#362: make the plugin visible, and fix the drift that made it unusable

CB-527 shipped a Claude Code plugin and a marketplace in this repo. Nothing in
CLAUDE.md or docs/ ever named it, so a later session planned the same feature
from scratch. The wiki Features entry existed and was correct, but wiki/ is a
submodule whose pointer is never advanced, so no session reads it.

Visibility:
- CLAUDE.md addendum now names plugin/ and both structural limits, so every
  session sees it. This is the change that stops the rebuild happening again.
- wiki/11-Features.md records the rename and why the entry alone was not enough.

Drift (each measured against the code, not assumed):
- mount name fleetd -> fleet, matching PeerLauncher.MCP_MOUNT_NAME. The old name
  gave a lead with both a project .mcp.json and the plugin two mounts of one
  daemon and a duplicated fleet_* tool set.
- url is now ${FLEETD_MCP_URL} instead of a hardcoded address, so one plugin can
  serve hosts running the daemon on different ports. Plain ${VAR}, the form
  kb-alms proves works here; ${VAR:-default} is untested and not used.
- plugin claude-bridge -> fleet, marketplace claude-bridge -> fleetd, version
  0.2.0. Breaking for a 0.1.0 install: mcp__fleetd__* becomes mcp__fleet__*.
- README install path ltms/claude-bridge -> the fleet/fleetd remote.
- the setup skill's §5 told operators to pin primary.terminal:. CB-579 replaced
  that with fleet.leaders.*.tab. Replaced, with the duplicate-tab warning (#359).

Scope: the plugin is lead-side only, and cannot be otherwise. The launcher adds
--agent only when <worktree>/.claude/agents/<role>.md exists in the member's own
tree (ClaudeCodeLauncher.java:371,391), and a member's CLAUDE_CONFIG_DIR points
at its profile's config dir (ClaudeCodeLauncher.java:285), so a member never
reads the operator's plugin store. On this Mac all four Claude profiles set
configDir, and the four ccs instances hold four separate copies of the plugin
store -- same md5, different inodes. Seeding member skills through the worktree
is #362 scope item 3, implemented separately.

Note for anyone verifying a plugin: `claude plugin validate` does NOT read
.mcp.json. Replacing it with `{ this is not json at all` still passes, exit 0.

Refs #362, #359

.claude-plugin/marketplace.json

@@ -1,15 +1,15 @@
 {
-  "name": "claude-bridge",
+  "name": "fleetd",
   "description": "Tooling for orchestrating a fleet of delegated coding agents through the fleetd MCP gateway.",
   "owner": {
     "name": "LTMS"
   },
   "plugins": [
     {
-      "name": "claude-bridge",
+      "name": "fleet",
       "source": "./plugin",
-      "description": "Make a project bridge-ready: mount the fleetd MCP gateway and apply standard Claude Code settings so a session can orchestrate delegated workers. Ships no credentials.",
-      "version": "0.1.0",
+      "description": "Mount the fleetd MCP gateway and apply standard Claude Code settings so a session can orchestrate delegated workers. Ships no credentials.",
+      "version": "0.2.0",
       "author": {
         "name": "LTMS"
       }

CLAUDE.md

@@ -210,6 +210,17 @@ must obey belongs in the charter, not here.
 - **Primary-side skills** (not delegation playbooks — a worker cannot use them):
   `port-to-opencode` (make an OpenCode session a participant in this workspace) and
   `fleets-status` (report every fleet that shares one LavinMQ instance).
+- **This repo is also a Claude Code marketplace, and ships a plugin.** `.claude-plugin/marketplace.json`
+  points at `plugin/`, which carries the MCP mount and the `setup` skill
+  (`/claude-bridge:setup` — make any project bridge-ready). It was added in CB-527 and then went
+  unmentioned by every instruction file, so it drifted and a later session planned it from scratch
+  (#362). **Read `plugin/` before designing anything about onboarding a project.** Two limits are
+  structural, not bugs: a plugin cannot carry the role agent files, because
+  `ClaudeCodeLauncher.java:371` requires `<cwd>/.claude/agents/<role>.md` in the member's own
+  worktree; and a plugin cannot deliver anything to members at all, because
+  `ClaudeCodeLauncher.java:285` exports `CLAUDE_CONFIG_DIR` and every Claude profile here sets it,
+  so a member never reads the operator's plugin store. **The plugin is the lead-side surface;
+  member-facing assets travel in the worktree.**
 - **Never commit** `.mcp.json` (the primary's local copy, flagged `--skip-worktree`) or `wiki/`
   (a submodule with its own remote).
 - **A provisioned worktree neutralizes `.mcp.json`, `opencode.json` and `.autoenv`** — the repo's

fleetd/src/main/java/dev/ltms/fleet/inject/CompletionResolver.java

@@ -380,7 +380,9 @@ public final class CompletionResolver implements TurnListener {
                         + "matched the profile's exhausted pattern): {}", target, reason);
                 // CB-578 stage B: only on the resolution that actually won the race — a late
                 // duplicate must never quarantine a credential twice for one refusal.
-                exhaustionSink.onExhausted(target, reason);
+                if (startsWithExhaustion(matchedLine, exhausted)) {
+                    exhaustionSink.onExhausted(target, reason);
+                }
             }
             return;
         }
@@ -478,7 +480,9 @@ public final class CompletionResolver implements TurnListener {
                         + "usable assistant block; no fleet_reply): {}", target, reason);
                 // CB-578 stage B: only on the resolution that actually won the race — a late
                 // duplicate must never quarantine a credential twice for one refusal.
-                exhaustionSink.onExhausted(target, reason);
+                if (startsWithExhaustion(matchedLine, exhausted)) {
+                    exhaustionSink.onExhausted(target, reason);
+                }
             }
             return true;
         }
@@ -613,6 +617,46 @@ public final class CompletionResolver implements TurnListener {
         return null;
     }
 
+    /**
+     * True when nothing before the match on this pane line ends a sentence — that is, the match is
+     * still inside the line's first sentence rather than inside prose a member wrote about it.
+     * Used to decide whether an exhaustion match may quarantine a credential (fleetd #348).
+     *
+     * <p><strong>Why this is looser than {@link #startsWithBackendError}.</strong> An
+     * {@code exhaustedPattern} is written per profile and may name only the decisive words of a
+     * provider message — {@code "usage limit has been reached"} without its leading {@code "The"}.
+     * A start-of-line check would then reject the genuine refusal. That is the false negative
+     * fleetd #348's invariant 1 calls the worse direction: an unrecorded exhaustion leaves the
+     * fleet spawning into a credential with no capacity, and a quarantine runs 1800s against the
+     * backend-error cooldown's fixed 60s.
+     *
+     * <p>This rule accepts a superset of what a start-of-line check accepts: if the match begins
+     * right after the chrome, there is nothing in front of it, so there is no sentence ending
+     * either. So moving to it cannot add a false negative.
+     *
+     * <p><strong>No chrome skipping here, deliberately.</strong> The first version of this method
+     * copied {@code startsWithBackendError}'s leading-chrome loop. Measured on merge: deleting that
+     * loop left all 1369 tests green, and it must — the scan only looks for {@code . ! ?}, and no
+     * terminal chrome character is one of those. A step that cannot change the result is worse than
+     * no step, because the next reader takes it as evidence that chrome was handled.
+     *
+     * <p>It stays a heuristic. Prose whose <em>first</em> sentence carries the pattern still
+     * notifies the sink, and a genuine refusal behind an earlier full stop (a hostname, a version
+     * number) still does not. Both are known and neither is fixed here.
+     */
+    private static boolean startsWithExhaustion(String line, Pattern pattern) {
+        var matcher = pattern.matcher(line);
+        if (!matcher.find()) {
+            return false;
+        }
+        for (int prefix = 0; prefix < matcher.start(); prefix++) {
+            if (".!?".indexOf(line.charAt(prefix)) >= 0) {
+                return false;
+            }
+        }
+        return true;
+    }
+
     /**
      * True when the error pattern begins the matched pane line, rather than appearing in prose.
      *

fleetd/src/main/java/dev/ltms/fleet/member/HerdrPeerLauncher.java

@@ -937,14 +937,20 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
      */
     @Override
     public void stop(String idOrPane) {
-        // Teardown knows only the pane, not which profile spawned it. Attempt tab cleanup when any
-        // profile uses tab placement (so the bridge may have created a dedicated peer tab); the
-        // single-occupant check below is what actually protects the user's shared tabs.
+        // Teardown knows only the pane, not which profile spawned it — and, when this call is
+        // routed here through CompositePeerLauncher's single-daemon stop() shortcut (fleetd #342),
+        // not even which adapter's config actually governed the spawn: the shortcut can hand the
+        // pane to a delegate that never spawned it, whose own profiles say nothing about how THIS
+        // pane was placed. So the decision to look for a tab to clean up is made from the pane's
+        // actual state, not from this delegate's static profile config: resolve the tab
+        // unconditionally and let {@link WorkspaceControl#locatePane} tolerate "not found" (it
+        // returns null rather than throwing); the single-occupant check below is what actually
+        // protects the user's shared tabs, exactly as it always has.
         String paneId = paneByAgentId.remove(idOrPane);
         if (paneId == null) {
             paneId = idOrPane; // raw-pane fallback (reap, gate timeout, pane-addressed callers)
         }
-        WorkspaceControl.PaneLocation loc = usesTabPlacement() ? spaces.locatePane(paneId) : null;
+        WorkspaceControl.PaneLocation loc = spaces.locatePane(paneId);
         try {
             agents.close(paneId);
         } catch (HerdrException e) {
@@ -985,11 +991,6 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
         }
     }
 
-    /** Whether any configured profile places peers in their own tab (so tabs may need cleanup). */
-    private boolean usesTabPlacement() {
-        return profiles.values().stream().anyMatch(FleetConfig.Profile::tabPlacement);
-    }
-
     /** True when a herdr error means the target is already gone (safe to treat as done). */
     private static boolean isAlreadyGone(HerdrException e) {
         return e.code() != null && e.code().endsWith("_not_found");

fleetd/src/main/java/dev/ltms/fleet/msg/AmqpReplyInbox.java

@@ -8,6 +8,7 @@ import com.rabbitmq.client.DeliverCallback;
 import com.rabbitmq.client.Recoverable;
 import com.rabbitmq.client.RecoveryListener;
 import com.rabbitmq.client.Return;
+import com.rabbitmq.client.impl.DefaultExceptionHandler;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
@@ -153,17 +154,22 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
     /** As {@link #open(String)}, with an explicit consumer prefetch (CB-527: caps the held backlog per target). */
     public static AmqpReplyInbox open(String uri, int prefetch) {
         try {
-            ConnectionFactory factory = new ConnectionFactory();
-            factory.setUri(uri);
-            // Self-heal transient blips; topology recovery re-declares queues and re-attaches consumers.
-            factory.setAutomaticRecoveryEnabled(true);
-            factory.setTopologyRecoveryEnabled(true);
-            return new AmqpReplyInbox(factory.newConnection("fleetd-reply-inbox"), prefetch);
+            return new AmqpReplyInbox(connectionFactory(uri).newConnection(AmqpConnectionFailureLogger.REPLY_INBOX), prefetch);
         } catch (Exception e) {
             throw new IllegalStateException("cannot connect to AMQP broker at " + uri, e);
         }
     }
 
+    static ConnectionFactory connectionFactory(String uri) throws Exception {
+        ConnectionFactory factory = new ConnectionFactory();
+        factory.setUri(uri);
+        // Self-heal transient blips; topology recovery re-declares queues and re-attaches consumers.
+        factory.setAutomaticRecoveryEnabled(true);
+        factory.setTopologyRecoveryEnabled(true);
+        factory.setExceptionHandler(new AmqpConnectionFailureLogger(AmqpConnectionFailureLogger.REPLY_INBOX, log));
+        return factory;
+    }
+
     /** Wrap an already-open connection with {@link #DEFAULT_PREFETCH} (injection seam for the contract test). */
     AmqpReplyInbox(Connection connection) {
         this(connection, DEFAULT_PREFETCH);
@@ -571,3 +577,44 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
         }
     }
 }
+
+/**
+ * Keeps RabbitMQ's forgiving exception behaviour while adding the connection identity that its
+ * default logger drops. Package-private so both AMQP connections use the same two names.
+ */
+final class AmqpConnectionFailureLogger extends DefaultExceptionHandler {
+
+    static final String REPLY_INBOX = "fleetd-reply-inbox";
+    static final String LEAD_MAILBOX = "fleetd-lead-mailbox";
+
+    private final String connectionName;
+    private final Logger logger;
+
+    AmqpConnectionFailureLogger(String connectionName, Logger logger) {
+        this.connectionName = connectionName;
+        this.logger = logger;
+    }
+
+    String connectionName() {
+        return connectionName;
+    }
+
+    @Override
+    protected void log(String message, Throwable cause) {
+        if (isSocketClosedOrConnectionReset(cause)) {
+            logger.warn("AMQP connection {}: {} (Exception message: {})", connectionName, message, cause.getMessage());
+        } else {
+            logger.error("AMQP connection {}: {}", connectionName, message, cause);
+        }
+    }
+
+    private static boolean isSocketClosedOrConnectionReset(Throwable cause) {
+        // Deliberate copy of ForgivingExceptionHandler's private static helper; check it on amqp-client upgrades.
+        if (!(cause instanceof IOException)) {
+            return false;
+        }
+        return "Connection reset".equals(cause.getMessage())
+                || "Socket closed".equals(cause.getMessage())
+                || "Connection reset by peer".equals(cause.getMessage());
+    }
+}

fleetd/src/main/java/dev/ltms/fleet/msg/LeadMailbox.java

@@ -122,17 +122,22 @@ public final class LeadMailbox implements LeadChannel, AutoCloseable {
     /** As {@link #open(String, String)}, with an explicit consumer prefetch. */
     public static LeadMailbox open(String uri, String selfCoordId, int prefetch) {
         try {
-            ConnectionFactory factory = new ConnectionFactory();
-            factory.setUri(uri);
-            // Self-heal transient blips; topology recovery re-declares the queue and re-attaches the consumer.
-            factory.setAutomaticRecoveryEnabled(true);
-            factory.setTopologyRecoveryEnabled(true);
-            return new LeadMailbox(factory.newConnection("fleetd-lead-mailbox"), selfCoordId, prefetch);
+            return new LeadMailbox(connectionFactory(uri).newConnection(AmqpConnectionFailureLogger.LEAD_MAILBOX), selfCoordId, prefetch);
         } catch (Exception e) {
             throw new IllegalStateException("cannot connect to AMQP coordination broker at " + uri, e);
         }
     }
 
+    static ConnectionFactory connectionFactory(String uri) throws Exception {
+        ConnectionFactory factory = new ConnectionFactory();
+        factory.setUri(uri);
+        // Self-heal transient blips; topology recovery re-declares queues and re-attaches consumers.
+        factory.setAutomaticRecoveryEnabled(true);
+        factory.setTopologyRecoveryEnabled(true);
+        factory.setExceptionHandler(new AmqpConnectionFailureLogger(AmqpConnectionFailureLogger.LEAD_MAILBOX, log));
+        return factory;
+    }
+
     /** Wrap an already-open connection with {@link #DEFAULT_PREFETCH} (injection seam for tests). */
     LeadMailbox(Connection connection, String selfCoordId) {
         this(connection, selfCoordId, DEFAULT_PREFETCH);

fleetd/src/main/java/dev/ltms/fleet/msg/MessageService.java

@@ -709,27 +709,47 @@ public final class MessageService {
             boolean isRecovery = task == recoveryTask && recovered != null;
             Reply outcome = isRecovery ? recovered : new Reply(Outcome.WORKER_FAILED, reason);
             String turnId = task.turnId;
-            if (task.future.complete(outcome)) {
-                if (outcome.outcome() == Outcome.WORKER_FAILED) {
-                    asyncFailed = true;
+            // fleetd #335: completing THIS task's future must not depend on any other task's
+            // cleanup succeeding — every task in `matching` is owed its own outcome regardless of
+            // what happens below, so decide and record that before doing anything that can throw.
+            boolean completedHere = task.future.complete(outcome);
+            if (completedHere && outcome.outcome() == Outcome.WORKER_FAILED) {
+                asyncFailed = true;
+            }
+            try {
+                if (abandonCleanupHookForTest != null) {
+                    // Test-only (fleetd #335 site 1): see the field's own javadoc.
+                    abandonCleanupHookForTest.run();
                 }
-                if (turnId != null) {
-                    // #275: whether this task was swept out of ASKING or was already answered and
-                    // only waiting on its resumed turn's real reply (#137), nothing will ever
-                    // complete this turnId now — drop it from this class's own bookkeeping AND the
-                    // reverse-rendezvous itself, so hasAsyncQuestion(target) stops reporting a turn
-                    // that is actually done, and a late answer() sees it as lapsed rather than
-                    // resolving a question nothing is listening for any more.
-                    asyncTasksByTurn.remove(turnId, task);
-                    rendezvous.closeAsk(turnId);
+                if (completedHere) {
+                    if (turnId != null) {
+                        // #275: whether this task was swept out of ASKING or was already answered and
+                        // only waiting on its resumed turn's real reply (#137), nothing will ever
+                        // complete this turnId now — drop it from this class's own bookkeeping AND the
+                        // reverse-rendezvous itself, so hasAsyncQuestion(target) stops reporting a turn
+                        // that is actually done, and a late answer() sees it as lapsed rather than
+                        // resolving a question nothing is listening for any more.
+                        asyncTasksByTurn.remove(turnId, task);
+                        rendezvous.closeAsk(turnId);
+                    }
+                } else if (isRecovery) {
+                    // The recovered reply was already drained out of the inbox, but this task resolved
+                    // through another path (e.g. a concurrent reply() or a second abandon() racing this
+                    // one) between us choosing it and completing it here. Put the reply back rather than
+                    // lose it silently — it may still belong to some other still-open task, or the next
+                    // caller that drains this target's inbox.
+                    inbox.publish(target, UUID.randomUUID().toString(), recovered.text());
                 }
-            } else if (isRecovery) {
-                // The recovered reply was already drained out of the inbox, but this task resolved
-                // through another path (e.g. a concurrent reply() or a second abandon() racing this
-                // one) between us choosing it and completing it here. Put the reply back rather than
-                // lose it silently — it may still belong to some other still-open task, or the next
-                // caller that drains this target's inbox.
-                inbox.publish(target, UUID.randomUUID().toString(), recovered.text());
+            } catch (RuntimeException e) {
+                // fleetd #335: inbox.publish reaches a broker (AmqpReplyInbox throws
+                // IllegalStateException on an unroutable/unconfirmed/interrupted publish) and this
+                // loop has no other teardown path — a caller on the release path, or the health
+                // monitor's GONE/NEVER_READY sweep. Losing this exception uncaught would abort the
+                // loop and leave every task still to come in `matching` PENDING forever (fleetd
+                // #335 site 1). completedHere is already recorded above, so only this task's
+                // best-effort bookkeeping is lost — log it and let the loop reach the rest.
+                log.error("abandon: per-task cleanup failed for ticket {} (target {}, turnId {})",
+                        task.ticket, target, turnId, e);
             }
         }
         if (failed) {
@@ -945,13 +965,41 @@ public final class MessageService {
             return new AskResult(AskOutcome.ANSWERED, answer);
         } catch (TimeoutException e) {
             log.debug("fleet_ask from {} went unanswered in {}ms", workerSession, timeoutMillis);
-            // fleetd #307: mark the task BEFORE clearAsyncQuestion(forgetTurn=true) below drops it out of
-            // asyncTasksByTurn and nulls its turnId — that forgetting is deliberate and stays (it is
-            // what keeps the target from staying BUSY forever), but it would otherwise also erase
-            // askAnsweredAsyncTasks' only signal that the worker's eventual real fleet_reply still
-            // belongs to this task, stranding it in the inbox with a false "never replied" verdict.
-            markAskTimedOut(ticket.turnId());
-            clearAsyncQuestion(ticket.turnId(), true);
+            // Only the fresh owner tears down the shared turn (mirrors the finally block below).
+            // A duplicate's own timeoutMillis says nothing about whether the SHARED ask is actually
+            // done — it must leave the close/forget bookkeeping to the fresh owner, exactly as it
+            // already leaves closeAsk to it.
+            if (ticket.fresh()) {
+                // fleetd #334: close the ask turn BEFORE forgetting this task's turnId mapping below.
+                // Before this fix the order was reversed — the mapping was forgotten here first, and
+                // rendezvous.closeAsk only ran afterward, in the shared finally. A primary's answer()
+                // call racing this exact timeout could then find rendezvous.askSession(turnId) still
+                // non-null (the ask still "answerable") after the Task mapping was already gone:
+                // answer()'s own asyncTasksByTurn lookup returned null, its task != null guard skipped
+                // the completion, and the async ticket sat at PENDING forever even though answer()
+                // itself reported the worker's real reply. Closing here first removes that window:
+                // any answer() call that still observes askSession(turnId) != null is necessarily
+                // racing a point BEFORE the forgetting below runs (both happen on this one thread, in
+                // this order, with nothing that yields in between), so the Task mapping is still there
+                // for it to find; any call that observes askSession(turnId) == null now correctly
+                // bails out STALE_TURN (see answer()'s own top check) before ever reaching
+                // asyncTasksByTurn. rendezvous.closeAsk is idempotent — a no-op once the turn is
+                // already removed, see its own javadoc — so the shared finally below re-running it
+                // for this same fresh call is harmless.
+                rendezvous.closeAsk(ticket.turnId());
+                if (askTimeoutRaceHookForTest != null) {
+                    // Test-only (fleetd #334): see the field's own javadoc.
+                    askTimeoutRaceHookForTest.run();
+                }
+                // fleetd #307: mark the task BEFORE clearAsyncQuestion(forgetTurn=true) below drops it
+                // out of asyncTasksByTurn and nulls its turnId — that forgetting is deliberate and
+                // stays (it is what keeps the target from staying BUSY forever), but it would
+                // otherwise also erase askAnsweredAsyncTasks' only signal that the worker's eventual
+                // real fleet_reply still belongs to this task, stranding it in the inbox with a false
+                // "never replied" verdict.
+                markAskTimedOut(ticket.turnId());
+                clearAsyncQuestion(ticket.turnId(), true);
+            }
             return new AskResult(AskOutcome.TIMED_OUT, null);
         } catch (ExecutionException e) {
             Throwable cause = e.getCause();
@@ -1050,19 +1098,28 @@ public final class MessageService {
                 // the chained ask deliberately left open.
                 //
                 // A null task is NOT only "this was never an async ticket". That reading was in this
-                // comment when #329 merged and it is wrong. A genuine async ticket also lands here
-                // with task == null, because ask()'s timeout path runs clearAsyncQuestion(turnId,
-                // true) — which drops the asyncTasksByTurn entry — in its catch block, while
-                // rendezvous.closeAsk(turnId) runs later, in its finally. Between those two the ask
-                // is still answerable but the map entry is already gone, so the lookup at :991
-                // returns null and this ticket is never completed. Measured on 2026-09-04: a probe
-                // firing only that first half before answer() runs printed
-                // "answer=REPLIED phase=PENDING reply=null" — the same stranded ticket #329 set out
-                // to fix, one step earlier in the same race. The probe used forgetTurnForTest, which
-                // omits ask()'s markAskTimedOut; that cannot change the outcome, because askTimedOut
-                // is read only by askAnsweredAsyncTasks, and reply() never reaches it while this
-                // method's own waiter is live. So #329 narrows this window rather than closing it.
-                // Open as fleetd #334 — do not read this guard as complete.
+                // comment when #329 merged and it is wrong; it is still not the whole story after
+                // #334. A genuine async ticket can still land here with task == null — a blocking
+                // (wait:true) send's fleet_ask never has a Task at all, so that case is expected and
+                // fine. What #334 fixed was a SECOND, unintended way to get here with task == null:
+                // ask()'s timeout path used to run clearAsyncQuestion(turnId, true) — which drops the
+                // asyncTasksByTurn entry — in its catch block, while rendezvous.closeAsk(turnId) ran
+                // later, in its finally. Between those two the ask was still answerable but the map
+                // entry was already gone, so the lookup at :1053 returned null and this ticket was
+                // never completed. Measured on 2026-09-04: a probe firing only that first half before
+                // answer() ran printed "answer=REPLIED phase=PENDING reply=null" — the same stranded
+                // ticket #329 set out to fix, one step earlier in the same race; the probe used
+                // forgetTurnForTest, which omits ask()'s markAskTimedOut, and that omission does not
+                // change the outcome, because askTimedOut is read only by askAnsweredAsyncTasks, and
+                // reply() never reaches it while this method's own waiter is live. #334's fix
+                // reorders ask()'s timeout catch to run closeAsk before the forgetting (see the
+                // fresh-owner block there), which removes this path entirely rather than narrowing
+                // it further: once closeAsk has run, rendezvous.askSession(turnId) is null and
+                // answer() returns STALE_TURN from its own top check, before it ever reaches this
+                // lookup — see aLateAnswerDuringAskTimeoutTeardownStillCompletesTheAsyncTicket in
+                // MessageServiceTest, which pins the exact window with askTimeoutRaceHookForTest.
+                // So by the time this line runs, task == null means only the ordinary blocking-send
+                // case (or #329's own already-fixed race elsewhere) — not this one.
                 if (result.outcome() != Outcome.QUESTION && task != null) {
                     finishAsyncTask(task, result);
                 }
@@ -1122,7 +1179,18 @@ public final class MessageService {
             // class javadoc on sendAsync/CB-107.
             task.future.whenComplete((reply, ex) -> {
                 boolean failed = ex != null || reply == null || !reply.completed();
-                pushLoop.onTicketTerminal(ticket, target, failed);
+                try {
+                    pushLoop.onTicketTerminal(ticket, target, failed);
+                } catch (Throwable t) {
+                    // fleetd #335 (site 2): this stage's own CompletableFuture is discarded, so an
+                    // uncaught throw here (e.g. a RejectedExecutionException from
+                    // ReplyPushLoop's scheduler, already shut down while this in-flight send's
+                    // whenComplete fires during the daemon's own shutdown sequence — messages.close()
+                    // only stops accepting NEW async work, it does not cancel a delivery already
+                    // running) vanishes with no log line and no metric, and the push loop never learns
+                    // the ticket went terminal — the exact thing this hook exists to tell it.
+                    log.error("push loop failed to learn ticket {} (target {}) went terminal", ticket, target, t);
+                }
             });
         }
         asyncExecutor.submit(() -> {
@@ -1421,6 +1489,56 @@ public final class MessageService {
         clearAsyncQuestion(turnId, true);
     }
 
+    /**
+     * Null in production; test seam for fleetd #335 (site 1) — invoked from {@link #abandon(String,
+     * String, boolean)}'s per-task loop, once per task, right before that task's own cleanup
+     * (turnId bookkeeping, or the stranded-reply put-back) runs. A test installs this to inject a
+     * throw at that exact point deterministically.
+     *
+     * <p>The one production call there that can really throw is {@code inbox.publish} in the
+     * put-back branch — {@link AmqpReplyInbox#publish} reaches a broker and throws {@link
+     * IllegalStateException} on an unroutable, unconfirmed, or interrupted publish — but reaching
+     * that branch requires a second completion of the very task {@code abandon} is about to
+     * complete to win the race first (see the branch's own comment), and the #137 follow-up
+     * investigation above already found the combination this needs (a stranded reply coinciding
+     * with an open matching task) unreachable through the public API, not merely hard to time.
+     * This hook reproduces the resulting shape — a per-task cleanup throw — directly, the same
+     * technique {@link #finishAsyncTaskRaceHook} and {@link
+     * #afterFinishAsyncTaskCompleteHookForTest} already use for their own hard-to-time races.
+     */
+    private volatile Runnable abandonCleanupHookForTest;
+
+    /**
+     * Test-only (fleetd #335, site 1): install {@link #abandonCleanupHookForTest}. Package-private
+     * so the test, in the same package, can reach it without widening any production API.
+     */
+    void setAbandonCleanupHookForTest(Runnable hook) {
+        this.abandonCleanupHookForTest = hook;
+    }
+
+    /**
+     * Null in production; test seam for fleetd #334 — invoked from {@link #ask}'s {@code
+     * TimeoutException} catch, only for the fresh owner, right after {@code rendezvous.closeAsk}
+     * has run and before {@link #markAskTimedOut} / {@link #clearAsyncQuestion} forget this task's
+     * turnId mapping. A test installs this to call {@link #answer} for the very same {@code turnId}
+     * synchronously from inside that exact window, deterministically reproducing the race a real
+     * concurrent {@code answer()} call could otherwise only win by timing luck: with the ask already
+     * closed, that call must see {@code rendezvous.askSession(turnId) == null} and return {@link
+     * Outcome#STALE_TURN} immediately, never reaching {@code asyncTasksByTurn} at all — proving the
+     * window fleetd #334 describes (mapping forgotten while the ask was still "answerable") is
+     * closed, rather than merely narrowed the way fleetd #329 narrowed the sibling race in {@link
+     * #finishAsyncTask}.
+     */
+    private volatile Runnable askTimeoutRaceHookForTest;
+
+    /**
+     * Test-only (fleetd #334): install {@link #askTimeoutRaceHookForTest}. Package-private so the
+     * test, in the same package, can reach it without widening any production API.
+     */
+    void setAskTimeoutRaceHookForTest(Runnable hook) {
+        this.askTimeoutRaceHookForTest = hook;
+    }
+
     /** A new send must not open a waiter while an async ticket owns this worker's paused turn. */
     private boolean hasAsyncQuestion(String target) {
         return asyncTasksByTurn.values().stream().anyMatch(task -> target.equals(task.target));

fleetd/src/test/java/dev/ltms/fleet/config/FleetConfigWithDefaultsPreservesEveryComponentTest.java

@@ -0,0 +1,192 @@
+package dev.ltms.fleet.config;
+
+import org.junit.jupiter.api.Test;
+
+import java.lang.reflect.Constructor;
+import java.lang.reflect.RecordComponent;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Locale;
+import java.util.Map;
+import java.util.Objects;
+import java.util.Set;
+import java.util.TreeSet;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+
+/**
+ * Guards against a "defect factory" built into this file's own established pattern, found live
+ * while building the (parked) idle-sleep-guard PR: every time a component is added to
+ * {@link FleetConfig}, the record grows by one arg AND a new back-compat constructor is added at
+ * the OLD arity, so existing callers keep compiling. That is correct and required — see the
+ * constructor ladder just below the record header. But {@link #withDefaults()}'s own {@code return
+ * new FleetConfig(...)} call sits in this same file, written at a literal argument count. The very
+ * next time a component is added, the freshly-added back-compat constructor at the OLD arity
+ * silently captures that stale call, because it is now a legal overload at that arg count too. It
+ * compiles. Every other test passes, because nothing else exercises the new field. The new
+ * component is defaulted away — {@code null}, or whatever that back-compat overload defaults it to
+ * — on every {@link FleetConfig#load}. Measured, not theoretical: this exact sequence happened
+ * live when the {@code idleSleepGuard} component was added on a sibling branch; it was caught only
+ * because that branch's own new tests happened to assert on the new field's value.
+ *
+ * <p>This test proves the opposite property, and does it in a way that survives the next field
+ * being added without being rewritten: reflectively enumerate {@link FleetConfig}'s own record
+ * components (never a hardcoded count — the arity is exactly what changes over time), build one
+ * config through the true canonical constructor with a real, distinctive, non-null value in EVERY
+ * component (reusing the exact reflective-construction pattern
+ * {@link ConfigRefTopLevelReportingCoverageTest} already established for this file:
+ * {@code getDeclaredConstructor(exact record-component types)}, which resolves the canonical
+ * constructor by its true shape, not by binding to whichever overload happens to match arg count —
+ * the same way Jackson resolves it), call the real {@link FleetConfig#withDefaults()}, and assert
+ * every one of those values survives unchanged.
+ *
+ * <p>Why this is a valid check for every component, not just some: {@link #withDefaults()}'s own
+ * comments document that it only ever REPLACES a component when the incoming value is {@code null}
+ * (or blank, for {@code placement}) — {@code broker}/{@code primary}/{@code leadHeartbeat}/
+ * {@code configReload}/{@code coordinator}/{@code worktreeGroup}/{@code memberLoginShell} are left
+ * as-is unconditionally, and {@code bind}/{@code guard}/{@code lifecycle}/{@code auth}/
+ * {@code fleet}/{@code quarantineCooldownSeconds}/{@code memberCredentials}/{@code placement} are
+ * replaced only on null/blank input. A value that is never null or blank going in must therefore
+ * never change coming out, for every current component. No exclusion is needed today.
+ *
+ * <p>{@link #EXCLUDED_FROM_SURVIVAL_CHECK} exists anyway, kept deliberately empty and size-pinned
+ * by {@link #exclusionListSizeIsPinned()}: a future component that {@code withDefaults()} is
+ * <em>documented</em> to transform unconditionally (unlike every field today) would legitimately
+ * need one. Pinning the size at 0 means growing that set to make a failure go away is itself a
+ * visible diff to this test, not a silent one — a checker that can be silenced by adding to its
+ * own escape hatch is not a checker.
+ */
+class FleetConfigWithDefaultsPreservesEveryComponentTest {
+
+    private static final RecordComponent[] COMPONENTS = FleetConfig.class.getRecordComponents();
+
+    /** See the class javadoc — deliberately empty today; grow it only with a matching justification. */
+    private static final Set<String> EXCLUDED_FROM_SURVIVAL_CHECK = Set.of();
+
+    /** One real, distinctive, non-null (non-blank where blankness would mean "unset") value per component. */
+    private static Map<String, Object> baseValues() {
+        Map<String, Object> v = new LinkedHashMap<>();
+        v.put("bind", new FleetConfig.Bind("127.0.0.1", 8765));
+        v.put("herdrSocket", "~/.config/herdr/guard.sock");
+        v.put("memberHerdrSocket", "~/.config/herdr/member-guard.sock");
+        v.put("profiles", Map.of("sonnet", minimalProfile("sonnet")));
+        v.put("guard", new FleetConfig.Guard(List.of("host-guard")));
+        v.put("worktreeRoot", "/wt/guard");
+        v.put("lifecycle", new FleetConfig.Lifecycle(300, 5, 30, true));
+        v.put("spawnReadyTimeoutMs", 12_345);
+        v.put("spawnReadyPollMs", 234);
+        v.put("broker", new FleetConfig.Broker("amqp://guard", null, 7));
+        v.put("primary", new FleetConfig.Primary("term-guard", 4, 4000));
+        v.put("fleet", new FleetConfig.Fleet(
+                Map.of("opus", new FleetConfig.Leader("sonnet", "lead: opus-guard", 1, null, 10,
+                        "claude", null, null, null)),
+                Map.of(), Map.of(), Map.of(), Map.of(), "{role}: {profile} #{n}"));
+        v.put("leadHeartbeat", new FleetConfig.LeadHeartbeat(301, 61_000L, 4));
+        v.put("health", new FleetConfig.Health(true, 31, 601, 61, null));
+        v.put("placement", "round-robin");
+        v.put("auth", new FleetConfig.Auth("loopback-trust", null));
+        v.put("configReload", new FleetConfig.ConfigReload(true, 11));
+        v.put("quarantineCooldownSeconds", 1801);
+        v.put("memberCredentials", new FleetConfig.MemberCredentials(
+                FleetConfig.MemberCredentials.POLICY_DENY_BY_DEFAULT,
+                List.of("git"), List.of("git", "ssh"), null));
+        v.put("coordinator", new FleetConfig.Coordinator("amqp://coord-guard", null, "self-guard", 3));
+        v.put("worktreeGroup", "group-guard");
+        v.put("memberLoginShell", "/bin/zsh");
+        assertNamesMatchComponents(v);
+        return v;
+    }
+
+    /** A minimal, otherwise-null {@link FleetConfig.Profile} — just enough to name one in a map. */
+    private static FleetConfig.Profile minimalProfile(String name) {
+        return new FleetConfig.Profile(name, null, null, null, null, null, null, null, null, null,
+                null, null, null, null, null, null, null, null, null, null, null, null, null, null,
+                null, null);
+    }
+
+    /**
+     * Guards {@link #baseValues()} itself against drifting from the record's real shape — the same
+     * assurance {@link ConfigRefTopLevelReportingCoverageTest} already relies on. This is what makes
+     * "no hardcoded arity" true in practice: forgetting to add a new component here fails this
+     * assertion by name, rather than silently checking one component fewer than the record has.
+     */
+    private static void assertNamesMatchComponents(Map<String, Object> values) {
+        Set<String> names = new TreeSet<>();
+        for (RecordComponent rc : COMPONENTS) {
+            names.add(rc.getName());
+        }
+        assertEquals(names, new TreeSet<>(values.keySet()),
+                "this test's value map has drifted from FleetConfig's actual top-level components — "
+                        + "update baseValues() alongside the record");
+    }
+
+    /**
+     * Builds a {@link FleetConfig} through the TRUE canonical constructor — resolved by the record's
+     * own component types, not by argument count — so this never accidentally exercises a
+     * back-compat overload the way a literal {@code new FleetConfig(...)} call risks doing.
+     */
+    private static FleetConfig configOf(Map<String, Object> values) throws ReflectiveOperationException {
+        Class<?>[] types = Arrays.stream(COMPONENTS).map(RecordComponent::getType).toArray(Class<?>[]::new);
+        Object[] args = Arrays.stream(COMPONENTS).map(rc -> values.get(rc.getName())).toArray();
+        Constructor<FleetConfig> ctor = FleetConfig.class.getDeclaredConstructor(types);
+        return ctor.newInstance(args);
+    }
+
+    @Test
+    void exclusionListSizeIsPinned() {
+        assertEquals(0, EXCLUDED_FROM_SURVIVAL_CHECK.size(),
+                "EXCLUDED_FROM_SURVIVAL_CHECK grew from 0 — every entry needs a justification in "
+                        + "this test class's javadoc AND this assertion re-pinned to the new size; a "
+                        + "growing exclusion list that silences failures on its own is not a guard");
+    }
+
+    /**
+     * The mutation this is built to catch: make {@code withDefaults()}'s final constructor call
+     * literal at some arg count, add one more component to the record with a new back-compat
+     * constructor at the old arity, and the stale call silently rebinds. Every component here is
+     * real and non-null (non-blank for the one String — {@code placement} — where blank has
+     * meaning), so none of it should be replaced by {@code withDefaults()}; any component that
+     * comes back different was silently dropped.
+     */
+    @Test
+    void everyComponentGivenARealValueSurvivesWithDefaults() throws ReflectiveOperationException {
+        Map<String, Object> base = baseValues();
+        FleetConfig config = configOf(base);
+        FleetConfig defaulted = config.withDefaults();
+
+        List<String> dropped = new ArrayList<>();
+        int checked = 0;
+        for (RecordComponent rc : COMPONENTS) {
+            String name = rc.getName();
+            if (EXCLUDED_FROM_SURVIVAL_CHECK.contains(name)) {
+                continue;
+            }
+            checked++;
+            Object expected = base.get(name);
+            Object actual;
+            try {
+                actual = rc.getAccessor().invoke(defaulted);
+            } catch (ReflectiveOperationException e) {
+                throw new RuntimeException("failed to read FleetConfig." + name + "()", e);
+            }
+            if (!Objects.equals(expected, actual)) {
+                dropped.add(String.format(Locale.ROOT,
+                        "%s: withDefaults() was given a real, non-null value (%s) for '%s' but "
+                                + "returned %s — a component silently dropped by withDefaults(), the "
+                                + "shape of the defect this test exists to catch (its final "
+                                + "\"return new FleetConfig(...)\" call binding to a back-compat "
+                                + "constructor instead of the true canonical one)",
+                        name, expected, name, actual));
+            }
+        }
+
+        System.out.printf(Locale.ROOT,
+                "FleetConfig.withDefaults() component-survival coverage — %d components, %d checked, "
+                        + "%d excluded, %d survived%n",
+                COMPONENTS.length, checked, EXCLUDED_FROM_SURVIVAL_CHECK.size(), checked - dropped.size());
+        assertEquals(List.of(), dropped,
+                "withDefaults() silently dropped these real, given components: " + dropped);
+    }
+}

fleetd/src/test/java/dev/ltms/fleet/inject/CompletionResolverTest.java

@@ -484,6 +484,27 @@ class CompletionResolverTest {
 
     // --- CB-578 stage A: backend-exhausted classification ---------------------------------
 
+    @Test
+    void aNormalMemberReportMentioningTheExhaustionPatternDoesNotNotifyTheSink() {
+        String block = "⏺ I reviewed capacity handling. The usage limit has been reached means no more work can start.\n❯ ";
+        FakeHerdr herdr = new FakeHerdr().readText(block);
+        Rendezvous rendezvous = new Rendezvous();
+        ExhaustedPatternLookup patterns = target -> Pattern.compile("usage limit has been reached");
+        java.util.List<String> notified = new java.util.ArrayList<>();
+        ExhaustionSink sink = (target, reason, profile) -> notified.add(target + ": " + reason);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, patterns, sink);
+
+        var waiter = rendezvous.open("term_a");
+        resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
+
+        assertEquals(Rendezvous.Kind.BACKEND_EXHAUSTED, waiter.getNow(null).kind(),
+                "a matching report still fails the send as exhausted");
+        assertTrue(waiter.getNow(null).text().contains("I reviewed capacity handling."),
+                "the exhausted result keeps the whole matched pane line");
+        assertTrue(notified.isEmpty(),
+                "a normal report mentioning an exhaustion pattern must not quarantine a credential");
+    }
+
     @Test
     void classifiesAMatchingScrapeAsBackendExhaustedInsteadOfACompletedReply() {
         String block = "⏺ Working on it...\nThe usage limit has been reached. Try again later.\n❯ ";
@@ -534,6 +555,53 @@ class CompletionResolverTest {
                 "the sink is told the matched reason: " + notified.get(0));
     }
 
+    @Test
+    void aRealExhaustionBehindTerminalChromeStillNotifiesTheSink() {
+        String block = "⏺ │ The usage limit has been reached. Try again later.\n❯ ";
+        FakeHerdr herdr = new FakeHerdr().readText(block);
+        Rendezvous rendezvous = new Rendezvous();
+        ExhaustedPatternLookup patterns = target -> Pattern.compile("usage limit has been reached");
+        java.util.List<String> notified = new java.util.ArrayList<>();
+        ExhaustionSink sink = (target, reason, profile) -> notified.add(target + ": " + reason);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, patterns, sink);
+
+        var waiter = rendezvous.open("term_a");
+        resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
+
+        assertEquals(Rendezvous.Kind.BACKEND_EXHAUSTED, waiter.getNow(null).kind(),
+                "a real exhaustion must still fail the send as exhausted");
+        assertEquals(1, notified.size(),
+                "a real exhaustion behind terminal chrome must reach the sink");
+    }
+
+    /**
+     * The live fleet configures {@code exhaustedPattern: "The usage limit has been reached"} — with
+     * the leading {@code "The"}. Every other test here uses a pattern without it, which is the shape
+     * that made fleetd #348 need a looser rule than a start-of-line check. This pins the deployed
+     * shape as well, so a later tightening of {@link CompletionResolver} cannot silently stop
+     * recording the exhaustion this fleet actually reports.
+     *
+     * <p>What it does not prove: that this is the only pattern shape an operator will write.
+     */
+    @Test
+    void anExhaustionPatternCarryingItsLeadingWordsStillNotifiesTheSink() {
+        String block = "⏺ │ The usage limit has been reached. Try again later.\n❯ ";
+        FakeHerdr herdr = new FakeHerdr().readText(block);
+        Rendezvous rendezvous = new Rendezvous();
+        ExhaustedPatternLookup patterns = target -> Pattern.compile("The usage limit has been reached");
+        java.util.List<String> notified = new java.util.ArrayList<>();
+        ExhaustionSink sink = (target, reason, profile) -> notified.add(target + ": " + reason);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, patterns, sink);
+
+        var waiter = rendezvous.open("term_a");
+        resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
+
+        assertEquals(Rendezvous.Kind.BACKEND_EXHAUSTED, waiter.getNow(null).kind(),
+                "the live pattern shape must still fail the send as exhausted");
+        assertEquals(1, notified.size(),
+                "the live pattern shape must still reach the sink");
+    }
+
     @Test
     void aLosingBackendExhaustedClassificationNeverNotifiesTheExhaustionSink() {
         // The waiter was already resolved (e.g. by the worker's own reply) before this scrape landed —

fleetd/src/test/java/dev/ltms/fleet/member/CompositePeerLauncherTest.java

@@ -299,6 +299,41 @@ class CompositePeerLauncherTest {
                 "two adapter kinds sharing one daemon keep the fallback route");
     }
 
+    @Test
+    void stopThroughTheSingleDaemonShortcutStillClosesTheTabWhenTheFallbackDelegateUsesPanePlacement() {
+        // fleetd #342: the single-daemon shortcut (spawnedBy empty, herdrDaemonCount()==1) always
+        // routes stop() through delegates.getFirst() — here the claude adapter, configured for
+        // PANE placement (its own profiles never create a dedicated tab). The pane being torn
+        // down here actually belongs to the opencode adapter's TAB placement, sharing the same
+        // herdr daemon — mixing placements is the point: every existing stop-fallback test in this
+        // class configures BOTH adapters as "tab", so usesTabPlacement() was true either way and
+        // the mis-routing never showed.
+        //
+        // Before the fix, HerdrPeerLauncher#stop gated tab resolution on usesTabPlacement() of the
+        // delegate it happened to be called through, so the wrongly-routed (pane-placement) claude
+        // adapter never even looked for a tab to close, and the now-empty tab leaked with nothing
+        // to reap it. The fix (fleetd #342) resolves the pane's real tab unconditionally, so the
+        // decision follows the pane's actual placement rather than the fallback delegate's static
+        // config.
+        FakeHerdr herdr = new FakeHerdr();
+        FleetConfig.Profile claudePane = new FleetConfig.Profile("claude", "http://gx00.gw:8000",
+                "coder", null, "FLEETD_WORKER_TOKEN", List.of("claude"), "pane", "fleetd-workers",
+                "w #{n}", null, null, null);
+        ClaudeCodeLauncher claude = new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
+                new SubscriptionGuard(Set.of("gx00.gw")), Map.of("claude", claudePane), "claude", _ -> null);
+        PeerLauncher composite = new CompositePeerLauncher(List.of(claude, opencodeAdapter(herdr)), "claude");
+
+        // "w9:pW" was never spawned through this composite instance, so spawnedBy has no entry for
+        // it (the same in-memory-cache-miss shape a daemon restart leaves behind) — stop() falls
+        // through to the single-daemon shortcut and hands it to delegates.getFirst() (claude).
+        composite.stop("w9:pW");
+
+        assertTrue(herdr.called("pane.close"), "the pane itself is still closed on every routed path");
+        assertTrue(herdr.called("tab.close"),
+                "the pane's real (sole-occupant) tab must be closed even though the fallback routed "
+                        + "through a delegate configured for pane placement");
+    }
+
     @Test
     void listKeepsBothPanesWhenTwoDaemonsShareAPaneId() {
         // herdr pane ids are per-daemon counters, so two daemons really can both hold w1:p1 on

fleetd/src/test/java/dev/ltms/fleet/msg/AmqpConnectionFailureLoggerTest.java

@@ -0,0 +1,134 @@
+package dev.ltms.fleet.msg;
+
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
+import com.rabbitmq.client.Channel;
+import com.rabbitmq.client.ConnectionFactory;
+import com.rabbitmq.client.impl.DefaultExceptionHandler;
+import org.junit.jupiter.api.Test;
+import org.slf4j.LoggerFactory;
+
+import java.io.IOException;
+import java.lang.reflect.Proxy;
+import java.util.concurrent.atomic.AtomicInteger;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertFalse;
+import static org.junit.jupiter.api.Assertions.assertInstanceOf;
+import static org.junit.jupiter.api.Assertions.assertNotEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+class AmqpConnectionFailureLoggerTest {
+
+    @Test
+    void installedHandlersLogTheirOwnConnectionNamesAtErrorWithTheCause() throws Exception {
+        ConnectionFactory inboxFactory = AmqpReplyInbox.connectionFactory("amqp://127.0.0.1");
+        ConnectionFactory mailboxFactory = LeadMailbox.connectionFactory("amqp://127.0.0.1");
+
+        AmqpConnectionFailureLogger inboxHandler = installedStrictHandler(inboxFactory, "reply inbox");
+        AmqpConnectionFailureLogger mailboxHandler = installedStrictHandler(mailboxFactory, "lead mailbox");
+        assertEquals(AmqpConnectionFailureLogger.REPLY_INBOX, inboxHandler.connectionName());
+        assertEquals(AmqpConnectionFailureLogger.LEAD_MAILBOX, mailboxHandler.connectionName());
+
+        ListAppender<ILoggingEvent> inboxEvents = attach(AmqpReplyInbox.class);
+        ListAppender<ILoggingEvent> mailboxEvents = attach(LeadMailbox.class);
+        IllegalStateException inboxFailure = new IllegalStateException("inbox failure");
+        IllegalStateException mailboxFailure = new IllegalStateException("mailbox failure");
+        try {
+            inboxHandler.handleUnexpectedConnectionDriverException(null, inboxFailure);
+            mailboxHandler.handleConnectionRecoveryException(null, mailboxFailure);
+
+            assertError(inboxEvents, "AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred",
+                    inboxFailure, "inbox failure line");
+            assertError(mailboxEvents, "AMQP connection fleetd-lead-mailbox: Caught an exception during connection recovery!",
+                    mailboxFailure, "mailbox recovery line");
+        } finally {
+            detach(AmqpReplyInbox.class, inboxEvents);
+            detach(LeadMailbox.class, mailboxEvents);
+        }
+    }
+
+    @Test
+    void connectionResetKeepsForgivingHandlerWarningSemantics() {
+        AmqpConnectionFailureLogger handler = new AmqpConnectionFailureLogger(
+                AmqpConnectionFailureLogger.REPLY_INBOX, LoggerFactory.getLogger(AmqpReplyInbox.class));
+        ListAppender<ILoggingEvent> events = attach(AmqpReplyInbox.class);
+        try {
+            handler.handleUnexpectedConnectionDriverException(null, new IOException("Connection reset"));
+            assertEquals(1, events.list.size(), "the handler must still log a reset");
+            ILoggingEvent event = events.list.getFirst();
+            assertEquals(Level.WARN, event.getLevel(), "ForgivingExceptionHandler logs connection resets at WARN");
+            assertEquals("AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred "
+                            + "(Exception message: Connection reset)", event.getFormattedMessage());
+            assertTrue(event.getThrowableProxy() == null, "ForgivingExceptionHandler does not attach a reset stack trace");
+        } finally {
+            detach(AmqpReplyInbox.class, events);
+        }
+    }
+
+    @Test
+    void connectionNamesStayDistinct() {
+        assertNotEquals(AmqpConnectionFailureLogger.REPLY_INBOX, AmqpConnectionFailureLogger.LEAD_MAILBOX,
+                "reply-inbox and lead-mailbox failures must be distinguishable");
+    }
+
+    @Test
+    void strictConsumerExceptionStillClosesItsChannel() {
+        AtomicInteger closes = new AtomicInteger();
+        Channel channel = (Channel) Proxy.newProxyInstance(getClass().getClassLoader(), new Class<?>[] {Channel.class},
+                (_, method, _) -> switch (method.getName()) {
+                    case "close" -> {
+                        closes.incrementAndGet();
+                        yield null;
+                    }
+                    case "toString" -> "test-channel";
+                    default -> throw new UnsupportedOperationException(method.getName());
+                });
+        AmqpConnectionFailureLogger handler = new AmqpConnectionFailureLogger(
+                AmqpConnectionFailureLogger.REPLY_INBOX, LoggerFactory.getLogger(AmqpReplyInbox.class));
+
+        handler.handleConsumerException(channel, new IllegalStateException("consumer failed"), null, "tag", "handleDelivery");
+
+        assertEquals(1, closes.get(), "DefaultExceptionHandler must close a channel after a consumer exception");
+    }
+
+    @Test
+    void handlerOnlyChangesDefaultHandlerLogging() {
+        assertEquals(DefaultExceptionHandler.class,
+                AmqpConnectionFailureLogger.class.getSuperclass());
+        assertFalse(java.util.Arrays.stream(AmqpConnectionFailureLogger.class.getDeclaredMethods())
+                        .anyMatch(method -> method.getName().startsWith("handle")),
+                "all exception-handling methods must remain inherited from DefaultExceptionHandler");
+    }
+
+    private static ListAppender<ILoggingEvent> attach(Class<?> owner) {
+        Logger logger = (Logger) LoggerFactory.getLogger(owner);
+        logger.setLevel(Level.DEBUG);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        logger.addAppender(appender);
+        return appender;
+    }
+
+    private static void detach(Class<?> owner, ListAppender<ILoggingEvent> appender) {
+        ((Logger) LoggerFactory.getLogger(owner)).detachAppender(appender);
+    }
+
+    private static AmqpConnectionFailureLogger installedStrictHandler(ConnectionFactory factory, String connection) {
+        assertInstanceOf(DefaultExceptionHandler.class, factory.getExceptionHandler(),
+                connection + " must keep DefaultExceptionHandler: replacing the strict handler with a forgiving one "
+                        + "changes when a channel is closed");
+        return assertInstanceOf(AmqpConnectionFailureLogger.class, factory.getExceptionHandler());
+    }
+
+    private static void assertError(ListAppender<ILoggingEvent> events, String message, Throwable cause, String name) {
+        assertEquals(1, events.list.size(), name);
+        ILoggingEvent event = events.list.getFirst();
+        assertEquals(Level.ERROR, event.getLevel(), name);
+        assertEquals(message, event.getFormattedMessage(), name);
+        assertEquals(cause.toString(), event.getThrowableProxy().getClassName() + ": "
+                + event.getThrowableProxy().getMessage(), name);
+    }
+}

fleetd/src/test/java/dev/ltms/fleet/msg/MessageServiceTest.java

@@ -377,6 +377,126 @@ class MessageServiceTest {
         assertEquals(MessageService.Outcome.QUESTION, q.outcome());
     }
 
+    /**
+     * fleetd #334. {@code ask()}'s {@code TimeoutException} catch used to forget this task's
+     * {@code turnId} mapping ({@code clearAsyncQuestion(turnId, true)}) BEFORE closing the ask
+     * ({@code rendezvous.closeAsk}, in the shared {@code finally}). A primary's {@code answer()}
+     * call racing that exact window found the ask still "answerable" ({@code
+     * rendezvous.askSession(turnId)} still non-null) while the {@code Task} was already forgotten,
+     * so its {@code task != null} guard skipped the completion and the async ticket sat at
+     * {@code PENDING} forever even though {@code answer()} itself reported a result. The fix
+     * (closing the ask first) makes this window impossible: a racing {@code answer()} call either
+     * still finds the ask open (and the {@code Task} mapping guaranteed intact) or finds it already
+     * closed (and bails {@code STALE_TURN} before ever touching the {@code Task}). This test pins
+     * the exact window with {@code askTimeoutRaceHookForTest} and proves both invariants the ticket
+     * named: (1) a late/racing {@code answer()} sees the ask as already lapsed ({@code STALE_TURN}),
+     * never made answerable again, and (2) the async ticket still resolves {@code DONE} once the
+     * worker's real {@code fleet_reply} lands — it is never stranded {@code PENDING}.
+     */
+    /**
+     * fleetd #334 gated the ask-timeout teardown on {@code ticket.fresh()}, matching the {@code
+     * finally} block that already did. This pins that gate. A coalesced duplicate passes its own
+     * {@code timeoutMillis}, which says nothing about whether the shared ask is done — so a
+     * duplicate timing out first must leave the fresh owner's still-open ask answerable.
+     *
+     * <p>Measured on merge: without this test, removing the {@code ticket.fresh()} gate left all
+     * 1371 tests green. The gate shipped with the reorder and nothing held it there.
+     *
+     * <p>What this does not prove: anything about the ordering inside the gate — that is
+     * {@code aLateAnswerDuringAskTimeoutTeardownStillCompletesTheAsyncTicket}'s job.
+     */
+    @Test
+    void aCoalescedDuplicateAskTimingOutLeavesTheFreshOwnersAskOpen() throws Exception {
+        String ticket = messages.sendAsync(T, "long task");
+        awaitWaiting();
+        injector.onStatus(T, AgentStatus.IDLE);    // deliver
+        injector.onStatus(T, AgentStatus.WORKING); // worker picks it up, then pauses to ask
+
+        CompletableFuture<MessageService.AskResult> fresh =
+                CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
+
+        MessageService.TaskView asking = null;
+        long deadline = System.currentTimeMillis() + 2000;
+        while ((asking == null || asking.phase() != MessageService.Phase.ASKING)
+                && System.currentTimeMillis() < deadline) {
+            asking = messages.poll(ticket);
+            //noinspection BusyWait
+            Thread.sleep(5);
+        }
+        assertNotNull(asking, "the fresh owner's question must surface before the duplicate asks");
+        String turnId = asking.turnId();
+        assertNotNull(turnId, "an ASKING view carries the turnId to answer on");
+
+        // A coalesced duplicate on the same session, with its own much shorter timeout.
+        MessageService.AskResult duplicate = messages.ask(T, "which config file?", 100);
+        assertEquals(MessageService.AskOutcome.TIMED_OUT, duplicate.outcome(),
+                "the duplicate's own timeout elapses first");
+
+        CompletableFuture<MessageService.Reply> answered =
+                CompletableFuture.supplyAsync(() -> messages.answer(turnId, "fleetd.yaml", 500));
+
+        MessageService.AskResult a = fresh.get(5, TimeUnit.SECONDS);
+        assertEquals(MessageService.AskOutcome.ANSWERED, a.outcome(),
+                "a duplicate's timeout must not lapse the ask the fresh owner still holds");
+        assertEquals("fleetd.yaml", a.answer());
+        assertFalse(answered.get(5, TimeUnit.SECONDS).outcome() == MessageService.Outcome.STALE_TURN,
+                "the answer must not be rejected as stale");
+    }
+
+    @Test
+    void aLateAnswerDuringAskTimeoutTeardownStillCompletesTheAsyncTicket() throws Exception {
+        String ticket = messages.sendAsync(T, "long task");
+        awaitWaiting();
+        injector.onStatus(T, AgentStatus.IDLE);    // deliver
+        injector.onStatus(T, AgentStatus.WORKING); // worker picks it up, then pauses to ask
+
+        CompletableFuture<MessageService.AskResult> ask =
+                CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 200));
+
+        // Wait for the question to actually surface (poll sees ASKING) before racing the timeout.
+        MessageService.TaskView asking = null;
+        long deadline = System.currentTimeMillis() + 2000;
+        while ((asking == null || asking.phase() != MessageService.Phase.ASKING)
+                && System.currentTimeMillis() < deadline) {
+            asking = messages.poll(ticket);
+            //noinspection BusyWait
+            Thread.sleep(5);
+        }
+        assertNotNull(asking, "the question must surface before the ask times out");
+        String turnId = asking.turnId();
+        assertNotNull(turnId, "an ASKING view carries the turnId to answer on");
+
+        CompletableFuture<MessageService.Reply> lateAnswer = new CompletableFuture<>();
+        messages.setAskTimeoutRaceHookForTest(() ->
+                lateAnswer.complete(messages.answer(turnId, "too late", 500)));
+        try {
+            MessageService.AskResult a = ask.get(5, TimeUnit.SECONDS);
+            assertEquals(MessageService.AskOutcome.TIMED_OUT, a.outcome());
+
+            MessageService.Reply late = lateAnswer.get(5, TimeUnit.SECONDS);
+            assertEquals(MessageService.Outcome.STALE_TURN, late.outcome(),
+                    "a late answer racing the timeout teardown must see the ask as already lapsed");
+
+            // The worker resumes on its own (per the ask() contract) and eventually sends its real
+            // fleet_reply; the async ticket must still resolve with it, not strand at PENDING.
+            assertTrue(messages.reply(T, "real result"), "the worker's real reply must still be accepted");
+        } finally {
+            messages.setAskTimeoutRaceHookForTest(null);
+        }
+
+        MessageService.TaskView done = null;
+        deadline = System.currentTimeMillis() + 2000;
+        while ((done == null || done.phase() == MessageService.Phase.PENDING)
+                && System.currentTimeMillis() < deadline) {
+            done = messages.poll(ticket);
+            //noinspection BusyWait
+            Thread.sleep(5);
+        }
+        assertNotNull(done);
+        assertEquals(MessageService.Phase.DONE, done.phase(), "the async ticket must not be stranded PENDING");
+        assertEquals("real result", done.reply());
+    }
+
     @Test
     void answeringAnUnknownTurnIsStale() {
         MessageService.Reply r = messages.answer(T + "#999", "too late", 500);
@@ -808,6 +928,49 @@ class MessageServiceTest {
         assertFailedTicket(third, "agent target term_a not found");
     }
 
+    // --- fleetd #335 (site 1): a per-task cleanup failure inside the abandon() loop must not -----
+    // strand the tasks that come after it. abandon()'s own comment on the loop documents the one
+    // real production call that can throw there (inbox.publish, in the stranded-reply put-back
+    // branch, reached when a concurrent reply() or a second abandon() races this one) — but
+    // reaching that branch requires the exact combination the #137 follow-up above already found
+    // unreachable through the public API. abandonCleanupHookForTest reproduces the resulting SHAPE
+    // (one task's cleanup throws) directly instead, the same technique this file already uses for
+    // fleetd #324/#329's own hard-to-time races.
+    @Test
+    void aPerTaskCleanupFailureDoesNotStrandTheRemainingMatchingTasks() throws Exception {
+        ListAppender<ILoggingEvent> appender = attachMessageServiceLog();
+        try {
+            String first = messages.sendAsync(T, "first task");
+            awaitWaiting(); // first task owns the target lock and rendezvous waiter
+            String second = messages.sendAsync(T, "second task"); // parked on the same lock
+            String third = messages.sendAsync(T, "third task"); // parked too — the whole sweep must survive
+
+            java.util.concurrent.atomic.AtomicInteger calls = new java.util.concurrent.atomic.AtomicInteger();
+            messages.setAbandonCleanupHookForTest(() -> {
+                if (calls.getAndIncrement() == 0) {
+                    throw new RuntimeException("PROBE-335-SITE1");
+                }
+            });
+
+            assertTrue(messages.abandon(T, "agent target term_a not found"));
+
+            // Every task in the loop still gets its own outcome — the one whose cleanup threw
+            // included — even though the loop had no way to know in advance which one that would be.
+            assertFailedTicket(first, "agent target term_a not found");
+            assertFailedTicket(second, "agent target term_a not found");
+            assertFailedTicket(third, "agent target term_a not found");
+
+            assertTrue(appender.list.stream().anyMatch(e ->
+                            e.getLevel() == Level.ERROR
+                                    && e.getThrowableProxy() != null
+                                    && "PROBE-335-SITE1".equals(e.getThrowableProxy().getMessage())),
+                    "a per-task cleanup failure must still reach the log, not vanish silently");
+        } finally {
+            messages.setAbandonCleanupHookForTest(null);
+            detachMessageServiceLog(appender);
+        }
+    }
+
     // --- #137 follow-up: abandon() must not guess when more than one task is open ---------------
     //
     // A test combining a genuine stranded reply (hasStrandedReply(T)==true) with two simultaneously
@@ -1495,6 +1658,42 @@ class MessageServiceTest {
         }
     }
 
+    // --- fleetd #335 (site 2): task.future.whenComplete's own returned stage is discarded, so an --
+    // uncaught throw from ReplyPushLoop.onTicketTerminal used to vanish with no log line and no
+    // metric. The real production trigger is the daemon's own shutdown sequence (Fleetd's shutdown
+    // hook): messages.close() only stops the async executor from taking NEW work — it does not
+    // cancel a send already in flight — while pushLoop.close() shuts its scheduler down immediately
+    // right after, so a ticket that completes in that narrow window has onTicketTerminal's own
+    // scheduler.schedule(...) throw a real RejectedExecutionException. Reproduced here by shutting
+    // the very same scheduler down before the ticket resolves — no test-only hook needed, this
+    // reachable path throws for real.
+    @Test
+    void aTicketTerminalPushFailureDoesNotVanishSilently() throws Exception {
+        ListAppender<ILoggingEvent> appender = attachMessageServiceLog();
+        try (var wiring = wireWithPushLoop(1, 50)) {
+            String ticket = wiring.service().sendAsync(T, "long task");
+            awaitWaiting();
+            wiring.scheduler().shutdownNow(); // simulate pushLoop.close() racing an in-flight send
+            injectDelivery();
+            assertTrue(rendezvous.resolve(T, "async result"));
+
+            // The ticket's own outcome must be unaffected by the swallowed exception — finishAsyncTask
+            // completes task.future before whenComplete's action (and thus onTicketTerminal) ever runs.
+            MessageService.TaskView done = awaitTicketPhaseOn(wiring.service(), ticket, MessageService.Phase.DONE);
+            assertEquals("async result", done.reply());
+
+            assertTrue(appender.list.stream().anyMatch(e ->
+                            e.getLevel() == Level.ERROR
+                                    && e.getFormattedMessage().contains(ticket)
+                                    && e.getThrowableProxy() != null
+                                    && "java.util.concurrent.RejectedExecutionException"
+                                            .equals(e.getThrowableProxy().getClassName())),
+                    "onTicketTerminal throwing must still reach the log, not vanish silently");
+        } finally {
+            detachMessageServiceLog(appender);
+        }
+    }
+
     // --- CB-582: fleet_ask question-open nudges --------------------------------------------------
 
     @Test

fleetd/src/test/java/dev/ltms/fleet/rest/FleetAppTest.java

@@ -711,13 +711,21 @@ class FleetAppTest {
 
     @Test
     void stopWorkerInPanePlacementClosesOnlyThePane() throws Exception {
-        FakeHerdr herdr = new FakeHerdr();
+        // fleetd #342: tab-cleanup resolution is no longer skipped based on a profile's declared
+        // placement — a stop() routed through the wrong delegate (e.g. CompositePeerLauncher's
+        // single-daemon shortcut after a daemon restart) could carry a placement config that says
+        // nothing true about how THIS pane was actually placed. So the pane's real tab is now
+        // always resolved, and the single-occupant check below is what protects a pane-placement
+        // peer's shared tab, exactly as it always protected a tab-placement one. Model that
+        // realistically: the peer's pane was split into an existing tab that already held another
+        // occupant, so the tab must never be closed.
+        FakeHerdr herdr = new FakeHerdr().withWorkerTabPaneCount(2);
         int port = start(herdr, "http://gx00.gw:8000", Set.of("gx00.gw"), "pane");
 
         assertEquals(204, req(port, "DELETE", "/members/w9:pW").statusCode());
         assertTrue(herdr.called("pane.close"));
-        assertFalse(herdr.called("tab.close"), "pane placement owns no tab to close");
-        assertFalse(herdr.called("pane.get"), "no tab resolution in pane placement");
+        assertTrue(herdr.called("pane.get"), "tab resolution now always runs, regardless of placement");
+        assertFalse(herdr.called("tab.close"), "pane placement's shared tab must never be closed");
     }
 
     @Test

plans/fleet-plugin/plan.md

@@ -0,0 +1,337 @@
+# Fleet as a Claude Code plugin — plan
+
+Status: draft for architect review. Not implemented.
+Author: primary (lead `opus`, Mac fleet). Date: 2026-09-05.
+
+## 0. Correction — this already exists, and that changes the plan
+
+I wrote sections below as if the plugin were new work. It is not. **This repo is already a Claude
+Code marketplace and already ships a plugin**, added in `ef1e014` (CB-527) and last touched in
+`2e138a1` (CB-634):
+
+```
+.claude-plugin/marketplace.json          -> name "claude-bridge", plugins: [ ./plugin ]
+plugin/.claude-plugin/plugin.json        -> name "claude-bridge", version 0.1.0
+plugin/.mcp.json                         -> mounts "fleetd" at http://127.0.0.1:8765/mcp
+plugin/skills/setup/SKILL.md             -> a full onboarding skill
+plugin/README.md
+```
+
+The `setup` skill is good and covers most of what section 5 proposes: preflight, merge-not-clobber
+into `.mcp.json`, read-only permissions only, credentials by env-var name, and a verify step that
+insists on a **real spawn** because a green `/healthz` proves nothing.
+
+So the operator's question — "can we pack things into plugins?" — is already answered *yes, and it
+was built*. The real question is why it did nothing for the kb session. The answer is drift plus
+invisibility.
+
+### The drift, measured
+
+| # | Finding | Evidence |
+|---|---|---|
+| 1 | **Mount name mismatch.** The plugin mounts the server as `fleetd`; the daemon's own constant is `fleet` | `plugin/.mcp.json` vs `PeerLauncher.java:34` `String MCP_MOUNT_NAME = "fleet"` |
+| 2 | **URL hardcoded**, no env indirection, so one plugin cannot serve two hosts or ports | `plugin/.mcp.json` |
+| 3 | **Ships no worker skills and no agents** | `plugin/` has 1 skill (`setup`); `.claude/skills/` has 5 and `.claude/agents/` has 3, none of them in `plugin/` |
+| 4 | **Stale identity advice.** `setup` §5 tells the operator to pin `primary.terminal:` | CB-579 replaced that with `fleet.leaders.*.tab`. `record Primary` still exists (`FleetConfig.java:954`), so the advice is not dead — but it is no longer the mechanism |
+| 5 | **Stale install path.** README says `/plugin marketplace add ltms/claude-bridge` | the repo is `fleet/fleetd` since CB-623 |
+| 6 | **Stale names.** Plugin and marketplace are both `claude-bridge` | the project renamed to `fleetd` in CB-634 |
+| 7 | **Nothing references it.** `grep -rn "plugin/" CLAUDE.md docs/*.md` returns nothing | so no session is ever told the plugin exists — which is exactly why I planned it from scratch |
+
+Finding 7 is the root cause of the other six. A shipped capability that no instruction file
+mentions gets no maintenance, and the next person rebuilds it. That is the same failure the
+`CLAUDE.md` "Features" rule was written to stop.
+
+Finding 3 is the one that matters most for the operator's actual problem. A worker spawned into a
+**kb** worktree has no `implementer` skill, because only `claude-bridge` carries one in
+`.claude/skills/`. Every brief that says "Load the implementer skill" is a no-op outside this repo.
+The plugin is the right home for those skills and does not carry them yet.
+
+## 1. The problem, restated
+
+A project is "fleet-enabled" today by hand-edits nobody wrote down in one place:
+
+- a `fleet.leaders.<name>` entry in a host's gitignored `fleetd.yaml`;
+- the repo must carry `.claude/skills/*` for a worker to load `implementer` or `reviewer`;
+- the repo must carry the canonical bridge block in its `CLAUDE.md`;
+- the MCP mount arrives only because `LeadLauncher` and `ClaudeCodeLauncher` add `--mcp-config`
+  to the argv they build.
+
+Shown live on 2026-09-05: an operator opened `claude` by hand in `/home/ltms/LTMS/kb` on fleet01
+and the session had **no `fleet_*` tools at all**, because a hand-started agent never gets the
+launcher's `--mcp-config`. The plugin would have fixed that — if it had been installed, and if it
+had been mentioned anywhere.
+
+## 2. What was verified, and how
+
+| Claim | Evidence |
+|---|---|
+| A plugin can install globally | `~/.claude/plugins/installed_plugins.json` — scopes in use are `project` (5), `local` (4), **`user` (1)** |
+| A plugin can mount an MCP server | `~/.claude/plugins/marketplaces/kb-alms/.mcp.json` mounts `memory` at `"url": "${KB_MEMORY_URL}"` |
+| A plugin can carry skills, agents, commands, hooks | `kb-alms` ships `skills/` + `hooks/hooks.json`; `umputun-cc-thingz/plugins/planning` ships `agents/` + `skills/` |
+| Env vars interpolate in a plugin's `.mcp.json` | same `kb-alms` file: `${KB_MEMORY_URL}`, `${MEMORY_MCP_TOKEN}` |
+| A marketplace can be a plain git repo | `known_marketplaces.json` — `mgnl-code-review` has `"source": "git", "url": "https://..."` |
+| **This repo is already such a marketplace** | `.claude-plugin/marketplace.json`, committed in `ef1e014` |
+| A lead already binds to a project directory | `FleetConfig.java:1017` `record Leader(..., String workspace, String cwd)`; used at `LeadLauncher.java:193-199` |
+| The lead's mount comes from argv, not config | `LeadLauncher.java:253` adds `--mcp-config` |
+
+## 2b. Architect review + one measurement changed the design
+
+The architect verified the plan against the code and returned **build it with these changes**. Two
+of its findings are load-bearing. I checked both myself.
+
+### A plugin cannot carry the agent definitions — confirmed
+
+`ClaudeCodeLauncher.java:371` calls
+`agentDefinitionFile(spec.cwd(), spec.role(), ".claude", "agents")`, and
+`HerdrPeerLauncher.java:359-365` returns a path only when
+`<cwd>/.claude/agents/<role>.md` `Files.isRegularFile`. `ClaudeCodeLauncher.java:391-393` adds
+`--agent` **only** when that returns non-null. `OpenCodeLauncher` does the same for
+`.opencode/agent`.
+
+So the file must exist **in the member's worktree**. Moving `.claude/agents/*.md` into the plugin
+would silently stop every member getting `--agent`. **The agents stay in the repo.** My plan had
+this wrong.
+
+### The plugin does not reach members at all — confirmed, and worse than the architect could see
+
+`ClaudeCodeLauncher.java:285` does `putIfPresent(workerEnv, "CLAUDE_CONFIG_DIR", cfg.configDir())`.
+A member with `configDir` set reads that directory, not the operator's `~/.claude`.
+
+The architect could not check how far that goes, because `fleetd.yaml` is gitignored. I measured it:
+
+- **All four Claude profiles on the live fleet set `configDir`** — `local`, `local-direct`, `opus`,
+  `sonnet` (`grep -c configDir fleetd/fleetd.yaml` = 4).
+- Each `ccs` instance has its **own** `plugins/` directory: `gx10` (8 entries), `ltms` (11),
+  `ollama` (9), `work` (9).
+- Those directories are **four separate real directories with four separate inodes**, and
+  `installed_plugins.json` in each is a **separate inode with an identical md5**
+  (`51c6e1c853e32e656b817e123fbbfcc5`). They are *copies made once*, not links.
+
+So a plugin installed at user scope lands in exactly one instance's store. It would have to be
+installed once per `CLAUDE_CONFIG_DIR`, and each copy would then drift. **The plugin is not a
+delivery mechanism for member-facing assets on this host.**
+
+A side effect worth recording: my own session's `CLAUDE_CONFIG_DIR` is set, so the
+`~/.claude/plugins/*` evidence in section 2 is not even this session's store. The claims about what
+a plugin *can* do still hold — they were read from real manifests — but the directory I read them
+from is the wrong one for any conclusion about *this* session.
+
+### The design that follows
+
+Split by audience, not by mechanism:
+
+| Audience | Delivered by | Carries |
+|---|---|---|
+| operator / lead (a human opening any project) | **the plugin**, per config dir | the MCP mount, `setup`, the bridge charter |
+| member (a worker in a provisioned worktree) | **worktree provisioning** | `.claude/skills/*`, `.claude/agents/*` |
+
+The second row is not a new idea — it is what the code already does for agents, and it is why
+`agentDefinitionFile` looks in the worktree. Extending worktree provisioning to seed
+`.claude/skills/` from a fleetd-owned source is the consistent move, and it is what actually fixes
+"a worker in kb has no `implementer` skill". The plugin never could.
+
+### Other review findings I accepted
+
+- **Mount-name collision is real.** `PeerLauncher.java:34` is `fleet`; the plugin mounts `fleetd`.
+  A lead with both gets two mounts of one daemon and duplicate `fleet_*` tools. Rename the
+  plugin's server to `fleet`.
+- **Keep `--mcp-config` in `LeadLauncher`.** It is config→argv from `profile.mcpUrl()`, not drift,
+  and it is the only path that works for a member with its own `configDir`.
+- **I overstated #359.** `LeadCoordLoop.java:174-197` returns null and logs a warning that names the
+  fix; `tick()` leaves the message unacked, so the broker holds it and delivers once a lead is
+  named. It **stalls loudly and recovers** — it is not silent, and it is not data loss. My wording
+  in issue #361 needs the same correction.
+- **Stage 1 ends with tools that mostly cannot be used until stage 3**, because authority still
+  comes from the tab. Section 3 already said this; the stage table did not.
+
+## 3. What a plugin can and cannot do
+
+This is the part that decides the design, so it is stated before the design.
+
+**A plugin gives tools. It does not give authority.**
+
+`fleet_whoami` resolves a caller's role from the connection, not from what is mounted. A
+hand-started session in kb that mounts `fleet_*` through a plugin will be resolved as a **worker**
+and refused on every orchestration call, because its pane is not in a tab matching
+`fleet.leaders.*.tab`.
+
+So the plugin alone does not make a project fleet-enabled. It makes it *tool*-enabled. Registering
+the lead stays fleetd's job. Any plan that forgets this ships a plugin that looks installed and
+does nothing.
+
+```mermaid
+flowchart TB
+    P["fleet plugin<br/>(user scope, every session)"] --> T["fleet_* tools mounted"]
+    D["fleetd.yaml<br/>leaders.kb {tab, cwd}"] --> A["role = primary"]
+    T --> W["can call fleet_*"]
+    A --> W2["calls are authorized"]
+    W --> OK["working lead"]
+    W2 --> OK
+    T --> NO["tools mounted, every call refused"]
+    classDef good fill:#2f855a,stroke:#22543d,color:#ffffff;
+    classDef bad fill:#9b2c2c,stroke:#63171b,color:#ffffff;
+    class OK good
+    class NO bad
+```
+
+*Both halves are needed. The plugin is the left half only.*
+
+## 4. Proposed architecture — three layers
+
+### Layer 1: the plugin — lead-side only, fix the one that exists
+
+Keep it at `plugin/`, keep the marketplace at `.claude-plugin/marketplace.json`. Do not create a
+second one, and do not put member-facing assets in it (see 2b).
+
+```
+.claude-plugin/marketplace.json   -> rename to "fleetd"; keep source ./plugin
+plugin/
+  .claude-plugin/plugin.json      -> rename to "fleet"; bump version
+  .mcp.json                       -> mount name "fleet" (match PeerLauncher.MCP_MOUNT_NAME),
+                                     url "${FLEETD_MCP_URL}", 8765 default documented
+  skills/setup/SKILL.md           -> EXISTS. fix the stale primary.terminal advice (§5)
+  skills/bridge-charter/SKILL.md  -> NEW: the canonical CLAUDE.md block
+  README.md                       -> fix the install path (fleet/fleetd, not ltms/claude-bridge)
+```
+
+**Not in the plugin:** `agents/*.md` (the launcher requires them in the member's worktree —
+`ClaudeCodeLauncher.java:371,391`) and the three worker skills (a member with `configDir` set never
+reads the operator's plugin store — measured in 2b). Those belong to layer 1b.
+
+**A rename is a breaking change for anyone who installed 0.1.0.** The mount name goes `fleetd` ->
+`fleet`, so a project whose `.claude/settings.json` pre-allows `mcp__fleetd__fleet_whoami` stops
+matching. Only this fleet has it installed today, so the cost is small now and grows. Decide once.
+
+**This solves propagation of the charter.** `CLAUDE.md` says the bridge block "must stay
+byte-identical with the template in the wiki" and that "other projects carrying the block need the
+same edit" — a hand-copy the file itself admits is fragile, with a python snippet to check it. A
+plugin skill turns that into a version bump.
+
+### Layer 1b: worker skills reach members through the worktree, not the plugin
+
+This is the change that actually fixes "a worker in kb cannot load `implementer`".
+
+Worktree provisioning already writes into the member's tree — the parity overlay, the neutralised
+`.mcp.json`, the IDE overlay. Add one more: seed `<worktree>/.claude/skills/` from a fleetd-owned
+source directory, so every member gets `implementer`, `reviewer` and `hunter` whatever repo it is
+working in. `.claude/agents/` is already required there by the launcher, so this follows the
+grain of the design rather than cutting across it.
+
+Open question for implementation: copy or symlink, and where the source lives (a config key such
+as `memberSkills:`, or the plugin's own directory read by the daemon). A symlink is one source of
+truth but breaks if the member's tree is archived; a copy drifts but is self-contained.
+
+### Layer 2: host-global fleet settings
+
+`~/.fleet/fleetd.yaml` — the things that are true for the **machine**, not the project:
+
+- `broker:` and `coordinator:` (URIs come from env, no secrets in the file)
+- `profiles:` — backends, models, credentials, weights
+- `memberCredentials:` policy
+- `worktreeRoot`, `worktreeGroup`
+
+### Layer 3: per-project settings, committable
+
+`<project>/.fleet/project.yaml` — the things that are true for the **repo**:
+
+```yaml
+lead:
+  tab: "lead: kb"
+  profile: opus
+ide:
+  projectDir: ""          # kb is a Python repo, everything at the root
+worktree: true
+```
+
+**This split fixes a contradiction that exists today.** `ideProjectDir` is a property of a *repo*
+(fleetd's Maven module is a subdirectory; kb's code is at the root) but the config key is
+per-*profile*. One profile therefore cannot serve both repos — measured on fleet01 on 2026-09-05,
+where the key had to be commented out to make kb work. Moving it to a project file removes the
+contradiction rather than working around it.
+
+It is also committable, because it holds no secrets. A project that has been fleet-enabled once
+stays fleet-enabled for everyone who clones it.
+
+## 5. The `fleet-setup` skill
+
+What the operator actually asked for: one command that makes any project fleet-compatible.
+
+```mermaid
+sequenceDiagram
+    participant Op as Operator
+    participant Sk as fleet-setup skill
+    participant Fs as project files
+    participant Fd as fleetd
+    Op->>Sk: /fleet-setup (in any project)
+    Sk->>Fs: write .fleet/project.yaml
+    Sk->>Fs: add the bridge block to CLAUDE.md (if absent)
+    Sk->>Fd: register the lead (tab + cwd)
+    Fd-->>Sk: tab created, lead launched
+    Sk-->>Op: report what changed, and what is still manual
+```
+
+*The skill writes the project half and asks the daemon for the host half.*
+
+The registration step needs something that does not exist yet: an MCP tool such as
+`fleet_workspace_add{path, tab, profile}`, or a `fleetd` config include so a project file is picked
+up without hand-editing the host file. **This is the one genuinely new piece of daemon work.**
+
+## 6. Does this reduce fleetd's complexity?
+
+Honestly: **partly**. Claiming more than this would be wrong.
+
+**Yes, in three places.**
+
+1. Skill and agent delivery leaves the daemon and the repos entirely.
+2. Worktree config neutralisation (fleetd #134) gets safer. It blanks `.mcp.json` so the primary's
+   IDE and forge servers do not leak into a worker. Today the fleet mount survives only because
+   the launcher re-adds it by argv. With a user-scope plugin the fleet mount is outside the file
+   being neutralised, so the two concerns stop fighting.
+3. The `ideProjectDir` per-profile/per-repo contradiction disappears.
+
+**No, in the places that matter most.** fleetd still owns spawn, authorization, worktrees, herdr,
+the broker, tickets, and identity. A plugin cannot do any of those. The plugin is a **distribution**
+mechanism, not a replacement for the daemon.
+
+**And it adds one new risk.** The mount URL becomes a second source of truth. `fleetd.yaml` has the
+port; the plugin has the URL. Mitigation: the plugin reads `${FLEETD_MCP_URL}` only, and the host
+env is the single place it is set.
+
+## 7. Rollout stages
+
+| Stage | Content | Ends with |
+|---|---|---|
+| 0 | **Make it visible.** One `CLAUDE.md` line and one Features entry saying the plugin exists and where | nobody re-plans it a third time |
+| 1 | Fix the plugin's drift: mount name `fleet`, `${FLEETD_MCP_URL}`, names, README, stale `primary.terminal` advice | a lead in any project can install one plugin and get the mount |
+| 1b | Seed `.claude/skills/` into provisioned worktrees | **a worker in *kb* can load `implementer`** |
+| 2 | `.fleet/project.yaml` schema + `FleetConfig` reads it; `ideProjectDir` moves there | kb and fleetd both work off one profile |
+| 3 | Fix #359, then config-include for lead registration, wired into the existing `setup` skill | `/fleet:setup` in a fresh project produces a working lead |
+| 4 | Roll out to fleet01; retire the hand-copied CLAUDE.md block in favour of the skill | one `git pull` propagates the charter |
+
+Stage 0 is minutes of work and is the one that stops this happening again, so it goes first.
+**Stage 1b carries most of the value** and is independent of the plugin — it could ship first if the
+plugin rename needs more thought. Stage 1 alone ends with tools a hand-started session mostly
+cannot use, because authority still comes from the tab; that is fixed in stage 3, not stage 1.
+#359 moves ahead of stage 3 on the architect's advice, because stage 3 is what creates the second
+lead.
+
+## 8. Questions for the architect
+
+1. **Is the layer-2 / layer-3 split right?** Specifically: should `profiles:` stay host-global, or
+   should a project be able to pin which profiles it uses? Cost of getting this wrong is a config
+   that has to be re-split later.
+2. **Config include, or a new MCP tool, for registering a project's lead?** An include is passive
+   and survives a restart; a tool is live but writes to a gitignored file the daemon owns.
+3. **What happens when the plugin is absent?** Should `LeadLauncher` keep its `--mcp-config`
+   belt-and-braces, or is that the drift risk we should remove? Note opencode members cannot use
+   Claude plugins at all, so `OpenCodeLauncher` keeps its ephemeral config either way.
+4. **Does a user-scope plugin mount leak into members in a way we do not want?** Members already
+   inherit user-scope MCP servers (`--mcp-config` adds, it does not replace). A worker getting
+   `fleet_*` is correct and already happens. Confirm nothing else in the plugin should be
+   worker-invisible.
+5. **Two leads on one host both hold a subscription seat.** Is per-project leads the right unit, or
+   should one lead serve several projects by changing cwd?
+6. **Blocking defect to fix first or alongside:** `LeadCoordLoop.resolveLocalLead()` (lines
+   174-190) routes a peer message to "the sole lead" when no lead is *named* after
+   `coordinator.selfId`. The moment a host has two leads — exactly what this plan encourages —
+   cross-host coordination silently stops. Tracked as #359.

plugin/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
-  "name": "claude-bridge",
-  "description": "Make a project bridge-ready: mount the fleetd MCP gateway and set up standard Claude Code settings so this session can orchestrate a fleet of delegated workers. Ships no credentials.",
-  "version": "0.1.0",
+  "name": "fleet",
+  "description": "Make a project fleet-ready: mount the fleetd MCP gateway and set up standard Claude Code settings so this session can orchestrate a fleet of delegated workers. Lead-side only — member skills and agents travel in the worktree. Ships no credentials.",
+  "version": "0.2.0",
   "author": {
     "name": "LTMS"
   },

plugin/.mcp.json

@@ -1,8 +1,8 @@
 {
   "mcpServers": {
-    "fleetd": {
+    "fleet": {
       "type": "http",
-      "url": "http://127.0.0.1:8765/mcp"
+      "url": "${FLEETD_MCP_URL}"
     }
   }
 }

plugin/README.md

@@ -1,11 +1,24 @@
-# claude-bridge (Claude Code plugin)
+# fleet (Claude Code plugin)
 
-Makes a project **bridge-ready**: mounts the `fleetd` MCP gateway and applies standard Claude Code
+Makes a project **fleet-ready**: mounts the `fleetd` MCP gateway and applies standard Claude Code
 settings, so the session can orchestrate a fleet of delegated workers.
 
 **This plugin ships no credentials.** Every secret is referenced by environment-variable *name*;
 the values stay with the user. Nothing the plugin writes is unsafe to commit.
 
+## Scope — lead-side only
+
+This plugin configures **the session you are sitting in**: a lead, or any human-started Claude Code
+session that wants to talk to the daemon. It deliberately does **not** carry the worker playbook
+skills or the role agent definitions, and it cannot:
+
+- the launcher adds `--agent` only when `<worktree>/.claude/agents/<role>.md` exists in the
+  member's own tree (`ClaudeCodeLauncher.java:371,391`), so agent files must live in the repo;
+- a member's `CLAUDE_CONFIG_DIR` points at its profile's config directory
+  (`ClaudeCodeLauncher.java:285`), so it never reads the operator's plugin store.
+
+Member-facing assets travel in the worktree, not in this plugin. See fleetd #362.
+
 ## What it is not
 
 The plugin is the **client-side setup**, not the bridge. `fleetd` is a separate daemon and `herdr`
@@ -16,22 +29,35 @@ not try to install system services on your behalf.
 ## Install
 
 ```shell
-/plugin marketplace add ltms/claude-bridge
-/plugin install claude-bridge@claude-bridge
+/plugin marketplace add https://git.ltms.dev/fleet/fleetd
+/plugin install fleet@fleetd
+```
+
+Export the gateway URL — the plugin mounts `${FLEETD_MCP_URL}`, not a hardcoded address, so one
+plugin serves hosts that run the daemon on different ports:
+
+```shell
+export FLEETD_MCP_URL=http://127.0.0.1:8765/mcp
 ```
 
 Then, in the project you want to onboard:
 
 ```shell
-/claude-bridge:setup
+/fleet:setup
 ```
 
 ## What you get
 
 | Component | Effect |
 |---|---|
-| `.mcp.json` | mounts `fleetd` at `http://127.0.0.1:8765/mcp` for any session with the plugin enabled |
-| `skills/setup` | `/claude-bridge:setup` — preflight, project settings, credential guidance, and verification |
+| `.mcp.json` | mounts `fleet` at `${FLEETD_MCP_URL}` for any session with the plugin enabled |
+| `skills/setup` | `/fleet:setup` — preflight, project settings, credential guidance, and verification |
+
+The server is named **`fleet`** on purpose: that is `PeerLauncher.MCP_MOUNT_NAME` in the daemon and
+the name a spawned member's own mount carries. Version 0.1.0 named it `fleetd`, which produced two
+mounts of one daemon for anyone who also had a project-level `.mcp.json`. Upgrading from 0.1.0 is a
+**breaking change** — a project that pre-allowed `mcp__fleetd__fleet_whoami` in
+`.claude/settings.json` must be updated to `mcp__fleet__*`.
 
 Because the plugin carries its own `.mcp.json`, an installed plugin needs no project-level MCP
 file at all. The setup skill writes one only when you want the mount to work *without* the plugin —

plugin/skills/setup/SKILL.md

@@ -36,9 +36,17 @@ a time.
 command -v herdr  && herdr --version 2>&1 | head -1   || echo "MISSING: herdr"
 command -v ccs    && ccs version 2>&1 | head -1       || echo "MISSING: ccs (needed for worker profiles)"
 command -v codex  && codex --version 2>&1 | head -1   || echo "absent: codex (optional)"
-curl -s -m 5 http://127.0.0.1:8765/healthz || echo "MISSING: fleetd daemon is not reachable"
+curl -s -m 5 "${FLEETD_MCP_URL%/mcp}/healthz" 2>/dev/null \
+  || curl -s -m 5 http://127.0.0.1:8765/healthz \
+  || echo "MISSING: fleetd daemon is not reachable"
+[ -n "$FLEETD_MCP_URL" ] && echo "FLEETD_MCP_URL is set" || echo "MISSING: FLEETD_MCP_URL"
 ```
 
+**`FLEETD_MCP_URL` is required.** The plugin's own `.mcp.json` mounts `${FLEETD_MCP_URL}` rather
+than a hardcoded address, so one plugin can serve hosts that run the daemon on different ports. If
+it is unset the mount does not resolve. The usual value is `http://127.0.0.1:8765/mcp`; tell the
+user to export it, do not write it into a file for them.
+
 A healthy daemon answers with its status **and the herdr protocol it negotiated**:
 
 ```json
@@ -70,7 +78,7 @@ The entry to add, exactly:
 ```json
 {
   "mcpServers": {
-    "fleetd": {
+    "fleet": {
       "type": "http",
       "url": "http://127.0.0.1:8765/mcp"
     }
@@ -78,8 +86,13 @@ The entry to add, exactly:
 }
 ```
 
-If `.mcp.json` already exists, add only the `fleetd` key and leave every other server untouched.
-If a `fleetd` entry is already there with a different URL, **ask** rather than assuming yours is
+**The server must be named `fleet`.** That is `PeerLauncher.MCP_MOUNT_NAME` in the daemon, the name
+a spawned member's mount carries, and the name the `mcp__fleet__*` role heuristic in `CLAUDE.md`
+keys on. An earlier version of this plugin named it `fleetd`, which gave a lead with both a project
+file and the plugin **two mounts of the same daemon** and a duplicated `fleet_*` tool set.
+
+If `.mcp.json` already exists, add only the `fleet` key and leave every other server untouched.
+If a `fleet` entry is already there with a different URL, **ask** rather than assuming yours is
 right — a non-default port usually means a deliberate second daemon.
 
 > **If this plugin is installed, you can skip this step entirely.** The plugin ships its own
@@ -111,11 +124,11 @@ project already set.
   "$schema": "https://json.schemastore.org/claude-code-settings.json",
   "permissions": {
     "allow": [
-      "mcp__fleetd__fleet_whoami",
-      "mcp__fleetd__fleet_list",
-      "mcp__fleetd__fleet_status",
-      "mcp__fleetd__fleet_profiles",
-      "mcp__fleetd__fleet_poll"
+      "mcp__fleet__fleet_whoami",
+      "mcp__fleet__fleet_list",
+      "mcp__fleet__fleet_status",
+      "mcp__fleet__fleet_profiles",
+      "mcp__fleet__fleet_poll"
     ]
   }
 }
@@ -161,19 +174,31 @@ fleet_whoami
 ```
 
 - `{"role":"primary"}` — correct, you are done with this step.
-- `{"role":"worker", …}` — **this is the trap.** If the primary runs inside a herdr pane, the
-  daemon resolves it to a terminal and classifies it as a worker, refusing `spawn`/`send`/`stop`:
-  every verb an orchestrator exists to call. It is **self-locking**, because the daemon can only
-  *learn* the primary's terminal from those same refused calls. The only way out is an
-  operator-set pin in the daemon's config:
+- `{"role":"worker", …}` — **this is the trap.** If the lead runs inside a herdr pane whose tab the
+  daemon does not recognise, it is classified as a worker and refused on `spawn`/`send`/`stop`:
+  every verb an orchestrator exists to call. It is **self-locking**, because those are the same
+  calls that would tell the daemon who you are.
+
+  Identity is the **tab label**, matched exactly and case-insensitively:
 
   ```yaml
-  primary:
-    terminal: term_xxxxxxxxxxxx     # the terminalId fleet_whoami just reported
+  fleet:
+    leaders:
+      kb:                        # name it after coordinator.selfId if this host uses lead-to-lead
+        profile: opus
+        tab: "lead: kb"          # the exact label of the tab this lead sits in
+        cwd: /path/to/the/project
   ```
 
-  The daemon reads this **at boot**, so it needs a restart. Re-pin whenever the primary moves
-  panes — a stale pin fails exactly as silently as no pin.
+  A tab label is stable across restarts of the agent inside it, which is why CB-579 replaced the
+  older `primary.terminal:` pin — a herdr `terminal_id` changed on every restart and cost a config
+  edit each time. `primary.terminal:` still parses, but it is no longer the mechanism; do not
+  reach for it.
+
+  The daemon reads `leaders:` **at boot**, so a new entry needs a restart. Two things to check
+  afterwards: that `fleet_whoami` now answers `primary`, and that no *stale* tab carries the same
+  label — duplicate lead tabs are their own failure (#359), and they stall lead-to-lead delivery
+  until one lead is named after `coordinator.selfId`.
 
 Then prove the fleet actually works, with a real spawn:
 

scripts/redeploy-fleetd.sh

@@ -116,6 +116,50 @@ check_log_path_matches_plist() {
   ok "log path check: script and plist agree ($resolved_out)"
 }
 
+# Classify ERROR lines in one fresh log region. AMQP failure messages now include the connection
+# name, so a recovery can clear only errors for its own connection. A candidate with neither name
+# remains unexplained: it must never be quieted by a recovery on the other connection.
+classify_amqp_connection_errors() {
+  local log_file="$1" line pending_inbox=0 pending_lead_mailbox=0
+  REDEPLOY_ERROR_COUNT=0
+  REDEPLOY_RECOVERED_AMQP_ERRORS=0
+  REDEPLOY_UNEXPLAINED_ERRORS=0
+
+  while IFS= read -r line || [ -n "$line" ]; do
+    case "$line" in
+      *' ERROR '*|*' SEVERE '*)
+        REDEPLOY_ERROR_COUNT=$((REDEPLOY_ERROR_COUNT + 1))
+        case "$line" in
+          *'AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred'*|*'AMQP connection fleetd-reply-inbox: Caught an exception during connection recovery!'*)
+            pending_inbox=$((pending_inbox + 1))
+            ;;
+          *'AMQP connection fleetd-lead-mailbox: An unexpected connection driver error occurred'*|*'AMQP connection fleetd-lead-mailbox: Caught an exception during connection recovery!'*)
+            pending_lead_mailbox=$((pending_lead_mailbox + 1))
+            ;;
+          *'AMQP connection'*'An unexpected connection driver error occurred'*|*'AMQP connection'*'Caught an exception during connection recovery!'*)
+            REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + 1))
+            ;;
+          *) REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + 1)) ;;
+        esac
+        ;;
+      *'AMQP connection recovered; cleared held replies for fresh redelivery'*)
+        if [ "$pending_inbox" -gt 0 ]; then
+          pending_inbox=$((pending_inbox - 1))
+          REDEPLOY_RECOVERED_AMQP_ERRORS=$((REDEPLOY_RECOVERED_AMQP_ERRORS + 1))
+        fi
+        ;;
+      *'AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery'*)
+        if [ "$pending_lead_mailbox" -gt 0 ]; then
+          pending_lead_mailbox=$((pending_lead_mailbox - 1))
+          REDEPLOY_RECOVERED_AMQP_ERRORS=$((REDEPLOY_RECOVERED_AMQP_ERRORS + 1))
+        fi
+        ;;
+    esac
+  done < "$log_file"
+
+  REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + pending_inbox + pending_lead_mailbox))
+}
+
 # CB-600: sourceable for testing. When this file is SOURCED (not executed) it stops here — nothing
 # below runs — so a test harness can `source` it to call check_log_path_matches_plist (or the
 # other pure helpers above) against a throwaway plist fixture without ever reaching the mutating
@@ -361,15 +405,21 @@ tail -n "+$((RESTART_MARK + 1))" "$OUT" 2>/dev/null \
   | grep -iE 'deferred|classification:|fleet health:|coverage' | tail -8 | sed 's/^/   /' \
   || echo "   (nothing reported)"
 
-# Errors since the restart, anchored to the marker so old noise cannot leak in.
-ERRS="$(tail -n "+$((RESTART_MARK + 1))" "$OUT" 2>/dev/null | grep -cE ' (ERROR|SEVERE) ' || true)"
+# Errors since the restart, anchored to the marker so old noise cannot leak in. Keep the fresh
+# region in a file because the classifier must preserve the order of errors and recoveries.
+FRESH_LOG="$(mktemp -t fleetd-fresh-log)"
+trap 'rm -f "$FRESH_LOG"' EXIT
+tail -n "+$((RESTART_MARK + 1))" "$OUT" > "$FRESH_LOG" 2>/dev/null || true
+classify_amqp_connection_errors "$FRESH_LOG"
 say "result"
 ok "pid $NEW_PID, jar $(jar_id)"
-if [ "${ERRS:-0}" -gt 0 ]; then
-  warn "$ERRS ERROR lines since restart:"
-  tail -n "+$((RESTART_MARK + 1))" "$OUT" | grep -E ' (ERROR|SEVERE) ' | tail -5 | sed 's/^/   /'
-else
+if [ "$REDEPLOY_ERROR_COUNT" -eq 0 ]; then
   ok "no ERROR lines since restart"
+elif [ "$REDEPLOY_UNEXPLAINED_ERRORS" -eq 0 ]; then
+  ok "$REDEPLOY_RECOVERED_AMQP_ERRORS AMQP connection reset ERROR lines recovered since restart"
+else
+  warn "$REDEPLOY_ERROR_COUNT ERROR lines since restart:"
+  grep -E ' (ERROR|SEVERE) ' "$FRESH_LOG" | tail -5 | sed 's/^/   /'
 fi
 echo
 echo "   Next: call fleet_whoami and confirm it still answers 'primary'. A lead whose tab label"

scripts/test-redeploy-fleetd.sh

@@ -0,0 +1,238 @@
+#!/usr/bin/env bash
+# Self-contained checks for the pure log classifier in redeploy-fleetd.sh.
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+TMP="$(mktemp -d "$ROOT/.redeploy-log-test.XXXXXX")"
+trap 'rm -rf "$TMP"' EXIT
+
+# Sourcing stops before redeploy-fleetd.sh can build, stop, or start the daemon.
+source "$ROOT/scripts/redeploy-fleetd.sh"
+
+fail() {
+  printf 'FAIL: %s\n' "$*" >&2
+  return 1
+}
+
+assert_equals() {
+  local expected="$1" actual="$2" description="$3"
+  [ "$expected" = "$actual" ] || fail "$description: expected $expected, got $actual"
+}
+
+classify_fixture() {
+  local name="$1"
+  classify_amqp_connection_errors "$TMP/$name"
+}
+
+test_no_errors() {
+  cat > "$TMP/no-errors.log" <<'LOG'
+2026-09-05 12:00:00 INFO fleetd listening
+LOG
+  classify_fixture no-errors.log
+  assert_equals 0 "$REDEPLOY_ERROR_COUNT" "no-errors total"
+  assert_equals 0 "$REDEPLOY_UNEXPLAINED_ERRORS" "no-errors unexplained"
+}
+
+test_recovery_patterns_match_source() {
+  grep -F 'AMQP connection {}: {}' "$ROOT/fleetd/src/main/java/dev/ltms/fleet/msg/AmqpReplyInbox.java" > /dev/null \
+    || fail "AMQP failure pattern no longer matches source"
+  grep -F 'AMQP connection recovered; cleared held replies for fresh redelivery' \
+    "$ROOT/fleetd/src/main/java/dev/ltms/fleet/msg/AmqpReplyInbox.java" > /dev/null \
+    || fail "reply-inbox recovery pattern no longer matches source"
+  grep -F 'AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery' \
+    "$ROOT/fleetd/src/main/java/dev/ltms/fleet/msg/LeadMailbox.java" > /dev/null \
+    || fail "lead-mailbox recovery pattern no longer matches source"
+}
+
+test_attributed_recovered_connection_error() {
+  cat > "$TMP/attributed-recovered.log" <<'LOG'
+2026-09-05 12:00:00 ERROR [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred
+2026-09-05 12:00:01 INFO [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection recovered; cleared held replies for fresh redelivery
+LOG
+  classify_fixture attributed-recovered.log
+  assert_equals 1 "$REDEPLOY_ERROR_COUNT" "attributed-recovered total"
+  assert_equals 1 "$REDEPLOY_RECOVERED_AMQP_ERRORS" "attributed-recovered errors"
+  assert_equals 0 "$REDEPLOY_UNEXPLAINED_ERRORS" "attributed-recovered unexplained"
+}
+
+test_source_derived_error_shapes_recover_by_connection() {
+  # These ERROR shapes come from AmqpConnectionFailureLogger on main. They need a live-log check
+  # after redeploy because the new code has not yet written a production line.
+  cat > "$TMP/source-derived.log" <<'LOG'
+17:37:53.537 ERROR [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred
+17:37:54.537 ERROR [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection fleetd-reply-inbox: Caught an exception during connection recovery!
+17:37:55.537 ERROR [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred
+17:37:56.537 ERROR [AMQP Connection broker:5672] d.ltms.fleet.msg.LeadMailbox - AMQP connection fleetd-lead-mailbox: An unexpected connection driver error occurred
+17:37:57.537 ERROR [AMQP Connection broker:5672] d.ltms.fleet.msg.LeadMailbox - AMQP connection fleetd-lead-mailbox: Caught an exception during connection recovery!
+17:37:58.537 ERROR [AMQP Connection broker:5672] d.ltms.fleet.msg.LeadMailbox - AMQP connection fleetd-lead-mailbox: An unexpected connection driver error occurred
+17:38:00.000 INFO  [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection recovered; cleared held replies for fresh redelivery
+17:38:01.000 INFO  [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection recovered; cleared held replies for fresh redelivery
+17:38:02.000 INFO  [AMQP Connection broker:5672] d.l.fleet.msg.AmqpReplyInbox - AMQP connection recovered; cleared held replies for fresh redelivery
+17:38:03.000 INFO  [AMQP Connection broker:5672] d.ltms.fleet.msg.LeadMailbox - AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery
+17:38:04.000 INFO  [AMQP Connection broker:5672] d.ltms.fleet.msg.LeadMailbox - AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery
+17:38:05.000 INFO  [AMQP Connection broker:5672] d.ltms.fleet.msg.LeadMailbox - AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery
+LOG
+  classify_fixture source-derived.log
+  assert_equals 6 "$REDEPLOY_ERROR_COUNT" "source-derived total"
+  assert_equals 6 "$REDEPLOY_RECOVERED_AMQP_ERRORS" "source-derived recovered"
+  assert_equals 0 "$REDEPLOY_UNEXPLAINED_ERRORS" "source-derived unexplained"
+}
+
+test_cross_connection_unattributable_errors_stay_loud() {
+  # This candidate has neither stable connection name, so LeadMailbox recovery must not consume it.
+  cat > "$TMP/cross-unattributable.log" <<'LOG'
+2026-09-05 12:00:00 ERROR [AMQP Connection broker:5672] unknown - AMQP connection: An unexpected connection driver error occurred
+2026-09-05 12:00:01 ERROR [AMQP Connection broker:5672] unknown - AMQP connection: An unexpected connection driver error occurred
+2026-09-05 12:00:02 INFO  [AMQP Connection 10.10.20.13:5672] d.ltms.fleet.msg.LeadMailbox - AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery
+2026-09-05 12:00:03 INFO  [AMQP Connection 10.10.20.13:5672] d.ltms.fleet.msg.LeadMailbox - AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery
+LOG
+  classify_fixture cross-unattributable.log
+  assert_equals 2 "$REDEPLOY_ERROR_COUNT" "cross-unattributable total"
+  assert_equals 0 "$REDEPLOY_RECOVERED_AMQP_ERRORS" "cross-unattributable recovered"
+  assert_equals 2 "$REDEPLOY_UNEXPLAINED_ERRORS" "cross-unattributable unexplained"
+}
+
+test_attributed_cross_connection_errors_stay_loud() {
+  # LeadMailbox recovery cannot heal AmqpReplyInbox errors.
+  cat > "$TMP/cross-attributed.log" <<'LOG'
+2026-09-05 12:00:00 ERROR AmqpReplyInbox - AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred
+2026-09-05 12:00:01 ERROR AmqpReplyInbox - AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred
+2026-09-05 12:00:02 INFO LeadMailbox - AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery
+2026-09-05 12:00:03 INFO LeadMailbox - AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery
+LOG
+  classify_fixture cross-attributed.log
+  assert_equals 2 "$REDEPLOY_ERROR_COUNT" "cross-attributed total"
+  assert_equals 0 "$REDEPLOY_RECOVERED_AMQP_ERRORS" "cross-attributed recovered"
+  assert_equals 2 "$REDEPLOY_UNEXPLAINED_ERRORS" "cross-attributed unexplained"
+}
+
+test_attributed_unrecovered_connection_error() {
+  cat > "$TMP/unrecovered.log" <<'LOG'
+2026-09-05 12:00:00 ERROR AmqpReplyInbox - AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred
+LOG
+  classify_fixture unrecovered.log
+  assert_equals 1 "$REDEPLOY_ERROR_COUNT" "unrecovered total"
+  assert_equals 0 "$REDEPLOY_RECOVERED_AMQP_ERRORS" "unrecovered AMQP errors"
+  assert_equals 1 "$REDEPLOY_UNEXPLAINED_ERRORS" "unrecovered unexplained"
+}
+
+test_other_error_is_unexplained() {
+  cat > "$TMP/other-error.log" <<'LOG'
+2026-09-05 12:00:00 ERROR dev.ltms.fleet.Fleetd - startup failed
+2026-09-05 12:00:01 INFO dev.ltms.fleet.msg.AmqpReplyInbox - AMQP connection recovered; cleared held replies for fresh redelivery
+LOG
+  classify_fixture other-error.log
+  assert_equals 1 "$REDEPLOY_ERROR_COUNT" "other-error total"
+  assert_equals 1 "$REDEPLOY_UNEXPLAINED_ERRORS" "other-error unexplained"
+}
+
+test_recovery_requirement_mutation_is_caught() {
+  classify_amqp_connection_errors() {
+    local log_file="$1" line
+    REDEPLOY_ERROR_COUNT=0
+    REDEPLOY_RECOVERED_AMQP_ERRORS=0
+    REDEPLOY_UNEXPLAINED_ERRORS=0
+    while IFS= read -r line || [ -n "$line" ]; do
+      case "$line" in
+        *' ERROR '*|*' SEVERE '*)
+          REDEPLOY_ERROR_COUNT=$((REDEPLOY_ERROR_COUNT + 1))
+          case "$line" in
+            *'AMQP connection fleetd-reply-inbox: An unexpected connection driver error occurred'*)
+              REDEPLOY_RECOVERED_AMQP_ERRORS=$((REDEPLOY_RECOVERED_AMQP_ERRORS + 1))
+              ;;
+            *) REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + 1)) ;;
+          esac
+          ;;
+      esac
+    done < "$log_file"
+  }
+
+  if test_attributed_unrecovered_connection_error > "$TMP/mutation-output" 2>&1; then
+    fail "mutation accepted an unrecovered connection error"
+  fi
+  grep -F 'FAIL: unrecovered AMQP errors: expected 0, got 1' "$TMP/mutation-output" > /dev/null \
+    || fail "mutation failed without the expected assertion"
+  printf 'Recovery mutation: FAIL: unrecovered AMQP errors: expected 0, got 1\n'
+}
+
+test_shared_counter_mutation_is_caught() {
+  classify_amqp_connection_errors() {
+    local log_file="$1" line pending=0
+    REDEPLOY_ERROR_COUNT=0
+    REDEPLOY_RECOVERED_AMQP_ERRORS=0
+    REDEPLOY_UNEXPLAINED_ERRORS=0
+    while IFS= read -r line || [ -n "$line" ]; do
+      case "$line" in
+        *' ERROR '*|*' SEVERE '*)
+          REDEPLOY_ERROR_COUNT=$((REDEPLOY_ERROR_COUNT + 1))
+          case "$line" in
+            *'AMQP connection'*'An unexpected connection driver error occurred'*|*'AMQP connection'*'Caught an exception during connection recovery!'*)
+              case "$line" in
+                *'fleetd-reply-inbox'*|*'fleetd-lead-mailbox'*) pending=$((pending + 1)) ;;
+                *) REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + 1)) ;;
+              esac
+              ;;
+            *) REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + 1)) ;;
+          esac
+          ;;
+        *'AMQP connection recovered; cleared held replies for fresh redelivery'*|*'AMQP lead mailbox connection recovered; cleared held messages for fresh redelivery'*)
+          if [ "$pending" -gt 0 ]; then
+            pending=$((pending - 1))
+            REDEPLOY_RECOVERED_AMQP_ERRORS=$((REDEPLOY_RECOVERED_AMQP_ERRORS + 1))
+          fi
+          ;;
+      esac
+    done < "$log_file"
+    REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + pending))
+  }
+
+  if test_attributed_cross_connection_errors_stay_loud > "$TMP/shared-mutation-output" 2>&1; then
+    fail "shared counter mutation accepted cross-connection recovery"
+  fi
+  grep -F 'FAIL: cross-attributed recovered: expected 0, got 2' "$TMP/shared-mutation-output" > /dev/null \
+    || fail "shared counter mutation failed without the expected assertion"
+  printf 'Shared-counter mutation: FAIL: cross-attributed recovered: expected 0, got 2\n'
+}
+
+test_unattributable_quiet_mutation_is_caught() {
+  classify_amqp_connection_errors() {
+    local log_file="$1" line
+    REDEPLOY_ERROR_COUNT=0
+    REDEPLOY_RECOVERED_AMQP_ERRORS=0
+    REDEPLOY_UNEXPLAINED_ERRORS=0
+    while IFS= read -r line || [ -n "$line" ]; do
+      case "$line" in
+        *' ERROR '*|*' SEVERE '*)
+          REDEPLOY_ERROR_COUNT=$((REDEPLOY_ERROR_COUNT + 1))
+          case "$line" in
+            *'AMQP connection'*'An unexpected connection driver error occurred'*|*'AMQP connection'*'Caught an exception during connection recovery!'*)
+              REDEPLOY_RECOVERED_AMQP_ERRORS=$((REDEPLOY_RECOVERED_AMQP_ERRORS + 1))
+              ;;
+            *) REDEPLOY_UNEXPLAINED_ERRORS=$((REDEPLOY_UNEXPLAINED_ERRORS + 1)) ;;
+          esac
+          ;;
+      esac
+    done < "$log_file"
+  }
+
+  if test_cross_connection_unattributable_errors_stay_loud > "$TMP/unattributable-mutation-output" 2>&1; then
+    fail "unattributable mutation accepted an unknown connection"
+  fi
+  grep -F 'FAIL: cross-unattributable recovered: expected 0, got 2' "$TMP/unattributable-mutation-output" > /dev/null \
+    || fail "unattributable mutation failed without the expected assertion"
+  printf 'Unattributable mutation: FAIL: cross-unattributable recovered: expected 0, got 2\n'
+}
+
+test_no_errors
+test_recovery_patterns_match_source
+test_attributed_recovered_connection_error
+test_source_derived_error_shapes_recover_by_connection
+test_cross_connection_unattributable_errors_stay_loud
+test_attributed_cross_connection_errors_stay_loud
+test_attributed_unrecovered_connection_error
+test_other_error_is_unexplained
+test_recovery_requirement_mutation_is_caught
+test_shared_counter_mutation_is_caught
+test_unattributable_quiet_mutation_is_caught
+printf 'PASS: redeploy log classifier\n'