ClawdDotNet/tests/ClawdDotNet.Core.Tests/Engine/HarnessHaertungTests.cs

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;

/// <summary>
/// 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.
/// </summary>
public class HarnessHaertungTests
{
    // ─────────────────────────── #547 ───────────────────────────

    [Fact]
    public async Task Der_Begleittext_zum_Werkzeugaufruf_bleibt_im_Verlauf()
    {
        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 verlauf = fixture.Client.ReceivedRequests[1].Messages;
        var mitAufruf = verlauf.Single(m => m.Role == "assistant" && m.ToolCalls is { Count: > 0 });

        Assert.Equal("Ich sehe zuerst in der Konfiguration nach.", mitAufruf.Content);
    }

    [Fact]
    public async Task Ohne_Begleittext_bleibt_die_Nachricht_ohne_Inhalt()
    {
        // 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 mitAufruf = fixture.Client.ReceivedRequests[1].Messages
            .Single(m => m.Role == "assistant" && m.ToolCalls is { Count: > 0 });

        Assert.Null(mitAufruf.Content);
    }

    // ─────────────────────────── #548 ───────────────────────────

    [Theory]
    [InlineData("length")]
    [InlineData("max_tokens")]
    [InlineData("LENGTH")]
    public void Ein_Abbruch_an_der_Ausgabegrenze_wird_erkannt(string grund)
        => Assert.True(AgentEngine.IsTruncated(grund));

    [Theory]
    [InlineData("stop")]
    [InlineData("tool_calls")]
    [InlineData("end_turn")]
    [InlineData(null)]
    public void Ein_regulaeres_Ende_gilt_nicht_als_Abbruch(string? grund)
        => Assert.False(AgentEngine.IsTruncated(grund));

    [Fact]
    public async Task Eine_abgeschnittene_Antwort_wird_nicht_als_fertig_gemeldet()
    {
        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 Eine_vollstaendige_Antwort_bleibt_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 Der_Denkverlauf_geht_unveraendert_zurueck()
    {
        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 mitAufruf = fixture.Client.ReceivedRequests[1].Messages
            .Single(m => m.Role == "assistant" && m.ToolCalls is { Count: > 0 });

        Assert.NotNull(mitAufruf.ReasoningDetails);

        // Signature and order must be preserved — by them the provider recognizes
        // its own block.
        var block = mitAufruf.ReasoningDetails!.Value[0];
        Assert.Equal("reasoning.text", block.GetProperty("type").GetString());
        Assert.Equal("sha256:abc", block.GetProperty("signature").GetString());
    }

    [Fact]
    public void Der_Denkverlauf_ueberlebt_Serialisierung_und_Rueckweg()
    {
        // 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 zurueck = JsonSerializer.Deserialize<ChatMessage>(json)!;

        Assert.Equal("Text", zurueck.Content);
        Assert.NotNull(zurueck.ReasoningDetails);
        Assert.Equal("reasoning.encrypted", zurueck.ReasoningDetails!.Value[0].GetProperty("type").GetString());
        Assert.Equal("AAAA", zurueck.ReasoningDetails!.Value[0].GetProperty("data").GetString());
    }

    [Fact]
    public void Ohne_Denkverlauf_steht_kein_leeres_Feld_in_der_Anfrage()
    {
        var ohne = ChatMessage.Assistant("Nur Text");
        var json = JsonSerializer.Serialize(ohne);

        Assert.DoesNotContain("reasoning", json);
    }

    // ─────────────────────────── #550 ───────────────────────────

    [Fact]
    public async Task Die_Anfrage_verlangt_einen_Anbieter_der_alles_beherrscht()
    {
        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 Die_Anbieterbindung_laesst_sich_je_Agent_abschalten()
    {
        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 Ohne_Denktiefe_geht_kein_reasoning_Feld_mit()
    {
        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 Die_Denktiefe_wird_richtig_uebersetzt(string einstellung, string? effort, int? tokens)
    {
        var fixture = new EngineFixture();
        var agent = fixture.AddAgent("a1");
        agent.ReasoningEffort = einstellung;
        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 Der_gemeldete_Betrag_wird_uebernommen_statt_geschaetzt()
    {
        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 Ohne_gemeldeten_Betrag_bleibt_das_Feld_leer()
    {
        // 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 Freigegebene_Werkzeuge_laufen_tatsaechlich_gleichzeitig()
    {
        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 Nicht_freigegebene_Werkzeuge_bleiben_seriell()
    {
        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 Die_Reihenfolge_der_Ergebnisse_folgt_den_Aufrufen()
    {
        // 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 StaffelTool();
        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 antworten = fixture.Client.ReceivedRequests[1].Messages
            .Where(m => m.Role == "tool")
            .ToList();

        Assert.Equal(["c1", "c2", "c3"], antworten.Select(m => m.ToolCallId));
        Assert.Equal(
            ["erste", "zweite", "dritte"],
            antworten.Select(m => m.Content?.ToString()?.Split('\n')[1]));
    }

    [Fact]
    public async Task Die_Grenze_der_Nebenlaeufigkeit_wird_eingehalten()
    {
        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 Bei_Grenze_eins_bleibt_alles_seriell()
    {
        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);
    }
}

/// <summary>
/// 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.
/// </summary>
internal sealed class StaffelTool : 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<ClawdDotNet.Core.Tools.ToolResult> ExecuteAsync(
        JsonElement input, ClawdDotNet.Core.Tools.AgentToolContext context, CancellationToken ct)
    {
        var delay = input.GetProperty("delay").GetInt32();
        var marke = input.GetProperty("marke").GetString()!;
        await Task.Delay(delay, ct);
        return ClawdDotNet.Core.Tools.ToolResult.Ok(marke);
    }
}