Equal( ["erste", "zweite", "dritte"], answers.Select(m => m.Content?.ToString()?.Split('\n')[1])); } [Fact] public async Task The_concurrency_limit_is_respected() { var tool = FakeTool.SlowParallelSafe(TimeSpan.FromMilliseconds(120)); var fixture = new EngineFixture().WithTool(tool); var agent = fixture.AddAgent("a1", "ParallelTool"); agent.MaxParallelToolCalls = 2; fixture.Client .RespondsWithToolCalls( ("ParallelTool", "{}", "c1"), ("ParallelTool", "{}", "c2"), ("ParallelTool", "{}", "c3"), ("ParallelTool", "{}", "c4")) .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); Assert.True(tool.MaxConcurrent <= 2, $"Grenze war 2, gleichzeitig liefen aber {tool.MaxConcurrent}."); } [Fact] public async Task With_limit_one_everything_stays_serial() { var tool = FakeTool.SlowParallelSafe(TimeSpan.FromMilliseconds(60)); var fixture = new EngineFixture().WithTool(tool); var agent = fixture.AddAgent("a1", "ParallelTool"); agent.MaxParallelToolCalls = 1; fixture.Client .RespondsWithToolCalls( ("ParallelTool", "{}", "c1"), ("ParallelTool", "{}", "c2")) .RespondsWithText("Fertig."); await fixture.Engine.RunAsync(agent, "Los", "inst", CancellationToken.None); Assert.Equal(1, tool.MaxConcurrent); } } /// /// Tool that returns a marker after a time given in the call. /// This allows checking whether the order of the results follows the calls /// and not the completion. /// internal sealed class DelayedMarkerTool : ClawdDotNet.Core.Tools.IAgentTool { public string Name => "StaffelTool"; public string Description => "Wartet und meldet eine Marke."; public bool IsParallelSafe => true; public JsonElement InputSchema { get; } = JsonDocument.Parse(""" { "type": "object", "properties": { "delay": { "type": "integer" }, "marke": { "type": "string" } } } """).RootElement.Clone(); public async Task ExecuteAsync( JsonElement input, ClawdDotNet.Core.Tools.AgentToolContext context, CancellationToken ct) { var delay = input.GetProperty("delay").GetInt32(); var mark = input.GetProperty("marke").GetString()!; await Task.Delay(delay, ct); return ClawdDotNet.Core.Tools.ToolResult.Ok(mark); } }