refactor(logging): partition the log stream by session — session.log per session, daemon.log sparse, OTEL the union (#1472)

feeds/skills/.system/files/netclaw-operations/SKILL.md

@@ -3,7 +3,7 @@ name: netclaw-operations
 description: "REQUIRED when the user asks about scheduling, reminders, cron jobs, timers, background jobs, diagnostics, troubleshooting, MCP tools, daemon health, identity updates, or Netclaw capabilities and self-maintenance."
 metadata:
   author: netclaw
-  version: "2.18.0"
+  version: "2.19.0"
 ---
 
 # Netclaw Operations

feeds/skills/.system/files/netclaw-operations/references/diagnostics.md

@@ -34,16 +34,19 @@ What to expect inside `session.log`:
   cross-referencing two files.
 - Two line shapes: session output audit lines (`User:`, `Assistant:`,
   `Thinking:`, `Tool call:`, `Tool result:`, `Usage:`, `Turn N completed`,
-  etc.) and `Diagnostic:` lines from MEL providers (LLM client, HTTP,
-  retry middleware) emitted under a session diagnostics scope.
+  etc.) and `Diagnostic:` lines. A diagnostic line reaches `session.log`
+  when its log event carries the session id; actor logs are tagged with it
+  automatically (`SessionId`), so session- and sub-agent-actor lifecycle
+  lines (startup, guards, timeouts, cancellation, completion) appear here —
+  routed by the tag on the line, not by any ambient context.
 - Best-effort observability. Individual lines may be dropped on transient
   IO errors and logged at Debug level in the daemon log; this is not a
   transactional audit trail. Cross-check the daemon log for warnings
   if a critical line appears missing.
-- Sidecar paths (compaction, title generation, sub-agents, memory
-  distillation) currently bypass the session diagnostics scope, so their
-  internal diagnostics may not appear in `session.log` even though their
-  output audit lines do. Tracked in netclaw-dev/netclaw#920.
+- Not everything is session-tagged: the LLM-client / HTTP / retry / failover
+  decorators log their internal diagnostics without a session id, so those
+  land in `daemon.log` only. For deep LLM-call internals, read `daemon.log`
+  (filter by the `SessionId` attribute where the line carries one).
 
 | Symptom | Check |
 |---------|-------|

src/Netclaw.Configuration.Tests/SessionDiagnosticsContextTests.cs → src/Netclaw.Actors.Tests/SubAgents/SubAgentSessionScopeTests.cs

@@ -1,22 +1,22 @@
 // -----------------------------------------------------------------------
-// <copyright file="SessionDiagnosticsContextTests.cs" company="Petabridge, LLC">
+// <copyright file="SubAgentSessionScopeTests.cs" company="Petabridge, LLC">
 //      Copyright (C) 2026 - 2026 Petabridge, LLC <https://petabridge.com>
 // </copyright>
 // -----------------------------------------------------------------------
-using Netclaw.Configuration;
+using Netclaw.Actors.SubAgents;
 using Xunit;
 
-namespace Netclaw.Configuration.Tests;
+namespace Netclaw.Actors.Tests.SubAgents;
 
-public sealed class SessionDiagnosticsContextTests
+public sealed class SubAgentSessionScopeTests
 {
     [Theory]
     [InlineData("ch/thr/subagent/skill", "ch/thr")]
     [InlineData("ch/thr", "ch/thr")]
     [InlineData("ch/thr/subagent/skill/inner", "ch/thr")]
     public void NormalizeSessionId_collapses_subagent_marker_to_parent_id(string input, string expected)
     {
-        Assert.Equal(expected, SessionDiagnosticsContext.NormalizeSessionId(input));
+        Assert.Equal(expected, SubAgentSessionScope.NormalizeSessionId(input));
     }
 
     [Theory]
@@ -25,6 +25,6 @@ public void NormalizeSessionId_collapses_subagent_marker_to_parent_id(string inp
     [InlineData("   ")]
     public void NormalizeSessionId_returns_null_for_blank(string? input)
     {
-        Assert.Null(SessionDiagnosticsContext.NormalizeSessionId(input));
+        Assert.Null(SubAgentSessionScope.NormalizeSessionId(input));
     }
 }

src/Netclaw.Actors.Tests/SubAgents/SubAgentSpawnObservabilityTests.cs

@@ -14,14 +14,12 @@
 namespace Netclaw.Actors.Tests.SubAgents;
 
 /// <summary>
-/// Regression coverage for sub-agent spawn observability. A sub-agent's own actor
-/// logs go through Akka's async logger bridge, where the diagnostics AsyncLocal is
-/// gone, so they never reach the per-session <c>session.log</c>. The spawn lifecycle
-/// is instead recorded by parent-side breadcrumbs that must run while the parent
-/// session scope is active — otherwise a refused or failed spawn is invisible in the
-/// session transcript. These tests assert the scope is active at log time, which is
-/// exactly the condition <c>RollingFileLoggerProvider</c> uses to route a line to
-/// <c>session.log</c>.
+/// Regression coverage for sub-agent spawn observability. The spawn lifecycle is
+/// recorded by parent-side breadcrumbs that carry the session id as a structured log
+/// field — exactly what <c>RollingFileLoggerProvider</c> reads off the log event to
+/// route a line to the per-session <c>session.log</c>. These tests assert each
+/// breadcrumb carries the session id; otherwise a refused or failed spawn would be
+/// invisible in the session transcript.
 /// </summary>
 public sealed class SubAgentSpawnObservabilityTests : IDisposable
 {
@@ -113,6 +111,17 @@ public void Log<TState>(
             TState state,
             Exception? exception,
             Func<TState, Exception?, string> formatter)
-            => Entries.Add((logLevel, formatter(state, exception), SessionDiagnosticsContext.SessionId));
+            => Entries.Add((logLevel, formatter(state, exception), ExtractSessionId(state)));
+
+        // Mirror RollingFileLoggerProvider: read the session id off the structured log
+        // state (the {SessionId} field the breadcrumbs carry), not an ambient context.
+        private static string? ExtractSessionId<TState>(TState state)
+        {
+            if (state is IEnumerable<KeyValuePair<string, object?>> fields)
+                foreach (var field in fields)
+                    if (field.Key == "SessionId" && field.Value is { } value)
+                        return value.ToString();
+            return null;
+        }
     }
 }

src/Netclaw.Actors/Memory/MemoryCurationActor.cs

@@ -298,7 +298,6 @@ private async Task<CurationDecision> EvaluateSingleAsync(
         try
         {
             using var cts = new CancellationTokenSource(LlmTimeout);
-            using var diagnosticsScope = SessionDiagnosticsContext.Push(sessionId.Value);
 
             var messages = new List<ChatMessage>
             {

src/Netclaw.Actors/Protocol/SessionLogDiagnostic.cs

@@ -4,16 +4,15 @@
 // </copyright>
 // -----------------------------------------------------------------------
 using Akka.Actor;
-using Netclaw.Configuration;
 
 namespace Netclaw.Actors.Protocol;
 
 /// <summary>
-/// Pre-formatted diagnostic line carried from the MEL logger provider
-/// to the <c>SessionLogDispatcher</c>. The provider snapshots
-/// <c>SessionDiagnosticsContext.SessionId</c> at log-emit time and embeds
-/// it here so the dispatcher routes the line by message field, not by
-/// any ambient context that would not flow across actor mailboxes.
+/// Pre-formatted diagnostic line carried from the MEL logger provider to the
+/// <c>SessionLogDispatcher</c>. The provider reads the session id off each log
+/// event's structured state and embeds it here so the dispatcher routes the line
+/// by message field, not by any ambient context that would not flow across actor
+/// mailboxes.
 /// </summary>
 public sealed record SessionLogDiagnostic(SessionId SessionId, string Line)
     : IWithSessionId, INoSerializationVerificationNeeded;

src/Netclaw.Actors/Sessions/LlmSessionActor.cs

@@ -1737,7 +1737,6 @@ internal static async Task InvokeMemoryExtractionCoreAsync(
         try
         {
             using var cts = new CancellationTokenSource(timeout);
-            using var diagnosticsScope = SessionDiagnosticsContext.Push(sessionId.Value);
             var extractionMessages = new List<AiChatMessage>
             {
                 new(Microsoft.Extensions.AI.ChatRole.System,

src/Netclaw.Actors/Sessions/Pipelines/SessionCompactionPipeline.cs

@@ -173,7 +173,6 @@ public static async Task ExecuteAsync(
         {
             using var cts = CancellationTokenSource.CreateLinkedTokenSource(cancellationToken);
             cts.CancelAfter(sidecarTimeout);
-            using var diagnosticsScope = SessionDiagnosticsContext.Push(sessionId.Value);
             var observerMessages = new List<AiChatMessage>
             {
                 new(Microsoft.Extensions.AI.ChatRole.System,