Merge wt/step-logs-t9: Task 9

src/ZB.MOM.WW.ScadaBridge.InboundAPI/EndpointExtensions.cs

@@ -280,6 +280,16 @@ public static class EndpointExtensions
             // production call so Replay of a real integration call is complete.
             trace: method.TraceAllCalls);
 
+        // Flow Debugger step logs (design 2026-09-09 D3): the executor sanitises the
+        // caller-facing message, so this stash is the ONLY place a method script's real
+        // exception survives for the inbound audit row. AuditWriteMiddleware consumes the
+        // slot once when it builds that row; the HTTP body below still carries only the
+        // safe ErrorMessage.
+        if (scriptResult.ErrorDetail is { } scriptErrorDetail)
+        {
+            httpContext.Items[AuditWriteMiddleware.ErrorDetailItemKey] = scriptErrorDetail;
+        }
+
         if (!scriptResult.Success)
         {
             // A client-aborted request is not a script failure.

src/ZB.MOM.WW.ScadaBridge.InboundAPI/InboundScriptExecutor.cs

@@ -599,8 +599,12 @@ public class InboundScriptExecutor
         catch (Exception ex)
         {
             _logger.LogError(ex, "Script execution failed for method {Method}", method.Name);
-            // Safe error message, no internal details
-            return new InboundScriptResult(false, null, "Internal script error", "SCRIPT_ERROR");
+            // Safe error message, no internal details — the caller only ever sees this.
+            // Flow Debugger step logs (design 2026-09-09 D3): the exception itself rides the
+            // additive ErrorDetail so the endpoint can put it on the inbound AUDIT row, where
+            // the Flow Debugger's inbound node reads it. It must never reach the HTTP body.
+            return new InboundScriptResult(
+                false, null, "Internal script error", "SCRIPT_ERROR", ErrorDetail: ex.ToString());
         }
         finally
         {
@@ -743,8 +747,18 @@ public class InboundScriptContext
 /// <c>SCRIPT_COMPILE_FAILED</c>, <c>SCRIPT_ERROR</c>). Null when the executor does not
 /// classify the failure; the endpoint then falls back to <c>SCRIPT_ERROR</c>.
 /// </param>
+/// <param name="ErrorDetail">
+/// Flow Debugger step logs (design 2026-09-09 D3): verbose, operator-only failure detail —
+/// the method script's exception <c>ToString()</c>. <b>Audit-row only.</b> The endpoint stashes
+/// it on <c>HttpContext.Items[AuditWriteMiddleware.ErrorDetailItemKey]</c> so it lands on the
+/// inbound audit row (and therefore on the Flow Debugger's inbound node); it is NEVER written
+/// to the HTTP response, which keeps carrying the sanitised <see cref="ErrorMessage"/>. Null
+/// for failures whose message already says everything (timeouts, client aborts, site-scope
+/// denials, compile failures).
+/// </param>
 public record InboundScriptResult(
     bool Success,
     string? ResultJson,
     string? ErrorMessage,
-    string? ErrorCode = null);
+    string? ErrorCode = null,
+    string? ErrorDetail = null);

src/ZB.MOM.WW.ScadaBridge.InboundAPI/Middleware/AuditWriteMiddleware.cs

@@ -79,6 +79,16 @@ public sealed class AuditWriteMiddleware
     /// </summary>
     public const string AuditActorItemKey = "ZB.MOM.WW.ScadaBridge.InboundAPI.AuditActor";
 
+    /// <summary>
+    /// <see cref="HttpContext.Items"/> key used by the endpoint handler to publish the
+    /// method script's exception detail when the script threw (Flow Debugger step logs,
+    /// design 2026-09-09 D3). The executor deliberately sanitises the caller-facing
+    /// message, so this slot is the only place the real cause survives; it is consumed
+    /// once here, onto the inbound audit row's <c>ErrorDetail</c>, and never written to
+    /// the HTTP response. An exception that ESCAPED the pipeline takes precedence.
+    /// </summary>
+    public const string ErrorDetailItemKey = "ScadaBridge.Audit.ErrorDetail";
+
     /// <summary>
     /// ParentExecutionId: <see cref="HttpContext.Items"/> key under
     /// which this middleware stashes the inbound request's per-request
@@ -318,6 +328,17 @@ public sealed class AuditWriteMiddleware
                 headerDict[header.Key] = value;
             }
 
+            // Flow Debugger step logs (design 2026-09-09 D3): an exception that ESCAPED the
+            // pipeline is the authoritative failure, so it wins; otherwise the endpoint may
+            // have stashed the method script's exception (which the executor keeps out of the
+            // HTTP body). Absent both, the row carries no detail.
+            var errorDetail = thrown?.ToString();
+            if (errorDetail is null
+                && ctx.Items.TryGetValue(ErrorDetailItemKey, out var stashedErrorDetail))
+            {
+                errorDetail = stashedErrorDetail as string;
+            }
+
             var extra = JsonSerializer.Serialize(new
             {
                 remoteIp = ctx.Connection.RemoteIpAddress?.ToString(),
@@ -347,6 +368,7 @@ public sealed class AuditWriteMiddleware
                 httpStatus: statusCode,
                 durationMs: (int)Math.Min(durationMs, int.MaxValue),
                 errorMessage: thrown?.Message,
+                errorDetail: errorDetail,
                 requestSummary: requestBody,
                 responseSummary: responseBody,
                 payloadTruncated: payloadTruncated,

tests/ZB.MOM.WW.ScadaBridge.InboundAPI.Tests/EndpointExtensionsTests.cs

@@ -432,7 +432,23 @@ public sealed class EndpointExtensionsTests : IDisposable
         var method = SeedMethod(1, "boom",
             """throw new System.InvalidOperationException("boom-msg");""");
 
-        using var host = await BuildHostAsync(method);
+        // Flow Debugger step logs (design 2026-09-09 D3): the detail the body must NOT carry
+        // is nonetheless stashed for AuditWriteMiddleware, so the inbound audit row (and the
+        // Flow Debugger's inbound node) can say why the run failed. A capture middleware reads
+        // the slot once the endpoint has finished, proving both halves in one request.
+        string? capturedDetail = null;
+        using var host = await BuildHostAsync(method, customize: builder =>
+        {
+            builder.Use(async (ctx, next) =>
+            {
+                await next();
+                if (ctx.Items.TryGetValue(
+                        AuditWriteMiddleware.ErrorDetailItemKey, out var stashed))
+                {
+                    capturedDetail = stashed as string;
+                }
+            });
+        });
         var token = await SeedKeyAsync(host, "key1", "caller", new[] { "boom" });
         var client = host.GetTestClient();
 
@@ -445,6 +461,10 @@ public sealed class EndpointExtensionsTests : IDisposable
         // message is not leaked (the executor sanitises before this point).
         Assert.Contains("error", body);
         Assert.DoesNotContain("boom-msg", body);
+
+        // …while the audit-only slot carries the real cause.
+        Assert.NotNull(capturedDetail);
+        Assert.Contains("boom-msg", capturedDetail);
     }
 
     [Fact]

tests/ZB.MOM.WW.ScadaBridge.InboundAPI.Tests/InboundScriptExecutorTests.cs

@@ -202,6 +202,56 @@ public class InboundScriptExecutorTests
         Assert.Equal("Internal script error", result.ErrorMessage);
     }
 
+    [Fact]
+    public async Task HandlerThrows_ErrorDetailCarriesTheException_WhileTheMessageStaysSafe()
+    {
+        // Flow Debugger step logs (design 2026-09-09 D3): the generic SCRIPT_ERROR catch keeps
+        // the caller-facing message sanitised, but the exception itself must survive on the
+        // additive ErrorDetail so the endpoint can put it on the inbound audit row (and the
+        // Flow Debugger's inbound node can show why the run failed).
+        var method = new ApiMethod("failing-detail", "throw new InvalidOperationException(\"real reason\");")
+        { Id = 1, TimeoutSeconds = 10 };
+        _executor.RegisterHandler("failing-detail", _ => throw new InvalidOperationException("real reason"));
+
+        var result = await _executor.ExecuteAsync(
+            method,
+            new Dictionary<string, object?>(),
+            _route,
+            TimeSpan.FromSeconds(10));
+
+        Assert.False(result.Success);
+        Assert.Equal("Internal script error", result.ErrorMessage);
+        Assert.Equal("SCRIPT_ERROR", result.ErrorCode);
+        Assert.NotNull(result.ErrorDetail);
+        Assert.Contains("InvalidOperationException", result.ErrorDetail);
+        Assert.Contains("real reason", result.ErrorDetail);
+        // The caller-safe fields must never carry it.
+        Assert.DoesNotContain("real reason", result.ErrorMessage!);
+    }
+
+    [Fact]
+    public async Task HandlerTimesOut_CarriesNoErrorDetail()
+    {
+        // A timeout is fully described by its message + TIMEOUT code — the
+        // OperationCanceledException adds nothing, so the row stays detail-less.
+        var method = new ApiMethod("slow-detail", "Thread.Sleep(60000);") { Id = 1, TimeoutSeconds = 1 };
+        _executor.RegisterHandler("slow-detail", async ctx =>
+        {
+            await Task.Delay(TimeSpan.FromSeconds(60), ctx.CancellationToken);
+            return "never";
+        });
+
+        var result = await _executor.ExecuteAsync(
+            method,
+            new Dictionary<string, object?>(),
+            _route,
+            TimeSpan.FromMilliseconds(100));
+
+        Assert.False(result.Success);
+        Assert.Equal("TIMEOUT", result.ErrorCode);
+        Assert.Null(result.ErrorDetail);
+    }
+
     [Fact]
     public async Task RoutedAskTimeout_Returns500WithSiteUnreachableCode()
     {

tests/ZB.MOM.WW.ScadaBridge.InboundAPI.Tests/Middleware/AuditWriteMiddlewareTests.cs

@@ -265,6 +265,74 @@ public class AuditWriteMiddlewareTests
         Assert.Equal("kaboom", evt.ErrorMessage);
     }
 
+    // ---------------------------------------------------------------------
+    // 4b. Error detail — the endpoint stashes a method script's exception on
+    //     HttpContext.Items (the HTTP body only ever carries the sanitised
+    //     message), and the middleware lifts it onto the row.
+    // ---------------------------------------------------------------------
+
+    [Fact]
+    public async Task ScriptErrorDetail_StashedOnItems_LandsOnTheRow()
+    {
+        var writer = new RecordingAuditWriter();
+        var ctx = BuildContext();
+        const string detail = "System.InvalidOperationException: real reason\n   at Script.<Main>";
+        var mw = CreateMiddleware(hc =>
+        {
+            hc.Items[AuditWriteMiddleware.ErrorDetailItemKey] = detail;
+            hc.Response.StatusCode = 500;
+            return Task.CompletedTask;
+        }, writer);
+
+        await mw.InvokeAsync(ctx);
+
+        var evt = Assert.Single(writer.Events);
+        Assert.Equal(AuditStatus.Failed, evt.Status);
+        Assert.Equal(detail, evt.ErrorDetail);
+    }
+
+    [Fact]
+    public async Task NoErrorDetailStashed_LeavesErrorDetailNull()
+    {
+        var writer = new RecordingAuditWriter();
+        var ctx = BuildContext();
+        var mw = CreateMiddleware(hc =>
+        {
+            hc.Response.StatusCode = 200;
+            return Task.CompletedTask;
+        }, writer);
+
+        await mw.InvokeAsync(ctx);
+
+        var evt = Assert.Single(writer.Events);
+        Assert.Null(evt.ErrorDetail);
+    }
+
+    [Fact]
+    public async Task ThrownException_ErrorDetail_WinsOverTheStashedItem()
+    {
+        // An exception that ESCAPED the pipeline is the richer, more authoritative
+        // source: it describes the failure the framework error path is about to report.
+        var writer = new RecordingAuditWriter();
+        var ctx = BuildContext();
+        var boom = new InvalidOperationException("kaboom");
+        var mw = CreateMiddleware(hc =>
+        {
+            hc.Items[AuditWriteMiddleware.ErrorDetailItemKey] = "stashed detail";
+            throw boom;
+        }, writer);
+
+        await Assert.ThrowsAsync<InvalidOperationException>(() => mw.InvokeAsync(ctx));
+
+        var evt = Assert.Single(writer.Events);
+        // Compared by content, not by an exact ToString(): the re-throw keeps appending
+        // stack frames after the row was built, so the captured text is a prefix of what
+        // the exception reads as by the time the assertion runs.
+        Assert.NotNull(evt.ErrorDetail);
+        Assert.StartsWith(boom.GetType().FullName + ": kaboom", evt.ErrorDetail);
+        Assert.DoesNotContain("stashed detail", evt.ErrorDetail);
+    }
+
     // ---------------------------------------------------------------------
     // 5. Actor resolution — the endpoint handler stashes the API key name
     //    AFTER successful auth so the middleware can pick it up from