#608 replace MessageService timing sleeps

Two bullets in the handover skill were telling every outgoing lead something
that is no longer true.

1. The skill said the bootstrap prompt "has never yet landed, and the fix is
   unproven (fleetd #489)", and told the lead to warn the operator it may fail.
   Measured today from fleetd/fleetd.out:

     grep -c "lead-rollover: rolled" -> 4
     grep -c "lead-rollover:"        -> 16   (positive control)
     grep -c "Unknown command"       -> 0

   Three of the four rolls ran on 2026-09-22 (10:01:43, 10:38:28, 11:15:47).
   Each cleared the old lead and bootstrapped a fresh one against the handover
   file. The old "Unknown command: /clearFresh" failure does not appear at all.
   The paragraph now carries the measured result and the three re-measure
   commands, including the control line, because a broken grep pattern returns
   a clean 0 that reads like good news.

   It also records that the "/clear was never observed as WORKING ... releasing
   rather than wedging the roll" WARN accompanies every successful roll. That
   is the safe branch, not a failure, and it was being misread as one.

2. The skill said requireOperatorConfirm "defaults to true and this is the only
   thing standing between a judgement call and a wiped session", which reads as
   if asking is always required. The default is still true
   (FleetConfig.java:1426), but this host set it to false on 2026-09-22 on the
   operator's explicit grant. The bullet now says to read the live value rather
   than assume, and notes the key is deferred, not hot.

   It also warns that until fleetd #621 merges, LeadHeartbeatLoop.contextNotice()
   still hardcodes "ask the operator" and takes no config, so the nudge text and
   the config disagree. Trust the config. That warning names the ticket that
   removes it.

Documentation only. No code or test changes.

.claude/skills/handover/SKILL.md

@@ -167,19 +167,59 @@ fails.
 - **There is no terminal or session parameter, on purpose.** The pane is always your own, resolved
   from your connection, so you can only ever roll yourself.
 - **`operatorConfirmed` is your report of what a human told you.** Do not pass `true` because you
-  are confident. Ask, wait for the answer, then pass what they said. `requireOperatorConfirm`
-  defaults to `true` and this is the only thing standing between a judgement call and a wiped
-  session.
+  are confident. Ask, wait for the answer, then pass what they said.
+
+- **Whether you must ask at all depends on `leadRollover.requireOperatorConfirm`. Check it; do not
+  assume.** The default is `true` (`FleetConfig.java:1426`), and then `confirm` refuses unless you
+  also pass `operatorConfirmed: true`. **This host set it to `false` on 2026-09-22**, on the
+  operator's explicit grant, because they do not want to approve routine context rolls. Where it is
+  `false`, the three handover-file checks are the whole gate: the file must exist, be fresher than
+  `maxDocAgeSeconds`, and have been modified after the open request.
+
+  Read the live value rather than trusting this line:
+
+  ```bash
+  grep -A1 'requireOperatorConfirm' fleetd/fleetd.yaml
+  ```
+
+  No match means the key is unset, so the default `true` applies and you must ask. The key is
+  **deferred, not hot** — it is read once at boot, so an edit does nothing until the daemon is
+  redeployed.
+
+  **Until fleetd #621 merges, the nudge text will tell you to ask the operator even where the
+  daemon no longer requires it.** `LeadHeartbeatLoop.contextNotice()` hardcodes "ask the operator"
+  and takes no config, so it cannot know. Trust the config value over the nudge text. Once #621 is
+  merged and deployed, the nudge matches the config and this warning can be deleted.
+
 - **The roll can still refuse after `confirm` returns**, and by then there is no caller to tell.
   Those outcomes are logged only, as `lead-rollover:` lines in the daemon log.
-- **The bootstrap prompt has never yet landed, and the fix is unproven (fleetd #489).** The first
-  real rollover, on 2026-09-12, joined `/clear` and the bootstrap text into one line and Claude Code
-  refused it as `Unknown command: /clearFresh`. The pane was never cleared and no context was lost,
-  so the failure was safe — the roll simply did nothing. PR #490 fixed the cause and is deployed,
-  but no roll has bootstrapped a fresh session end to end yet. **Assume it may still fail, and tell
-  the operator so before you confirm.** The recovery is the same either way: the file is already
-  written, so the operator starts a session and points it at the file. That is why you write the
-  file before you confirm, and never the other way round.
+- **The bootstrap prompt works end to end. Measured 2026-09-22.** This used to say the fix was
+  unproven (fleetd #489) and told you to expect a failure. That is no longer true. The daemon log
+  now holds four `lead-rollover: rolled` lines, and three of them ran on 2026-09-22 at 10:01:43,
+  10:38:28 and 11:15:47. Each one cleared the old lead and started a fresh session against the
+  handover file, with the configured `bootstrapText` arriving as its first message. No context was
+  lost. The old `Unknown command: /clearFresh` failure from 2026-09-12 does not appear in the log
+  at all. Re-measure both numbers with:
+
+  ```bash
+  grep -c "lead-rollover: rolled" fleetd/fleetd.out   # successful rolls
+  grep -c "lead-rollover:" fleetd/fleetd.out          # positive control: must be larger
+  grep -c "Unknown command" fleetd/fleetd.out         # the old failure: expect 0
+  ```
+
+  Run the control line too. A broken pattern returns a clean `0` that reads exactly like good news.
+  If the first number stops growing across rolls, or `Unknown command` returns anything above 0,
+  the bootstrap has regressed and this paragraph is stale again.
+
+  **You still write the file before you confirm, and never the other way round.** That order is not
+  about the bootstrap being unreliable. It is what the daemon checks: the handover file must have
+  been modified *after* the open request, or `confirm` refuses it as stale.
+
+- **One warning in the log is normal and is not a failure.** Every one of the three rolls above also
+  logged `/clear on term_… was never observed as WORKING after 8 consecutive IDLE/DONE polls —
+  releasing rather than wedging the roll`. The daemon could not see the pane go WORKING after
+  `/clear`, so it released instead of hanging. The roll then succeeded anyway. That is the safe
+  branch behaving correctly. Do not report it as a broken roll.
 
 ## Writing style
 

fleetd/fleetd.example.yaml

@@ -209,13 +209,13 @@ herdrSocket: ~/.config/herdr/herdr.sock
 #                      e.g. `env DISPLAY=:10.0 idea {dir}`. Best-effort: a failure is logged, never
 #                      fails the spawn. Omit to open the member's module by hand. There is no close
 #                      half yet — an opened module stays open until the operator closes it.
-#   autoCompactWindow → opt-in, default off. A bounded token window that forces a spawned member to
-#                      compact its context instead of running on the backend's own default and dying
-#                      mid-turn (losing its fleet_reply — the whole point of the turn — with it).
+#   autoCompactWindow → opt-in, default off. A bounded token window that forces a launched Claude Code
+#                      lead or member to compact its context instead of running on the backend's own
+#                      default. A member that runs out of context can die mid-turn and lose its fleet_reply.
 #                      Validated at config load to [100000, 1000000] — the band Claude Code's own
 #                      --autocompact flag accepts.
 #                      CROSS-BACKEND SEMANTICS DIFFER: on claude-code this is a launch-time
-#                      `--autocompact <tokens>` flag — the member compacts AT this window. opencode
+#                      `--autocompact <tokens>` flag — the Claude Code session compacts AT this window. opencode
 #                      has no equivalent flag (it only forces `compaction.auto: true`, unconditionally,
 #                      already), so this is instead applied as the model's `limit.context` in the
 #                      generated opencode.json — the member compacts WITHIN this window, not exactly
@@ -415,7 +415,7 @@ profiles:
     # ideMcpUrl: http://127.0.0.1:29170/index-mcp/streamable-http   # opt-in (CB-634): IDE code intelligence, pinned to the worktree
     # ideProjectDir: fleetd          # CB-634: module dir the IDE opens + the overlay pins (this repo's pom is in fleetd/)
     # ideOpenCommand: env DISPLAY=:10.0 idea {dir}   # CB-634 auto-open: opens {dir} in the IDE at spawn; omit to open by hand
-    # autoCompactWindow: 250000   # opt-in: bound member context; claude-code compacts AT this, opencode within it (model limit.context)
+    # autoCompactWindow: 250000   # opt-in: bound Claude Code lead/member context; claude-code compacts AT this, opencode within it (model limit.context)
   gx11:                              # a second backend, so `placement: weighted` has a choice
     baseUrl: http://gx01.gw:8000     # self-hosted; ccs handles the model + token
     placement: tab

fleetd/src/main/java/dev/ltms/fleet/config/FleetConfig.java

@@ -462,8 +462,8 @@ public record FleetConfig(
      *                       profile that does not opt in. Read live off the current config, so it is
      *                       HOT: a change takes effect on the next exhaustion classification / spawn,
      *                       no restart needed.
-     * @param autoCompactWindow  opt-in per-profile token window that forces a spawned member to
-     *                       auto-compact its context at (Claude Code) or within (opencode) a bound the
+     * @param autoCompactWindow  opt-in per-profile token window that forces a launched Claude Code
+     *                       session to auto-compact its context at, or an opencode session within, a bound the
      *                       operator chooses, instead of the backend's own default. {@code null} (the
      *                       default) leaves today's behaviour exactly — opencode already forces
      *                       {@code compaction.auto: true} unconditionally (CB-523) but has no absolute
@@ -1875,6 +1875,7 @@ public record FleetConfig(
             rejectDuplicateMemberSlots(yaml);
             rejectNegativeMaxLoad(yaml);
             rejectAutoCompactWindowOutOfRange(yaml);
+            warnConflictingAutoCompactWindows(yaml);
             rejectMalformedProfilePatterns(yaml);
             rejectUnknownKind(yaml);
             rejectUnknownAuthMode(yaml);
@@ -2225,6 +2226,73 @@ public record FleetConfig(
         }
     }
 
+    /**
+     * Warn (never refuse to start) about a Claude Code profile whose auto-compaction flag and
+     * environment setting disagree.
+     *
+     * <p>Renamed from {@code rejectConflictingAutoCompactWindows} (fleetd #601 review, measured
+     * 2026-09-22): that method threw {@link IllegalStateException}, so {@link #load(Path)} refused
+     * to start on a config carrying this conflict. On this host, four profiles trip it, including
+     * the lead's own profile and the one every worker spawns on — so the throw is not a rare edge
+     * case. Under launchd, a throw inside {@code load()} is a restart loop, not an error an operator
+     * reads once, and the config that would fix it ({@code fleetd/fleetd.yaml}) is gitignored, so
+     * the cause is invisible on the host where it bites. A WARN gives the operator the same
+     * information — which profiles, and now both values, so they can fix it without reading the
+     * source — without ever taking the fleet down.
+     *
+     * <p>fleetd #618 measured which of the two inputs Claude Code actually follows when they
+     * disagree: the environment variable wins, so {@code autoCompactWindow} is inert on a profile
+     * that also sets the env var. This method only detects and reports the disagreement — it does
+     * not correct it — see {@link dev.ltms.fleet.launch.ClaudeCodeArguments} for the full measured
+     * precedence.
+     *
+     * <p>Equal values never warn: either input then produces the same session window, so there is
+     * nothing to reconcile.
+     */
+    static void warnConflictingAutoCompactWindows(String yaml) {
+        Map<?, ?> raw;
+        try {
+            raw = YAML.readValue(yaml, Map.class);
+        } catch (IOException | IllegalArgumentException e) {
+            return;
+        }
+        if (raw == null || !(raw.get("profiles") instanceof Map<?, ?> profiles)) {
+            return;
+        }
+        List<String> names = new ArrayList<>();
+        List<String> detail = new ArrayList<>();
+        for (Map.Entry<?, ?> entry : profiles.entrySet()) {
+            if (!(entry.getValue() instanceof Map<?, ?> profile)
+                    || !(profile.get("autoCompactWindow") instanceof Number window)
+                    || !(profile.get("env") instanceof Map<?, ?> env)
+                    || !env.containsKey("CLAUDE_CODE_AUTO_COMPACT_WINDOW")) {
+                continue;
+            }
+            Object kind = profile.get("kind");
+            boolean claudeCode = kind == null || String.valueOf(kind).isBlank()
+                    || Profile.KIND_CLAUDE_CODE.equalsIgnoreCase(String.valueOf(kind));
+            Object envValue = env.get("CLAUDE_CODE_AUTO_COMPACT_WINDOW");
+            if (claudeCode && !String.valueOf(window).equals(String.valueOf(envValue))) {
+                String name = String.valueOf(entry.getKey());
+                names.add(name);
+                detail.add(name + " (autoCompactWindow=" + window
+                        + ", env.CLAUDE_CODE_AUTO_COMPACT_WINDOW=" + envValue + ")");
+            }
+        }
+        if (names.isEmpty()) {
+            return;
+        }
+        names.sort(String::compareTo);
+        detail.sort(String::compareTo);
+        log.warn("Claude Code profile(s) {} set disagreeing autoCompactWindow and env."
+                        + "CLAUDE_CODE_AUTO_COMPACT_WINDOW — the daemon starts anyway: {}. fleetd "
+                        + "#618 measured that CLAUDE_CODE_AUTO_COMPACT_WINDOW wins, so "
+                        + "autoCompactWindow is inert on these profiles. Set equal values on each "
+                        + "to resolve this — do not just delete the env var, since that LOWERS the "
+                        + "live window to autoCompactWindow's value rather than fixing anything.",
+                names, String.join(", ", detail));
+    }
+
     /**
      * Reject a profile whose {@code errorPattern} (fleetd #201 Unit 5) or {@code exhaustedPattern}
      * (CB-578 stage A) is not a valid Java regex, naming the profile, the key, and the parser's own

fleetd/src/main/java/dev/ltms/fleet/launch/ClaudeCodeArguments.java

@@ -0,0 +1,37 @@
+package dev.ltms.fleet.launch;
+
+import dev.ltms.fleet.config.FleetConfig;
+
+import java.util.ArrayList;
+import java.util.List;
+
+/** Arguments shared by every fleetd path that starts Claude Code. */
+public final class ClaudeCodeArguments {
+
+    private ClaudeCodeArguments() {
+    }
+
+    /**
+     * Append the configured Claude Code auto-compaction window when the profile opts in.
+     *
+     * <p>This flag and the environment variable {@code CLAUDE_CODE_AUTO_COMPACT_WINDOW} can
+     * disagree, and fleetd #618 measured which one Claude Code actually follows: the environment
+     * variable wins, ahead of this {@code --autocompact} flag, ahead of the settings file, ahead of
+     * clientdata, the experiment, and the model default. So when a profile sets both, the flag this
+     * method appends has NO effect — Claude Code reads {@code CLAUDE_CODE_AUTO_COMPACT_WINDOW}
+     * first and never consults the flag. {@link FleetConfig#load(java.nio.file.Path)} only WARNS
+     * when a Claude Code profile sets both to different values (see {@code
+     * FleetConfig.warnConflictingAutoCompactWindows}) — it does not stop the daemon from starting,
+     * and the launched session honours the env var, not this flag. Measured against Claude Code
+     * 2.1.278 (fleetd #618) — a later version could reorder this precedence.
+     */
+    public static List<String> withAutoCompactWindow(List<String> argv, FleetConfig.Profile profile) {
+        if (profile.autoCompactWindow() == null) {
+            return argv;
+        }
+        List<String> withAutoCompact = new ArrayList<>(argv);
+        withAutoCompact.add("--autocompact");
+        withAutoCompact.add(String.valueOf(profile.autoCompactWindow()));
+        return withAutoCompact;
+    }
+}

fleetd/src/main/java/dev/ltms/fleet/lead/LeadLauncher.java

@@ -8,6 +8,7 @@ import dev.ltms.fleet.herdr.PendingCloseMarker;
 import dev.ltms.fleet.herdr.Tab;
 import dev.ltms.fleet.herdr.Workspace;
 import dev.ltms.fleet.herdr.WorkspaceControl;
+import dev.ltms.fleet.launch.ClaudeCodeArguments;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
@@ -358,7 +359,8 @@ public final class LeadLauncher {
     }
 
     /**
-     * The lead's argv: the profile's own command, the model pin, and the bridge MCP mount.
+     * The lead's argv: the profile's own command, the model and auto-compaction pins, and the bridge
+     * MCP mount.
      *
      * <p>No {@code --append-system-prompt}. That flag carries the worker reply charter, and a lead
      * is not a worker — it reads its orchestration rules from the project's {@code CLAUDE.md} like
@@ -380,7 +382,7 @@ public final class LeadLauncher {
             argv.add("--model");
             argv.add(profile.model());
         }
-        return argv;
+        return profile.isOpenCode() ? argv : ClaudeCodeArguments.withAutoCompactWindow(argv, profile);
     }
 
     /**

fleetd/src/main/java/dev/ltms/fleet/lead/LeadRollover.java

@@ -231,7 +231,9 @@ public final class LeadRollover {
          * #status} could wrongly answer {@link #UNKNOWN} ("nothing was ever requested") for a roll
          * that is, in fact, actively running. This is not sticky: the deferred continuation
          * overwrites this same entry with a terminal state ({@link #ROLLED}, {@link
-         * #TURN_NEVER_SETTLED}, or {@link #CLEAR_NEVER_SETTLED}) once it finishes.
+         * #TURN_NEVER_SETTLED}, {@link #CLEAR_NEVER_SETTLED}, or {@link #FAILED}) once it finishes
+         * — including by throwing, which fleetd #615's catch in {@link #runRollover} now turns into
+         * {@link #FAILED} instead of leaving this entry stuck forever.
          */
         IN_PROGRESS,
         /**
@@ -253,6 +255,19 @@ public final class LeadRollover {
          * within {@code clearSettleSeconds} — {@code bootstrapText} was never sent.
          */
         CLEAR_NEVER_SETTLED,
+        /**
+         * fleetd #615: the deferred continuation threw a {@link RuntimeException} — most likely a
+         * {@link dev.ltms.fleet.herdr.HerdrException} out of one of the two unwrapped {@code
+         * agents.send} calls in {@link #runRollover} — and the continuation thread died with it.
+         * Before this state existed, that throw left {@link #outcomes} holding {@link #IN_PROGRESS}
+         * forever, because the production {@code continuationRunner} is a bare virtual thread with
+         * no uncaught-exception handler and nothing downstream of the throw ever ran to write a
+         * terminal outcome. {@code detail} names the exception, so a reader has something to act on
+         * — the same diagnostic style as {@link #TURN_NEVER_SETTLED} and {@link
+         * #CLEAR_NEVER_SETTLED}. The roll is dead at this point and does not retry itself; a stuck
+         * lead must {@link #open} a fresh request.
+         */
+        FAILED,
         /**
          * {@code token} names nothing this instance currently knows about: never issued by {@link
          * #open}, dropped by {@link #cancel}, or aged out of {@link #outcomes}'s bounded history.
@@ -493,8 +508,42 @@ public final class LeadRollover {
      * entirely after {@link #confirm} has returned to its caller — see this class's javadoc for the
      * four-step order. There is no result to return to by this point, so every outcome is logged
      * only.
+     *
+     * <p><strong>fleetd #615 — the whole body is wrapped in one {@code try}.</strong> The two {@code
+     * agents.send} calls below are not wrapped individually: {@code send} → {@code agentCall} →
+     * {@code herdr.call} can throw an unchecked {@link dev.ltms.fleet.herdr.HerdrException} (see
+     * {@code AgentControl.java}), and the production {@code continuationRunner} is a bare virtual
+     * thread with no uncaught-exception handler (see this class's public constructor). Before this
+     * fix, either throw killed the continuation thread silently, leaving the {@link
+     * RollState#IN_PROGRESS} entry {@link #confirm} wrote at hand-off stuck forever — {@link
+     * #status} had no way to tell a dead roll from one still genuinely running. The {@code catch}
+     * below is scoped to the method body rather than to each {@code send} call individually, so it
+     * also covers anything else added to this continuation later, not just today's two call sites —
+     * the same reasoning that put the write-a-terminal-outcome step at each of this method's other
+     * exits (see the {@link RollState#TURN_NEVER_SETTLED} and {@link RollState#CLEAR_NEVER_SETTLED}
+     * branches below) rather than inside the helpers that detect them.</p>
+     *
+     * <p>Only {@link RuntimeException} is caught, matching the local convention {@link
+     * #waitUntilAtTurnBoundary} already set around its own {@code agents.status} call — not the
+     * broader {@link Exception} or {@link Throwable}, which would also swallow something like an
+     * {@link OutOfMemoryError} this continuation has no business handling.</p>
      */
     private void runRollover(PendingRollover p, FleetConfig.LeadRollover cfg) {
+        try {
+            runRolloverUnguarded(p, cfg);
+        } catch (RuntimeException e) {
+            log.warn("lead-rollover: continuation for token={} lead={} threw {} — the roll is dead; "
+                            + "no further step in this continuation will run",
+                    p.token(), p.leadTerminal(), e.toString(), e);
+            outcomes.put(p.token(), new RollStatus(RollState.FAILED,
+                    "the roll's continuation threw " + e.toString() + " — the roll is dead and will "
+                            + "not retry itself; check the daemon log for the stack trace, then open() "
+                            + "a fresh rollover request"));
+        }
+    }
+
+    /** The actual body of {@link #runRollover}, unwrapped — see that method's javadoc for the catch. */
+    private void runRolloverUnguarded(PendingRollover p, FleetConfig.LeadRollover cfg) {
         String lead = p.leadTerminal();
         long rollStartMillis = nowMillis.getAsLong();
         TurnSettleResult turnResult = waitUntilAtTurnBoundary(lead, cfg.turnSettleSeconds());

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

@@ -8,6 +8,7 @@ import dev.ltms.fleet.guard.SubscriptionGuard;
 import dev.ltms.fleet.herdr.Agent;
 import dev.ltms.fleet.herdr.AgentControl;
 import dev.ltms.fleet.herdr.WorkspaceControl;
+import dev.ltms.fleet.launch.ClaudeCodeArguments;
 import dev.ltms.fleet.peer.Capability;
 import dev.ltms.fleet.peer.PeerLauncher;
 import org.slf4j.Logger;
@@ -298,7 +299,7 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
         // has neither MCP nor a charter — session flags must be added into a list we own.
         List<String> argv = mutableArgv(argvWithFleet(cfg, spec));
         String agentSessionId = applySessionIdentity(argv, spec.sessionName(), spec.resumeSessionId());
-        return new Launch(workerEnv, argvWithAutoCompact(argvWithModel(argv, cfg), cfg), agentSessionId);
+        return new Launch(workerEnv, ClaudeCodeArguments.withAutoCompactWindow(argvWithModel(argv, cfg), cfg), agentSessionId);
     }
 
     /**
@@ -908,30 +909,6 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
         return withModel;
     }
 
-    /**
-     * Pin a bounded auto-compaction window on the command line via {@code --autocompact <tokens>},
-     * opt-in per profile (CB-634's sibling ticket: a member that runs out of context dies mid-turn
-     * and its {@code fleet_reply} — the whole point of the turn — is lost with it; opencode already
-     * forces {@code compaction.auto: true} unconditionally, CB-523, but Claude Code has no equivalent
-     * and runs at the backend's own default window).
-     *
-     * <p>Mirrors {@link #argvWithModel}: appended after it, so it survives the {@code ccs <profile>}
-     * wrapper the same way {@code --model} does, and outranks env/settings and the operator's own
-     * {@code argv}. Verified: {@code claude 2.1.241 --help} lists {@code --autocompact <auto|tokens>}
-     * (either the literal {@code auto}, or an integer 100k–1M) — {@link FleetConfig#load} rejects a
-     * configured value outside that band before this ever runs, so the flag Claude Code receives here
-     * is always in range.
-     */
-    private static List<String> argvWithAutoCompact(List<String> argv, FleetConfig.Profile cfg) {
-        if (cfg.autoCompactWindow() == null) {
-            return argv;
-        }
-        List<String> withAutoCompact = mutableArgv(argv);
-        withAutoCompact.add("--autocompact");
-        withAutoCompact.add(String.valueOf(cfg.autoCompactWindow()));
-        return withAutoCompact;
-    }
-
     // --- Agent-returning convenience spawns (used by callers/tests that want the herdr Agent) ---
 
     /** Spawn a worker for the default profile in the resolved default cwd. */

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

@@ -1,8 +1,11 @@
 package dev.ltms.fleet.config;
 
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.spi.ILoggingEvent;
 import dev.ltms.fleet.auth.MemberRegistry;
 import dev.ltms.fleet.msg.LeadMailbox;
 import dev.ltms.fleet.peer.MemberRole;
+import dev.ltms.fleet.testing.CapturedLog;
 import org.junit.jupiter.api.Test;
 import org.junit.jupiter.api.io.TempDir;
 
@@ -93,6 +96,72 @@ class FleetConfigTest {
                 "unset means off — today's behaviour, unchanged");
     }
 
+    /**
+     * fleetd #601 review (measured 2026-09-22): this guard used to throw {@link
+     * IllegalStateException} and refuse to start. On a host with the conflict configured, that
+     * turned into a launchd restart loop with no readable cause, since {@code fleetd.yaml} is
+     * gitignored. It must now WARN and let the daemon start, and the warning must carry both values
+     * so an operator can fix the config without reading the source. Pinning the log line itself (via
+     * {@link CapturedLog}) rather than a snippet of production source text — the latter is the
+     * anti-pattern this repo avoids; the former is the actual observable behaviour a reader (or an
+     * alert on the log) depends on.
+     */
+    @Test
+    void aClaudeProfileWithConflictingAutoCompactFlagAndEnvironmentWindowLoadsAndWarnsWithBothValues(
+            @TempDir Path dir) throws Exception {
+        Path f = dir.resolve("conflicting-auto-compact-window.yaml");
+        Files.writeString(f, """
+                profiles:
+                  claude-profile:
+                    autoCompactWindow: 250000
+                    env:
+                      CLAUDE_CODE_AUTO_COMPACT_WINDOW: "300000"
+                """);
+
+        FleetConfig cfg;
+        List<String> warnings;
+        try (CapturedLog log = CapturedLog.at(FleetConfig.class, Level.WARN)) {
+            cfg = FleetConfig.load(f);
+            warnings = log.events().stream().map(ILoggingEvent::getFormattedMessage).toList();
+        }
+
+        assertEquals(250_000, cfg.profiles().get("claude-profile").autoCompactWindow(),
+                "the disagreement is reported, not corrected — the flag value still loads as-is");
+        assertEquals(1, warnings.size(), "exactly one warning for the one conflicting profile: " + warnings);
+        String warning = warnings.get(0);
+        assertTrue(warning.contains("claude-profile"), "names the offending profile: " + warning);
+        assertTrue(warning.contains("autoCompactWindow=250000"), "names the flag value: " + warning);
+        assertTrue(warning.contains("CLAUDE_CODE_AUTO_COMPACT_WINDOW=300000"), "names the env value: " + warning);
+    }
+
+    /**
+     * The negative probe paired with the test above (per fleetd #601 review): a warning that fires
+     * on every load and a warning that never fires read the same from a single test, so both must be
+     * checked. No conflict here — the flag and the env value agree — so no warning should be logged.
+     */
+    @Test
+    void aClaudeProfileWithEqualAutoCompactFlagAndEnvironmentWindowLoadsWithNoWarning(@TempDir Path dir)
+            throws Exception {
+        Path f = dir.resolve("equal-auto-compact-window.yaml");
+        Files.writeString(f, """
+                profiles:
+                  claude-profile:
+                    autoCompactWindow: 250000
+                    env:
+                      CLAUDE_CODE_AUTO_COMPACT_WINDOW: "250000"
+                """);
+
+        FleetConfig cfg;
+        List<ILoggingEvent> events;
+        try (CapturedLog log = CapturedLog.at(FleetConfig.class, Level.WARN)) {
+            cfg = FleetConfig.load(f);
+            events = log.events();
+        }
+
+        assertEquals(250_000, cfg.profiles().get("claude-profile").autoCompactWindow());
+        assertTrue(events.isEmpty(), "equal values must not warn: " + events);
+    }
+
     // ── fleetd #201 Unit 5: errorPattern ────────────────────────────────────────────────────────
 
     @Test

fleetd/src/test/java/dev/ltms/fleet/lead/LeadLauncherTest.java

@@ -32,13 +32,26 @@ class LeadLauncherTest {
                 Map.of("CLAUDE_CODE_AUTO_COMPACT_WINDOW", "300000"), null, null, true, null);
     }
 
+    private static FleetConfig.Profile profileWithAutoCompactWindow(String kind) {
+        return new FleetConfig.Profile(
+                "opus", null, "claude-opus-5", null, "FLEETD_WORKER_TOKEN",
+                List.of("ccs", "ltms"), "tab", "fleet", null,
+                "http://127.0.0.1:8765/mcp", null, null,
+                null, null, kind, Map.of(), null, null, true, null, null,
+                null, null, null, 250_000);
+    }
+
     private static FleetConfig configWith(FleetConfig.Leader lead) {
+        return configWith(lead, opusProfile());
+    }
+
+    private static FleetConfig configWith(FleetConfig.Leader lead, FleetConfig.Profile profile) {
         Map<String, FleetConfig.Leader> leaders = new LinkedHashMap<>();
         leaders.put("opus", lead);
         FleetConfig.Fleet fleet =
                 new FleetConfig.Fleet(leaders, Map.of(), Map.of(), Map.of(), null);
         return new FleetConfig(
-                null, null, Map.of("opus", opusProfile()), null, null, null, null, null,
+                null, null, Map.of("opus", profile), null, null, null, null, null,
                 null, null, fleet, null, "fixed", null).withDefaults();
     }
 
@@ -330,6 +343,36 @@ class LeadLauncherTest {
                 "--model is appended last so it outranks the ccs wrapper (CB-533)");
     }
 
+    @Test
+    void aClaudeLeadPassesItsConfiguredAutoCompactWindowToClaudeCode() {
+        FakeHerdr herdr = new FakeHerdr();
+        FleetConfig.Profile profile = profileWithAutoCompactWindow("claude-code");
+
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1), profile)).ensureLeads();
+
+        List<String> args = startedArgs(herdr);
+        assertEquals("250000", args.get(args.indexOf("--autocompact") + 1));
+    }
+
+    @Test
+    void aClaudeLeadWithNoAutoCompactWindowGetsNoAutoCompactFlag() {
+        FakeHerdr herdr = new FakeHerdr();
+
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads();
+
+        assertFalse(startedArgs(herdr).contains("--autocompact"));
+    }
+
+    @Test
+    void aNonClaudeLeadDoesNotGetAnAutoCompactFlag() {
+        FakeHerdr herdr = new FakeHerdr();
+        FleetConfig.Profile profile = profileWithAutoCompactWindow("opencode");
+
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1), profile)).ensureLeads();
+
+        assertFalse(startedArgs(herdr).contains("--autocompact"));
+    }
+
     /** A lead runs on the operator's subscription. Nothing may move it off. */
     @Test
     void theLeadEnvCarriesNoAnthropicBinding() {

fleetd/src/test/java/dev/ltms/fleet/lead/LeadRolloverTest.java

@@ -1359,4 +1359,92 @@ class LeadRolloverTest {
                     + "it were the measured wait duration: " + message);
         }
     }
+
+    // ---- fleetd #615: a HerdrException out of either unwrapped agents.send call must leave a ----
+    // ---- TERMINAL FAILED outcome, never a stuck IN_PROGRESS ---------------------------------------
+
+    @Test
+    @DisplayName("[fleetd #615 — 1] send() throwing on the /clear call leaves status(token) "
+            + "reporting FAILED, not stuck at IN_PROGRESS")
+    void sendThrowingOnClearLeavesStatusReportingFailed() throws IOException {
+        FakeHerdr fake = new FakeHerdr(); // default idle — the turn-settle wait passes immediately
+        HerdrClient throwsOnClear = new HerdrClient() {
+            @Override
+            public JsonNode call(String method, Object params) throws HerdrException {
+                if ("agent.prompt".equals(method) && String.valueOf(params).contains("/clear")) {
+                    throw new HerdrException("simulated herdr transport failure sending /clear");
+                }
+                return fake.call(method, params);
+            }
+
+            @Override
+            public void close() {
+                fake.close();
+            }
+        };
+        Path handover = writeHandover("handover contents");
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(throwsOnClear, cfg(handover.toString()), fixedClock(clock));
+
+        LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+        LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+        assertTrue(decision.accepted(), "every synchronous gate passes; the throw happens only "
+                + "inside the deferred continuation, which this test's synchronous runner has "
+                + "already run to completion by the time confirm() returns");
+
+        LeadRollover.RollStatus status = rollover.status(pending.token());
+        assertEquals(LeadRollover.RollState.FAILED, status.state(),
+                "a HerdrException out of the /clear send must leave a TERMINAL FAILED outcome — "
+                        + "before fleetd #615's fix, the continuation thread died silently and "
+                        + "status() was stuck reporting the IN_PROGRESS confirm() wrote at hand-off, "
+                        + "forever: got " + status.state() + " / " + status.detail());
+        assertNotEquals(LeadRollover.RollState.IN_PROGRESS, status.state());
+        assertTrue(status.detail().contains("HerdrException"), "the detail must name the exception "
+                + "so an operator reading status() has something to act on: " + status.detail());
+    }
+
+    @Test
+    @DisplayName("[fleetd #615 — 2] send() throwing on the bootstrap-text call (after /clear "
+            + "succeeded and the pane settled) also leaves status(token) reporting FAILED — a "
+            + "DIFFERENT exit from the /clear-throw case above")
+    void sendThrowingOnBootstrapTextLeavesStatusReportingFailed() throws IOException {
+        FakeHerdr fake = new FakeHerdr(); // default idle throughout — both settle waits pass promptly
+        HerdrClient throwsOnBootstrapText = new HerdrClient() {
+            @Override
+            public JsonNode call(String method, Object params) throws HerdrException {
+                if ("agent.prompt".equals(method) && String.valueOf(params).contains("read the handover file")) {
+                    throw new HerdrException("simulated herdr transport failure sending bootstrapText");
+                }
+                return fake.call(method, params);
+            }
+
+            @Override
+            public void close() {
+                fake.close();
+            }
+        };
+        Path handover = writeHandover("handover contents");
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(throwsOnBootstrapText, cfg(handover.toString()), fixedClock(clock));
+
+        LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+        LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+        assertTrue(decision.accepted(), "every synchronous gate passes; the throw happens only "
+                + "inside the deferred continuation, which this test's synchronous runner has "
+                + "already run to completion by the time confirm() returns");
+        assertEquals(1, promptCallCount(fake), "sanity: /clear was sent and settled — only the "
+                + "SECOND agent.prompt call (bootstrapText) threw");
+
+        LeadRollover.RollStatus status = rollover.status(pending.token());
+        assertEquals(LeadRollover.RollState.FAILED, status.state(),
+                "a HerdrException out of the bootstrapText send — a DIFFERENT exit from the /clear "
+                        + "throw, reached only after /clear already succeeded and the pane already "
+                        + "settled — must also leave a TERMINAL FAILED outcome, not a stuck "
+                        + "IN_PROGRESS: got " + status.state() + " / " + status.detail());
+        assertNotEquals(LeadRollover.RollState.IN_PROGRESS, status.state());
+        assertTrue(status.detail().contains("HerdrException"), "the detail must name the exception "
+                + "so an operator reading status() has something to act on: " + status.detail());
+    }
 }

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

@@ -1846,7 +1846,8 @@ class MessageServiceTest {
      * but backed by {@link ManualScheduler} instead of a real timer (fleetd #608): its tick never
      * fires on its own — a test drives it explicitly via {@link ManualScheduler#runDueTasks()}.
      */
-    private record ManualPushWiring(MessageService service, FakeHerdr leadHerdr, ManualScheduler scheduler)
+    private record ManualPushWiring(MessageService service, FakeHerdr leadHerdr, ManualScheduler scheduler,
+                                    ReplyPushLoop pushLoop)
             implements AutoCloseable {
         @Override
         public void close() {
@@ -1861,14 +1862,20 @@ class MessageServiceTest {
      * {@code anAlreadyCollectedTicketProducesNoNudge}, which sets it to 1 to prove exactly that.
      */
     private ManualPushWiring wireWithManualScheduler(int maxReminders, long backoffMs) {
+        return wireWithManualScheduler(maxReminders, backoffMs, System::nanoTime);
+    }
+
+    /** As above, with an injectable clock for tests that exercise terminal-ticket pruning. */
+    private ManualPushWiring wireWithManualScheduler(int maxReminders, long backoffMs,
+                                                      java.util.function.LongSupplier nowNanos) {
         PrimaryRegistry registry = new PrimaryRegistry(null);
         registry.recordDelegation(T, LEAD);
         FakeHerdr leadHerdr = new FakeHerdr();
         AgentControl leadAgents = new AgentControl(leadHerdr);
         ManualScheduler scheduler = new ManualScheduler();
         ReplyPushLoop pushLoop = new ReplyPushLoop(registry, leadAgents, inbox, scheduler, maxReminders, backoffMs);
-        MessageService service = new MessageService(agents, injector, rendezvous, inbox, pushLoop, null, System::nanoTime);
-        return new ManualPushWiring(service, leadHerdr, scheduler);
+        MessageService service = new MessageService(agents, injector, rendezvous, inbox, pushLoop, null, nowNanos);
+        return new ManualPushWiring(service, leadHerdr, scheduler, pushLoop);
     }
 
     private void awaitNudge(FakeHerdr leadHerdr) throws InterruptedException {
@@ -1959,24 +1966,35 @@ class MessageServiceTest {
 
     @Test
     void severalAsyncTicketsFinishingTogetherProduceOneCoalescedNudge() throws Exception {
-        try (var wiring = wireWithPushLoop(1, 300)) { // wide backoff: both tickets land before the tick fires
-            String first = wiring.service().sendAsync(T, "first task");
-            awaitWaiting();
-            injectDelivery();
-            assertTrue(rendezvous.resolve(T, "first done"));
-            // Settle without polling: poll() itself marks a ticket collected (that's the point of
-            // anAlreadyCollectedTicketProducesNoNudge above) — using it here to detect completion
-            // would collect the ticket before the coalescing this test checks ever gets a chance.
-            Thread.sleep(100);
+        // A 1ms backoff is due immediately. ManualScheduler still cannot run it until this test
+        // explicitly calls runDueTasks(), so both terminal tickets join one scheduled tick.
+        try (var wiring = wireWithManualScheduler(1, 1)) {
+            String first;
+            String second;
+            java.util.concurrent.CountDownLatch firstTerminalReached = new java.util.concurrent.CountDownLatch(1);
+            wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(firstTerminalReached::countDown);
+            try {
+                first = wiring.service().sendAsync(T, "first task");
+                awaitWaiting();
+                injectDelivery();
+                assertTrue(rendezvous.resolve(T, "first done"));
+                assertTrue(firstTerminalReached.await(5, TimeUnit.SECONDS),
+                        "the first ticket never reached its terminal phase");
 
-            String second = wiring.service().sendAsync(T, "second task");
-            awaitWaiting();
-            injectDelivery();
-            assertTrue(rendezvous.resolve(T, "second done"));
-            Thread.sleep(100);
+                java.util.concurrent.CountDownLatch secondTerminalReached = new java.util.concurrent.CountDownLatch(1);
+                wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(secondTerminalReached::countDown);
+                second = wiring.service().sendAsync(T, "second task");
+                awaitWaiting();
+                injectDelivery();
+                assertTrue(rendezvous.resolve(T, "second done"));
+                assertTrue(secondTerminalReached.await(5, TimeUnit.SECONDS),
+                        "the second ticket never reached its terminal phase");
+            } finally {
+                wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(null);
+            }
 
-            awaitNudge(wiring.leadHerdr());
-            Thread.sleep(200); // settle — nothing more should arrive beyond the one coalesced nudge
+            assertEquals(1, wiring.scheduler().runDueTasks(),
+                    "both terminal tickets must coalesce onto one scheduled tick");
             long nudgeCount = wiring.leadHerdr().calls.stream()
                     .filter(c -> c.method().equals("agent.prompt")).count();
             assertEquals(1, nudgeCount, "two tickets finishing together must produce ONE nudge, not two");
@@ -2055,7 +2073,8 @@ class MessageServiceTest {
 
     @Test
     void answeringAQuestionStopsFurtherNudgesAboutIt() throws Exception {
-        try (var wiring = wireWithPushLoop(5, 50)) {
+        // A 1ms backoff is due immediately, but ManualScheduler only ticks when this test asks it to.
+        try (var wiring = wireWithManualScheduler(5, 1)) {
             String ticket = wiring.service().sendAsync(T, "task that asks");
             awaitWaiting();
             injectDelivery();
@@ -2063,8 +2082,9 @@ class MessageServiceTest {
             CompletableFuture<MessageService.AskResult> ask = CompletableFuture.supplyAsync(
                     () -> wiring.service().ask(T, "which config file?", 5000));
             MessageService.TaskView asking = awaitTicketPhaseOn(wiring.service(), ticket, MessageService.Phase.ASKING);
+            awaitQuestionPendingOn(wiring.pushLoop(), LEAD, asking.turnId());
 
-            awaitNudge(wiring.leadHerdr());
+            assertEquals(1, wiring.scheduler().runDueTasks(), "the open question must have one scheduled tick");
             long callsBeforeAnswer = wiring.leadHerdr().calls.stream()
                     .filter(c -> c.method().equals("agent.prompt")).count();
 
@@ -2072,12 +2092,22 @@ class MessageServiceTest {
                     () -> wiring.service().answer(asking.turnId(), "config.yaml", 5000));
             assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
             awaitWaiting();
+            java.util.concurrent.CountDownLatch terminalReached = new java.util.concurrent.CountDownLatch(1);
+            wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(terminalReached::countDown);
             assertTrue(rendezvous.resolve(T, "done"));
+            try {
+                assertTrue(terminalReached.await(5, TimeUnit.SECONDS),
+                        "the answered ticket never reached its terminal phase");
+            } finally {
+                wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(null);
+            }
             answer.get(5, TimeUnit.SECONDS);
 
-            // Let several more ticks (and the ticket's own now-legitimate terminal nudge) fire —
-            // none of them may still name the question's turnId, which is closed.
-            Thread.sleep(300);
+            // Run the ticket's legitimate terminal nudge and every later scheduled tick through its
+            // reminder cap. None may still name the closed question.
+            for (int tick = 0; tick < 6; tick++) {
+                assertEquals(1, wiring.scheduler().runDueTasks(), "expected one scheduled reminder tick");
+            }
             boolean anyNamesClosedQuestion = wiring.leadHerdr().calls.stream()
                     .filter(c -> c.method().equals("agent.prompt"))
                     .skip(callsBeforeAnswer)
@@ -2160,35 +2190,45 @@ class MessageServiceTest {
         // decideTickets hits the cap and STOPs — activeLeads drops the lead, but (before the fix)
         // pendingTickets never drops the ticket. That is the exact "cap already STOPped" branch of
         // the bug report, reached deterministically rather than by timing it against a live tick.
-        try (var wiring = wireWithPushLoop(1, 50, clock::get)) {
-            String stale = wiring.service().sendAsync(T, "first task");
-            awaitWaiting();
-            injectDelivery();
-            assertTrue(rendezvous.resolve(T, "stale result"));
+        // A 1ms backoff is due immediately, but ManualScheduler runs only the ticks below.
+        try (var wiring = wireWithManualScheduler(1, 1, clock::get)) {
+            String stale;
+            String fresh;
+            java.util.concurrent.CountDownLatch staleTerminalReached = new java.util.concurrent.CountDownLatch(1);
+            wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(staleTerminalReached::countDown);
+            try {
+                stale = wiring.service().sendAsync(T, "first task");
+                awaitWaiting();
+                injectDelivery();
+                assertTrue(rendezvous.resolve(T, "stale result"));
+                assertTrue(staleTerminalReached.await(5, TimeUnit.SECONDS),
+                        "the stale ticket never reached its terminal phase");
 
-            // Let the reminder loop fire its one nudge and hit the cap (STOP removes it from
-            // activeLeads; pendingTickets is untouched either way — that asymmetry is the bug).
-            awaitNudge(wiring.leadHerdr());
-            Thread.sleep(300);
-            assertTrue(wiring.leadHerdr().lastCall("agent.prompt").params().toString().contains(stale),
-                    "sanity: the stale ticket's own reminder must have fired first");
+                // Fire the stale ticket's one nudge, then its cap tick (STOP removes it from activeLeads;
+                // pendingTickets is untouched either way — that asymmetry is the bug).
+                assertEquals(1, wiring.scheduler().runDueTasks(), "the stale ticket must have one nudge tick");
+                assertEquals(1, wiring.scheduler().runDueTasks(), "the stale ticket must have one cap tick");
+                assertTrue(wiring.leadHerdr().lastCall("agent.prompt").params().toString().contains(stale),
+                        "sanity: the stale ticket's own reminder must have fired first");
 
-            // Cross the TTL — a real clock would need 10 minutes; the injected one does it instantly.
-            clock.addAndGet(MessageService.TICKET_TTL_NANOS + TimeUnit.SECONDS.toNanos(1));
+                // Cross the TTL — a real clock would need 10 minutes; the injected one does it instantly.
+                clock.addAndGet(MessageService.TICKET_TTL_NANOS + TimeUnit.SECONDS.toNanos(1));
 
-            // A second, unrelated ticket to the same target/lead reaches sendAsync, which prunes.
-            String fresh = wiring.service().sendAsync(T, "second task");
-            awaitWaiting();
-            injectDelivery();
-            assertTrue(rendezvous.resolve(T, "fresh result"));
-
-            // The fresh ticket restarts the (now-dormant) reminder loop with its own nudge.
-            long before = wiring.leadHerdr().calls.stream().filter(c -> c.method().equals("agent.prompt")).count();
-            long deadline = System.currentTimeMillis() + 3000;
-            while (wiring.leadHerdr().calls.stream().filter(c -> c.method().equals("agent.prompt")).count() <= before
-                    && System.currentTimeMillis() < deadline) {
-                Thread.sleep(10);
+                // A second, unrelated ticket to the same target/lead reaches sendAsync, which prunes.
+                java.util.concurrent.CountDownLatch freshTerminalReached = new java.util.concurrent.CountDownLatch(1);
+                wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(freshTerminalReached::countDown);
+                fresh = wiring.service().sendAsync(T, "second task");
+                awaitWaiting();
+                injectDelivery();
+                assertTrue(rendezvous.resolve(T, "fresh result"));
+                assertTrue(freshTerminalReached.await(5, TimeUnit.SECONDS),
+                        "the fresh ticket never reached its terminal phase");
+            } finally {
+                wiring.service().setAfterFinishAsyncTaskCompleteHookForTest(null);
             }
+
+            // The fresh ticket restarts the now-dormant reminder loop with its own nudge.
+            assertEquals(1, wiring.scheduler().runDueTasks(), "the fresh ticket must have one nudge tick");
             String latestNudge = wiring.leadHerdr().lastCall("agent.prompt").params().toString();
             assertTrue(latestNudge.contains(fresh), "the fresh ticket's nudge must still arrive: " + latestNudge);
             assertFalse(latestNudge.contains(stale),