ClawdDotNet/tests/ClawdDotNet.Tools.Tests/Taskboard/SpotCheckToolTests.cs

using System.Text.Json;
using ClawdDotNet.Core.Audit;
using ClawdDotNet.Core.Config;
using ClawdDotNet.Core.Inspection;
using ClawdDotNet.Core.Organisation;
using ClawdDotNet.Core.Storage;
using ClawdDotNet.Core.Tasks;
using ClawdDotNet.Core.Tools;
using ClawdDotNet.Tools.Taskboard;
using ClawdDotNet.Tools.Tests.FileRW;
using Microsoft.Extensions.Logging.Abstractions;
using Shouldly;

namespace ClawdDotNet.Tools.Tests.Taskboard;

/// <summary>#693: The lead's side of the spot checks through the Escalation tool.</summary>
public sealed class SpotCheckToolTests : IDisposable
{
    private readonly string _directory = Path.Combine(Path.GetTempPath(), "clawd-tests", Guid.NewGuid().ToString("N"));
    private readonly SqliteStorage _storage;
    private readonly SqliteAuditRepository _audit;
    private readonly SpotChecks _checks;
    private readonly EscalationTool _tool = new();

    private readonly OrgChart _org = new(
        [
            new UnitConfig { Id = "purchasing", Name = "Purchasing", LeaderAgentId = "buy-lead", Inspection = new UnitInspection { Rate = 1, DenseRate = 1, PerDay = 5 } },
            new UnitConfig { Id = "sales", Name = "Sales", LeaderAgentId = "sales-lead" }
        ],
        [
            new AgentConfig { AgentId = "buy-lead", UnitId = "purchasing" },
            new AgentConfig { AgentId = "buyer", UnitId = "purchasing" },
            new AgentConfig { AgentId = "sales-lead", UnitId = "sales" },
            new AgentConfig { AgentId = "seller", UnitId = "sales" }
        ]);

    public SpotCheckToolTests()
    {
        _storage = new SqliteStorage(Path.Combine(_directory, "state.db"));
        _audit = new SqliteAuditRepository(_storage);
        var tasks = new SqliteTaskRepository(_storage);
        _checks = new SpotChecks(_storage, _audit,
            new RunInspector(_audit, tasks, null, () => RunInspector.VocabularyOf([_tool])),
            () => _org, new TaskboardService(tasks, Path.Combine(_directory, "tasks")), NullLogger.Instance);
    }

    public void Dispose()
    {
        SqliteStorage.ReleasePooledConnections(_directory);
        try { Directory.Delete(_directory, recursive: true); }
        catch { /* cleanup is secondary */ }
    }

    private Task<ToolResult> Run(string agent, object input)
        => _tool.ExecuteAsync(JsonSerializer.SerializeToElement(input),
            new AgentToolContext(agent, "inst", new Dictionary<string, object?>(), new NullStateStore(), NullLogger.Instance, CancellationToken.None)
            {
                Organisation = _org,
                SpotChecks = _checks
            }, CancellationToken.None);

    private async Task RecordAsync(string runId, string agent, string said)
    {
        await _audit.AppendAsync(new AuditEntry
        {
            RunId = runId, AgentId = agent, Tool = "FileRW", Arguments = "{\"action\":\"append\"}",
            Status = AuditStatus.Ok, OccurredAt = DateTime.UtcNow.AddMinutes(-2)
        }, default);
        await _audit.RecordReceiptAsync(new RunReceipt
        {
            RunId = runId, AgentId = agent, Model = "m", Status = "Completed", StepCount = 2,
            ResultRef = said, OccurredAt = DateTime.UtcNow.AddMinutes(-1)
        }, default);
    }

    [Fact]
    public async Task The_lead_sees_its_checks_with_the_audit_findings_and_closes_them()
    {
        await RecordAsync("run-b", "buyer", "Escalation doctrine_set was refused.");
        var check = (await _checks.SampleAsync(default)).Single();

        var list = await Run("buy-lead", new { action = "spot_checks" });
        list.Content.ShouldContain($"check {check.Id}");
        list.Content.ShouldContain("doctrine_set", customMessage: "the claim without a call is shown from the audit");

        var detail = await Run("buy-lead", new { action = "inspect_run", run = "run-b" });
        detail.Content.ShouldContain("FileRW.append: Ok");
        detail.Content.ShouldContain("its claim, not evidence");

        (await Run("buy-lead", new { action = "check_done", check = check.Id, verdict = "finding", note = "Reports a refusal that never happened." }))
            .Success.ShouldBeTrue();
        (await Run("buy-lead", new { action = "spot_checks" })).Content.ShouldContain("No spot check waits for you");
    }

    [Fact]
    public async Task A_lead_sees_no_runs_of_agents_it_does_not_lead()
    {
        await RecordAsync("run-s", "seller", "Done.");

        (await Run("buy-lead", new { action = "inspect_run", run = "run-s" })).ErrorMessage!.ShouldContain("do not lead");
        (await Run("buyer", new { action = "inspect_run", run = "run-s" })).Success.ShouldBeFalse();
    }

    [Fact]
    public async Task A_measure_needs_a_known_kind_and_goes_to_a_human()
    {
        (await Run("buy-lead", new { action = "propose_measure", agent = "buyer", measure = "fire", reasoning = "x" }))
            .ErrorMessage!.ShouldContain("'measure'");

        var proposed = await Run("buy-lead", new { action = "propose_measure", agent = "buyer", measure = "change_model", reasoning = "Fails every table task." });
        proposed.Success.ShouldBeTrue(proposed.ErrorMessage);
        proposed.Content.ShouldContain("with a human");
    }
}