fleetd#219: OpenCodeLauncher config/discovery roots must not assume fleetd's own filesystem

Site 1 (config root): under memberHerdrSocket, writeConfig() now places the
ephemeral opencode.json directory under worktreeRoot and shares it read-only
with worktreeGroup, reusing EnvAllowListScrub#shareWithGroup (widened to
package-private and generalized) — the same mechanism #213 built for the
ZDOTDIR scrub, rather than a second copy. Unlike the ZDOTDIR scrub's
degrade-to-overlay fallback, a missing worktreeRoot/worktreeGroup here
REFUSES the spawn (IllegalStateException from buildLaunch): this file is the
member's only way to learn where the bridge MCP is, so writing it somewhere
unreadable would just produce an undeliverable member with no signal
pointing at the cause. memberHerdrSocket absent stays byte-identical.

Site 2 (discovery root): under memberHerdrSocket, agentSessionId() now
declares session discovery unavailable and logs one WARN per launcher
instance instead of silently scanning fleetd's own $HOME (opencode.db lives
under the MEMBER's home under this config key). Decision + reasoning for why
this is a declare-unavailable rather than a new config key is in
defaultDiscoveryRoot()'s javadoc.

Widened HerdrPeerLauncher#memberHerdrSocketConfigured/memberScrubParentDir/
memberGroup to package-private so OpenCodeLauncher reuses the exact same
config resolution rather than re-deriving it.

Same-shape finding (not fixed, out of scope): ClaudeCodeLauncher#writeCharterFile
(line ~465) writes the role-charter temp file via Files.createTempFile with no
directory argument, i.e. under java.io.tmpdir — the same site-1 shape, unfixed
for the Claude Code adapter.

fleetd/fleetd.example.yaml

@@ -117,6 +117,15 @@ herdrSocket: ~/.config/herdr/herdr.sock
 # Optional socket for member panes. Omit this to use herdrSocket for both leads and members.
 # memberHerdrSocket: /Users/member/.config/herdr/herdr.sock
 
+# fleetd #213: the login shell the member OS user (memberHerdrSocket above) actually runs. ONLY
+# read when memberHerdrSocket is set — fleetd's own $SHELL says nothing about a pane running
+# under a different OS user, and there is no channel to ask herdr for that user's shell, so this
+# must be told rather than guessed. Absent, blank, or anything not ending in "zsh" is treated the
+# same as "not zsh": the memberCredentials.policy: allow-list ZDOTDIR scrub (see worktreeGroup
+# below) is skipped in favour of the weaker CB-596 sentinel overlay — a degraded control, never a
+# refusal to spawn. When memberHerdrSocket is absent this key is never consulted at all.
+# memberLoginShell: /bin/zsh
+
 # How member sessions are spawned. Define one or more named profiles (backends) under
 # `profiles`; each key is the profile name (also the ccs profile). A profile says only WHICH
 # BACKEND — model, CLI adapter, credentials, cost. It says nothing about what a member spawned on
@@ -643,6 +652,13 @@ guard:
 # CAUTION: this isolates credentials, not the repository — a member in the group can still
 # write the operator's git objects and refs in the shared repo. The operator running fleetd
 # must already be a member of the named group, or every provisioning spawn fails loudly.
+#
+# fleetd #213: this is also the ONE group the memberCredentials.policy: allow-list ZDOTDIR scrub
+# reuses when memberHerdrSocket is set — deliberately not a second config key. Under
+# memberHerdrSocket, the scrub directory is generated under worktreeRoot (never java.io.tmpdir,
+# which the member OS user cannot reach) and shared read-only with this group. If worktreeGroup
+# is unset while memberHerdrSocket is set, the scrub cannot be guaranteed reachable by the member,
+# so fleetd falls back to the weaker CB-596 sentinel overlay instead (a WARN names the gap).
 # worktreeGroup: fleet-workers
 
 # Session lifecycle limits (CB-303). All knobs are opt-in; omit or set to null to keep

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

@@ -84,6 +84,18 @@ import java.util.Set;
  *                      <strong>This isolates credentials, not the repository</strong>: a member in
  *                      the group can still write the operator's git objects and refs in the shared
  *                      repo. See {@link dev.ltms.fleet.session.Worktrees#shareWithGroup}.
+ * @param memberLoginShell fleetd #213: the login shell the member's OS user actually runs, ONLY
+ *                      meaningful (and only ever read) when {@code memberHerdrSocket} is
+ *                      configured — that mode spawns member panes under a different OS user than
+ *                      fleetd's own process, so fleetd's own {@code $SHELL} says nothing about what
+ *                      that pane runs. There is no channel to ask herdr for another user's shell, so
+ *                      this must be told, never guessed. {@code null}/blank (or a value not ending
+ *                      in {@code zsh}) is treated the same as "not zsh": the {@code
+ *                      memberCredentials.policy: allow-list} ZDOTDIR scrub is skipped in favour of
+ *                      the CB-596 sentinel overlay — a degraded control, never a refusal to spawn.
+ *                      When {@code memberHerdrSocket} is NOT configured this field is never
+ *                      consulted at all; fleetd keeps reading its own {@code $SHELL}, exactly as
+ *                      before this field existed.
  */
 @JsonIgnoreProperties(ignoreUnknown = true)
 public record FleetConfig(
@@ -107,7 +119,20 @@ public record FleetConfig(
         Integer quarantineCooldownSeconds,
         MemberCredentials memberCredentials,
         Coordinator coordinator,
-        String worktreeGroup) {
+        String worktreeGroup,
+        String memberLoginShell) {
+
+    /** Back-compat form before the {@code memberLoginShell} key was added. */
+    public FleetConfig(Bind bind, String herdrSocket, String memberHerdrSocket, Map<String, Profile> profiles,
+                         Guard guard, String worktreeRoot, Lifecycle lifecycle, Integer spawnReadyTimeoutMs,
+                         Integer spawnReadyPollMs, Broker broker, Primary primary, Fleet fleet,
+                         LeadHeartbeat leadHeartbeat, Health health, String placement, Auth auth,
+                         ConfigReload configReload, Integer quarantineCooldownSeconds,
+                         MemberCredentials memberCredentials, Coordinator coordinator, String worktreeGroup) {
+        this(bind, herdrSocket, memberHerdrSocket, profiles, guard, worktreeRoot, lifecycle, spawnReadyTimeoutMs,
+                spawnReadyPollMs, broker, primary, fleet, leadHeartbeat, health, placement, auth,
+                configReload, quarantineCooldownSeconds, memberCredentials, coordinator, worktreeGroup, null);
+    }
 
     /** Back-compat form before the {@code worktreeGroup} key was added. */
     public FleetConfig(Bind bind, String herdrSocket, String memberHerdrSocket, Map<String, Profile> profiles,
@@ -118,7 +143,7 @@ public record FleetConfig(
                          MemberCredentials memberCredentials, Coordinator coordinator) {
         this(bind, herdrSocket, memberHerdrSocket, profiles, guard, worktreeRoot, lifecycle, spawnReadyTimeoutMs,
                 spawnReadyPollMs, broker, primary, fleet, leadHeartbeat, health, placement, auth,
-                configReload, quarantineCooldownSeconds, memberCredentials, coordinator, null);
+                configReload, quarantineCooldownSeconds, memberCredentials, coordinator, null, null);
     }
 
     /** Back-compat form before the {@code coordinator:} block was added. */
@@ -1346,7 +1371,7 @@ public record FleetConfig(
             "bind", "herdrSocket", "memberHerdrSocket", "profiles", "guard", "worktreeRoot",
             "lifecycle", "spawnReadyTimeoutMs", "spawnReadyPollMs", "broker", "primary", "fleet",
             "leadHeartbeat", "health", "placement", "auth", "configReload", "quarantineCooldownSeconds",
-            "memberCredentials", "coordinator", "worktreeGroup");
+            "memberCredentials", "coordinator", "worktreeGroup", "memberLoginShell");
 
     /** Load and validate config from {@code path}. */
     public static FleetConfig load(Path path) {
@@ -1966,9 +1991,12 @@ public record FleetConfig(
         // and this ticket's Coordinator is config-only anyway (nothing yet reads it at startup).
         // worktreeGroup is left as-is (fleetd #185 stage 3): null/blank is "off", and there is no
         // sane non-null default — an OS group name is operator-specific.
+        // memberLoginShell is left as-is (fleetd #213), like worktreeGroup: null/blank is "not
+        // configured", and there is no sane non-null default — a member's login shell is
+        // operator-specific and only meaningful when memberHerdrSocket is also set.
         return new FleetConfig(b, herdrSocket, memberHerdrSocket, profiles, g, worktreeRoot, l, timeout, pollMs,
                 broker, primary, f, leadHeartbeat, health, placementOrDefault, a, configReload,
-                quarantineCooldown, mc, coordinator, worktreeGroup);
+                quarantineCooldown, mc, coordinator, worktreeGroup, memberLoginShell);
     }
 
     /**

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

@@ -355,7 +355,12 @@ public final class CompletionResolver implements TurnListener {
         }
         String backendError = firstMatchingLine(raw, BACKEND_ERROR);
         if (backendError != null) {
-            fail(target, turn, "member " + target + " ended on a backend error: " + backendError);
+            // Carry the pane, not just the matched line — the same fleetd#164 rule the normal path
+            // above applies. Here it matters more, not less: the trimmed block was empty, so the raw
+            // scrape is the ONLY copy of whatever the member managed to say. Clipped to the same cap
+            // the normal path uses, since a raw screen has no boundary trimming to bound it.
+            fail(target, turn, "member " + target + " ended on a backend error: " + backendError
+                    + "\n--- pane tail ---\n" + clip(raw));
             return true;
         }
         return false;

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

@@ -278,18 +278,22 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
 
     /**
      * Add the Claude-specific session-identity flags to {@code argv} and return the peer's OWN
-     * session id — the resume handle. A resume request passes the prior id via {@code -r} and
-     * returns that id; a fresh named session mints a new UUID, passes it via {@code --session-id},
-     * and returns the mint. The bridge's logical name rides along as {@code -n} when present. When
-     * <em>no</em> identity is requested (sessionName and resumeSessionId both blank) this adds
-     * nothing and returns {@code null}, keeping the legacy no-identity launch byte-identical.
+     * session id — the resume handle. A resume passes the prior id via {@code -r} and returns
+     * that id; every other spawn mints a new UUID, passes it via {@code --session-id}, and
+     * returns the mint. The bridge's logical name rides along as {@code -n} when present.
+     *
+     * <p>fleetd #214: the mint is unconditional. A plain {@code fleet_spawn} passes neither
+     * sessionName nor resumeSessionId, yet the member must still be resumable, and this id is
+     * the only resume handle a claude-code member has — unlike opencode, nothing resolves it
+     * after the launch. Checked against the real binary (claude 2.1.252): the flag is safe on
+     * every spawn. The binary takes only a valid UUID — it refuses any other value at argument
+     * parsing ("Invalid session ID. Must be a valid UUID.") — so the {@code UUID.randomUUID()}
+     * mint is required, not incidental. The only flag interaction the binary documents is with
+     * {@code -r} (both claim the session id), and the resume branch above never combines the two.
      */
     private static String applySessionIdentity(List<String> argv, String sessionName, String resumeSessionId) {
         boolean resuming = resumeSessionId != null && !resumeSessionId.isBlank();
         boolean named = sessionName != null && !sessionName.isBlank();
-        if (!resuming && !named) {
-            return null; // no identity requested — keep the legacy launch byte-identical
-        }
         if (named) {
             argv.add("-n");
             argv.add(sessionName);
@@ -320,11 +324,19 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
      *
      * <p>CB-618: Claude Code refuses to start when BOTH {@code --append-system-prompt} and
      * {@code --append-system-prompt-file} are on the command line ("Cannot use both ... Please use
-     * only one"), so the two charters can never travel on separate flags. When both are present they
-     * are concatenated into the one file, role charter first and reply charter last — last is where
-     * the reply rule must sit, because it is the rule that must survive. When only the reply charter
-     * is present it keeps its proven inline {@code --append-system-prompt} delivery, which is also
-     * the only form that reaches a member with no repo checkout.
+     * only one"), so the two charters can never travel on separate flags. They are concatenated
+     * into the one file, role charter first and reply charter last — last is where the reply rule
+     * must sit, because it is the rule that must survive.
+     *
+     * <p>fleetd #220: a lone reply charter used to ride inline on {@code --append-system-prompt},
+     * which put ~800 bytes of prose on the command line herdr types into the pane. That line is
+     * capped at {@value HerdrPeerLauncher#PANE_COMMAND_BYTE_LIMIT} bytes by the pty itself, and
+     * everything past the cap is dropped with no error from any layer. The charter alone left about
+     * 50 bytes of headroom, so adding one flag ({@code --session-id}, fleetd #214) truncated the
+     * LAST argument instead — {@code --autocompact 250000} arrived as {@code --autocompact 25},
+     * claude rejected it, and every claude-code spawn died as an unexplained readiness timeout.
+     * The charter now always travels as a file, which takes the prose off the command line for
+     * good; {@link HerdrPeerLauncher#checkPaneCommandFits} is the backstop for whatever grows next.
      */
     private List<String> argvWithFleet(FleetConfig.Profile cfg, LaunchSpec spec) {
         String roleCharter = nonBlank(spec.roleCharter());
@@ -341,16 +353,13 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
             argv.add("--mcp-config");
             argv.add(mcpConfigJson(cfg));
         }
-        // Combine the charters in order role -> reply, dropping any that are absent. When two
-        // or more survive they must ride one --append-system-prompt-file (CB-618 forbids the inline
-        // flag and the file flag together). A lone reply charter keeps its proven inline delivery.
+        // Combine the charters in order role -> reply, dropping any that are absent. They ride one
+        // --append-system-prompt-file (CB-618 forbids the inline flag and the file flag together),
+        // always — fleetd #220: charter prose on the command line overruns the pane's byte cap.
         List<String> charters = new java.util.ArrayList<>(2);
         if (roleCharter != null) charters.add(roleCharter);
         if (replyCharter != null) charters.add(replyCharter);
-        if (charters.size() == 1 && replyCharter != null && roleCharter == null) {
-            argv.add("--append-system-prompt");
-            argv.add(replyCharter);
-        } else if (!charters.isEmpty()) {
+        if (!charters.isEmpty()) {
             argv.add("--append-system-prompt-file");
             argv.add(writeCharterFile(String.join("\n\n", charters)).toString());
         }

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

@@ -7,6 +7,9 @@ import java.io.IOException;
 import java.io.UncheckedIOException;
 import java.nio.file.Files;
 import java.nio.file.Path;
+import java.nio.file.attribute.GroupPrincipal;
+import java.nio.file.attribute.PosixFileAttributeView;
+import java.nio.file.attribute.PosixFilePermissions;
 import java.time.Duration;
 import java.time.Instant;
 import java.util.ArrayList;
@@ -116,6 +119,77 @@ public final class EnvAllowListScrub {
         }
     }
 
+    /**
+     * fleetd #213: as {@link #generate(Path, Set)}, plus share the generated directory with
+     * {@code group} — the member OS user's group (the operator's existing {@code worktreeGroup:}
+     * name, reused rather than inventing a second one) — so a member running under a different OS
+     * user than fleetd's own process can still read what it needs from a directory placed outside
+     * {@code java.io.tmpdir}. {@code group} null/blank ⇒ identical to {@link #generate(Path, Set)};
+     * this is the single-daemon (no {@code memberHerdrSocket}) shape, where the pane is fleetd's own
+     * uid and no group sharing is needed.
+     *
+     * @throws UncheckedIOException also when {@code group} does not resolve on this host, or a
+     *                              group-ownership/permission call is refused — the same "fail
+     *                              loudly rather than start unprotected" contract as above: a scrub
+     *                              the configured member user cannot even read is not a working
+     *                              control.
+     */
+    public static Path generate(Path parentDir, Set<String> allowedNames, String group) {
+        Path dir = generate(parentDir, allowedNames);
+        if (group != null && !group.isBlank()) {
+            shareWithGroup(dir, group);
+        }
+        return dir;
+    }
+
+    /**
+     * chgrp/chmod-equivalent over a freshly generated directory and the flat files already written
+     * into it: owner keeps full access, {@code group} gets traverse+read on the directory ({@code
+     * rwxr-x---}, so a member process — a login shell reading it via {@code ZDOTDIR}, or another
+     * process simply opening a file under it — running under that group can find and read the
+     * files) and read-only on each file ({@code rw-r-----}) — deliberately no group WRITE anywhere,
+     * since a member never needs to add or change what fleetd generated. (For the ZDOTDIR scrub
+     * specifically, this also means the scrub script's own report write inside the pane fails
+     * closed rather than open — see {@code scrub.zsh}'s trailing {@code 2>/dev/null} — which {@link
+     * dev.ltms.fleet.member.HerdrPeerLauncher#releaseZdotdir} already treats as "cannot be
+     * confirmed to have run" rather than success.)
+     *
+     * <p>Package-private and named generically on purpose: fleetd #213 built this for the ZDOTDIR
+     * scrub directory, and fleetd #219 reuses it verbatim for {@link
+     * dev.ltms.fleet.member.OpenCodeLauncher}'s ephemeral {@code opencode.json} directory — both are
+     * "a fleetd-generated directory of flat files that a different-uid member process must read but
+     * never write," so the sharing mechanism is shared rather than copied a second time.
+     */
+    static void shareWithGroup(Path dir, String group) {
+        try {
+            GroupPrincipal principal = dir.getFileSystem().getUserPrincipalLookupService()
+                    .lookupPrincipalByGroupName(group);
+            setGroupAndPermissions(dir, principal, "rwxr-x---");
+            try (Stream<Path> entries = Files.list(dir)) {
+                for (Path file : entries.toList()) {
+                    setGroupAndPermissions(file, principal, "rw-r-----");
+                }
+            }
+        } catch (IOException e) {
+            throw new UncheckedIOException("cannot share generated directory " + dir + " with group '"
+                    + group + "' — the group must exist, and the fleetd operator ("
+                    + System.getProperty("user.name") + ") must be a member of it", e);
+        } catch (UnsupportedOperationException e) {
+            throw new UncheckedIOException("cannot share generated directory " + dir + " with group '"
+                    + group + "' — this filesystem does not support POSIX group ownership",
+                    new IOException(e));
+        }
+    }
+
+    private static void setGroupAndPermissions(Path path, GroupPrincipal group, String perms) throws IOException {
+        PosixFileAttributeView view = Files.getFileAttributeView(path, PosixFileAttributeView.class);
+        if (view == null) {
+            throw new IOException("POSIX file attributes are not supported for " + path);
+        }
+        view.setGroup(group);
+        Files.setPosixFilePermissions(path, PosixFilePermissions.fromString(perms));
+    }
+
     /** One operator-sourcing startup file: source the {@code $HOME} counterpart, change nothing else. */
     private static String homeSourcingFile(String name) {
         return """

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

@@ -682,6 +682,7 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
         // Protocol 19 resolves the executable from the agent kind (== namePrefix here), so
         // argv[0] — the configured executable — is dropped and only the extra args are passed.
         List<String> args = argv.isEmpty() ? argv : argv.subList(1, argv.size());
+        checkPaneCommandFits(cfg, argv);
         HerdrException last = null;
         for (int attempt = 0; attempt < NAME_RETRIES; attempt++) {
             long seq = nameSeq.incrementAndGet();
@@ -697,6 +698,59 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
         throw last;
     }
 
+    /**
+     * fleetd #220: herdr does not exec the launch command — it TYPES it into the pane as one line,
+     * and a pty line buffer holds only {@value #PANE_COMMAND_BYTE_LIMIT} bytes (BSD/macOS {@code
+     * MAX_CANON}). Everything past that byte is dropped. Nothing reports it: herdr answers "agent
+     * started", the backend exits on the mangled argument it was handed, the pane closes, and the
+     * only symptom is {@link #waitUntilInjectableOrThrow} timing out 20 seconds later with no
+     * reason. That is exactly how #214 broke every claude-code spawn — one 50-byte flag pushed a
+     * 978-byte command to 1028, and the tail that got cut was {@code --autocompact 250000}.
+     *
+     * <p>So measure it here and refuse, loudly and immediately, rather than spawn something that
+     * cannot work. The estimate is deliberately conservative: fleetd cannot see herdr's quoting, so
+     * every argument is charged its own bytes plus a separator and a quote pair. An over-estimate
+     * costs a clear error at a length that was already unsafe; an under-estimate would let the
+     * silent truncation back in.
+     *
+     * @throws PeerUnreachableException when the command cannot fit — the same failure the spawn
+     *                                  would have hit anyway, named at the point it is still
+     *                                  explainable
+     */
+    private void checkPaneCommandFits(FleetConfig.Profile cfg, List<String> argv) {
+        int bytes = 0;
+        String longest = null;
+        int longestBytes = 0;
+        for (String arg : argv) {
+            int argBytes = arg == null ? 0 : arg.getBytes(java.nio.charset.StandardCharsets.UTF_8).length;
+            bytes += argBytes + QUOTING_OVERHEAD_PER_ARG;
+            if (argBytes > longestBytes) {
+                longestBytes = argBytes;
+                longest = arg;
+            }
+        }
+        if (bytes <= PANE_COMMAND_BYTE_LIMIT) {
+            return;
+        }
+        String culprit = longest == null ? "<none>"
+                : longest.substring(0, Math.min(longest.length(), 60)) + (longest.length() > 60 ? "…" : "");
+        throw new PeerUnreachableException(
+                "launch command for profile " + cfg.profile() + " is about " + bytes + " bytes, over the "
+                        + PANE_COMMAND_BYTE_LIMIT + "-byte limit of the pane line herdr types it into. "
+                        + "The pty would drop the tail silently and the backend would exit on a mangled "
+                        + "argument. Longest argument is " + longestBytes + " bytes: " + culprit
+                        + " — move it off the command line (a file flag) or shorten it.");
+    }
+
+    /**
+     * The pty line buffer herdr types a launch command into: BSD/macOS {@code MAX_CANON}. Not a
+     * fleetd choice and not configurable — see {@link #checkPaneCommandFits}.
+     */
+    static final int PANE_COMMAND_BYTE_LIMIT = 1024;
+
+    /** Per-argument allowance for the separating space and a shell quote pair fleetd cannot see. */
+    private static final int QUOTING_OVERHEAD_PER_ARG = 3;
+
     /** Start the agent into {@code paneId}, waiting out the seed shell's boot with the sleeper. */
     private Agent startAwaitingShellPrompt(String name, List<String> args, String paneId) {
         HerdrException busy = null;
@@ -860,20 +914,51 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
      */
     private void waitUntilInjectableOrThrow(String paneId) {
         long deadline = nowMillis.getAsLong() + spawnReadyTimeoutMs;
+        Object lastStatus = null;
         while (nowMillis.getAsLong() < deadline) {
-            if (agents.status(paneId).injectable()) {
+            var status = agents.status(paneId);
+            lastStatus = status;
+            if (status.injectable()) {
                 log.debug("peer pane={} reached injectable state", paneId);
                 return;
             }
             sleeper.run();
         }
-        log.warn("peer pane={} did not become injectable within {}ms — closing", paneId, spawnReadyTimeoutMs);
+        // fleetd #220: read the pane BEFORE stop() closes it. Without this the gate says only that
+        // it timed out, which is true of every cause — a backend that never launched, a binary that
+        // rejected an argument and exited, a trust prompt, a login shell that hung. The pane holds
+        // the one copy of that answer and it is destroyed a line later.
+        log.warn("peer pane={} did not become injectable within {}ms (last status {}) — closing. "
+                        + "Pane tail:\n{}",
+                paneId, spawnReadyTimeoutMs, lastStatus, readPaneQuietly(paneId));
         stop(paneId);
         throw new PeerUnreachableException(
                 "worker pane " + paneId + " did not reach injectable state within "
                         + spawnReadyTimeoutMs + "ms");
     }
 
+    /**
+     * fleetd #220: the pane's recent output, clipped, for the readiness-gate timeout log — or a
+     * short note when it cannot be read. Best-effort by construction: this runs on a path that is
+     * already failing, so it must never replace the real error with one of its own.
+     */
+    private String readPaneQuietly(String paneId) {
+        try {
+            String pane = agents.read(paneId, "recent");
+            if (pane == null || pane.isBlank()) {
+                return "<pane read returned nothing>";
+            }
+            return pane.length() <= SPAWN_FAILURE_PANE_CHARS
+                    ? pane
+                    : pane.substring(pane.length() - SPAWN_FAILURE_PANE_CHARS);
+        } catch (RuntimeException e) {
+            return "<pane could not be read: " + e.getMessage() + ">";
+        }
+    }
+
+    /** How much of a failed spawn's pane the timeout log carries. */
+    private static final int SPAWN_FAILURE_PANE_CHARS = 4000;
+
     /**
      * A concrete {@link PeerHandle} wrapping herdr agent coordinates, the profile that spawned it,
      * the session identity the launch resolved (CB-547a): the bridge's logical name and the peer's
@@ -1077,6 +1162,29 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
      * name. It stays a one-off decision because it is a live handle to the operator's ssh-agent, not
      * a value — a member holding it can sign with every key the agent holds, so letting it ride in
      * on the generic {@code allow:} list would hand that out for an unrelated reason.
+     *
+     * <p>fleetd #213: which shell decides the zsh gate, and where the generated directory lives,
+     * both depend on whether {@code memberHerdrSocket:} is configured — see {@link
+     * #memberHerdrSocketConfigured()}'s javadoc for why fleetd's own {@code $SHELL} and {@code
+     * java.io.tmpdir} describe the wrong process once member panes run under a different OS user.
+     * <ul>
+     *   <li>{@code memberHerdrSocket} ABSENT (today's only mode): byte-identical to before this
+     *       fix — fleetd's own {@code $SHELL} decides zsh, and the directory is generated under
+     *       {@code java.io.tmpdir}.</li>
+     *   <li>{@code memberHerdrSocket} PRESENT: the configured {@code memberLoginShell:} decides
+     *       zsh instead — fleetd's own {@code $SHELL} is never consulted, since it names a
+     *       different user's shell, not the member's. Absent/non-zsh falls back exactly like the
+     *       non-zsh case below. When it IS zsh, the directory still cannot go under {@code
+     *       java.io.tmpdir} (mode 0700, unreadable by another uid — the exact gap fleetd #213
+     *       exists to close), so it is generated under {@code worktreeRoot} instead and shared
+     *       read-only with {@code worktreeGroup} — the same group {@link
+     *       dev.ltms.fleet.session.Worktrees#shareWithGroup} already uses, reused rather than
+     *       inventing a second group key. Either one missing means the scrub cannot be guaranteed
+     *       reachable by the member, which is the same "cannot guarantee the scrub runs" case as a
+     *       non-zsh shell, so it gets the identical fallback.</li>
+     * </ul>
+     * In every branch: never refuse to spawn. A degraded credential control must not become an
+     * outage for an opt-in feature.
      */
     private Path applyEnvironmentAllowListPolicy(FleetConfig.Profile cfg, Launch launch) {
         FleetConfig.MemberCredentials creds = memberCredentials == null ? null : memberCredentials.get();
@@ -1084,8 +1192,14 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
             return null;
         }
         Set<String> allowed = derivedAllowedNames(creds, launch);
-        String loginShell = resolveEnv("SHELL");
-        boolean zsh = loginShell != null && (loginShell.endsWith("/zsh") || loginShell.equals("zsh"));
+        boolean memberHerdrSocket = memberHerdrSocketConfigured();
+        // fleetd #213 defect 1: under memberHerdrSocket the member pane runs as a DIFFERENT OS
+        // user, so fleetd's own $SHELL says nothing about what that pane runs — resolveEnv("SHELL")
+        // must not even be called on this path, only the explicit memberLoginShell: config can
+        // answer it. With memberHerdrSocket absent, nothing here changes: fleetd's own $SHELL is
+        // still the input, exactly as before this fix.
+        String loginShell = memberHerdrSocket ? configuredMemberLoginShell() : resolveEnv("SHELL");
+        boolean zsh = isZshShell(loginShell);
         if (!zsh) {
             // A non-zsh login shell ignores ZDOTDIR entirely: NO scrub would run, so pretending
             // otherwise would be worse than saying so. Warn loudly and fall back to the CB-596
@@ -1099,13 +1213,35 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
             logCredentialGap(creds, null);
             return null;
         }
+        Path parentDir;
+        String group = null;
+        if (memberHerdrSocket) {
+            // fleetd #213 defect 2: java.io.tmpdir is fleetd's own per-user temp dir (mode 0700 on
+            // macOS) — a member running as a different uid cannot even traverse it, let alone read
+            // the generated files. worktreeRoot is the only configured location a different-uid
+            // member can be given access to, and only WITH worktreeGroup to grant that access —
+            // absent either, the scrub cannot be guaranteed reachable, so this falls back exactly
+            // like the non-zsh case above rather than generating a directory nothing can read.
+            parentDir = memberScrubParentDir();
+            group = memberGroup();
+            if (parentDir == null || group == null) {
+                warnCannotShareScrubDirectory();
+                overlayBlockedCredentials(launch.env(), creds);
+                logCredentialGap(creds, null);
+                return null;
+            }
+        } else {
+            parentDir = Path.of(System.getProperty("java.io.tmpdir"));
+        }
         // Only reached when the scrub is actually about to run — the count below describes that
         // scrub, so it must not be logged before this gate (see the non-zsh branch above). Same
         // reasoning gates logCredentialGap's wording: passing the derived `allowed` set (non-null)
         // here, and ONLY here, is what tells it the scrub will really blank an unkept name — #192.
         logAllowListCoverage(allowed);
         logCredentialGap(creds, allowed);
-        Path dir = EnvAllowListScrub.generate(Path.of(System.getProperty("java.io.tmpdir")), allowed);
+        Path dir = memberHerdrSocket
+                ? EnvAllowListScrub.generate(parentDir, allowed, group)
+                : EnvAllowListScrub.generate(parentDir, allowed);
         launch.env().put("ZDOTDIR", dir.toAbsolutePath().toString());
         log.info("memberCredentials policy=allow-list: profile={} generated ZDOTDIR {} — derived "
                         + "allow-list holds {} name(s); the pane reports allowed N of M at release",
@@ -1113,6 +1249,63 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
         return dir;
     }
 
+    /** True when {@code shell} is a zsh login shell path or bare name — the ZDOTDIR gate. */
+    private static boolean isZshShell(String shell) {
+        return shell != null && (shell.endsWith("/zsh") || shell.equals("zsh"));
+    }
+
+    /**
+     * fleetd #213: the configured {@code memberLoginShell:}, or {@code null} when unconfigured.
+     * Called ONLY from the {@code memberHerdrSocket}-configured branch of {@link
+     * #applyEnvironmentAllowListPolicy} — fleetd's own {@code $SHELL} is never read on that path.
+     * {@link #config} being {@code null} (an older test call site, or a launcher that never
+     * threaded the full config through) is treated the same as "not configured".
+     */
+    private String configuredMemberLoginShell() {
+        FleetConfig cfg = config == null ? null : config.get();
+        return cfg == null ? null : cfg.memberLoginShell();
+    }
+
+    /**
+     * fleetd #213: {@code worktreeRoot}, as the ZDOTDIR scrub's parent directory under {@code
+     * memberHerdrSocket}, or {@code null} when unconfigured — the same "cannot guarantee the scrub
+     * runs" gap as {@link #memberGroup()} being unset (see {@link
+     * #applyEnvironmentAllowListPolicy}). Deliberately no sibling-of-repo-root default here, unlike
+     * {@code GitWorktrees}' own {@code worktreeRoot} resolution: that default is a convenience for
+     * provisioning a worktree that will exist regardless, whereas an unconfigured value here means
+     * fleetd has no operator-endorsed location to put a credential-bearing directory a different OS
+     * user must reach, so falling back to the overlay is the honest answer, not a guess.
+     *
+     * <p>Package-private (fleetd #219) so {@link OpenCodeLauncher} can reuse the exact same
+     * "different OS user, put it under worktreeRoot instead of java.io.tmpdir" resolution for its
+     * own ephemeral {@code opencode.json} directory, rather than re-reading {@code config} a second
+     * time with a second copy of this null/blank handling.
+     */
+    Path memberScrubParentDir() {
+        FleetConfig cfg = config == null ? null : config.get();
+        if (cfg == null || cfg.worktreeRoot() == null || cfg.worktreeRoot().isBlank()) {
+            return null;
+        }
+        return Path.of(cfg.worktreeRoot());
+    }
+
+    /**
+     * fleetd #213: the configured {@code worktreeGroup:}, or {@code null} when unset/blank. Reuses
+     * the group {@link dev.ltms.fleet.session.Worktrees#shareWithGroup} already establishes for
+     * provisioned worktrees, rather than a second group key — see {@link
+     * #applyEnvironmentAllowListPolicy}.
+     *
+     * <p>Package-private (fleetd #219) — reused by {@link OpenCodeLauncher} alongside {@link
+     * #memberScrubParentDir()}; see that method's javadoc.
+     */
+    String memberGroup() {
+        FleetConfig cfg = config == null ? null : config.get();
+        if (cfg == null || cfg.worktreeGroup() == null || cfg.worktreeGroup().isBlank()) {
+            return null;
+        }
+        return cfg.worktreeGroup();
+    }
+
     /**
      * The full kept-name set for this spawn: the profile-derived names, unioned with {@code
      * memberCredentials.allow:} (CB-633 follow-up — previously ignored by this whole policy), the
@@ -1170,6 +1363,30 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
         }
     }
 
+    /** Guards {@link #warnCannotShareScrubDirectory} to one WARN per launcher instance. */
+    private final AtomicBoolean cannotShareScrubDirWarned = new AtomicBoolean();
+
+    /**
+     * fleetd #213: {@code memberHerdrSocket} is configured and the member login shell IS zsh, but
+     * {@code worktreeRoot} and/or {@code worktreeGroup} is missing, so the generated ZDOTDIR cannot
+     * be placed anywhere the member's OS user can reach — {@code java.io.tmpdir} is fleetd's own
+     * 0700 temp dir, unreadable by another uid, which is the exact gap this ticket exists to close.
+     * Say so once per launcher instance, instead of either generating a directory nothing can read
+     * (protection theatre) or refusing to spawn (turning a degraded credential control into an
+     * outage for an opt-in feature).
+     */
+    private void warnCannotShareScrubDirectory() {
+        if (cannotShareScrubDirWarned.compareAndSet(false, true)) {
+            log.warn("memberCredentials policy=allow-list: memberHerdrSocket is configured and the "
+                    + "member login shell is zsh, but worktreeRoot and/or worktreeGroup is not "
+                    + "configured — the generated ZDOTDIR cannot be placed where the member's OS "
+                    + "user can read it (java.io.tmpdir is fleetd's own, unreadable by another uid), "
+                    + "so the scrub cannot be guaranteed to run. Falling back to the CB-596 sentinel "
+                    + "overlay. Configure both worktreeRoot and worktreeGroup to enable the "
+                    + "allow-list scrub under memberHerdrSocket.");
+        }
+    }
+
     /**
      * CB-633 teardown half: read the pane's scrub report (the denominator report the generated
      * scrub wrote) and delete the directory. Called from {@link #stop}, which is the one funnel
@@ -1254,8 +1471,12 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
      * through (every production {@code HerdrPeerLauncher} does; a handful of older tests do not) —
      * treated the same as "not configured", which is the correct, permissive default: it is exactly
      * today's single-daemon behaviour.
+     *
+     * <p>Package-private (fleetd #219) — {@link OpenCodeLauncher} reuses this same gate to decide
+     * where its own ephemeral {@code opencode.json} directory (site 1) and its opencode session
+     * discovery (site 2) may run, rather than re-deriving "is this a multi-uid fleet" a second way.
      */
-    private boolean memberHerdrSocketConfigured() {
+    boolean memberHerdrSocketConfigured() {
         if (config == null) {
             return false;
         }

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

@@ -20,6 +20,8 @@ import java.util.EnumSet;
 import java.util.List;
 import java.util.Map;
 import java.util.Set;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.function.BooleanSupplier;
 import java.util.function.Function;
 import java.util.function.LongSupplier;
 import java.util.function.Supplier;
@@ -215,7 +217,43 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
         return Path.of(System.getProperty("java.io.tmpdir"));
     }
 
-    /** The default opencode storage root: {@code ~/.local/share/opencode} (the XDG data dir). */
+    /**
+     * The default opencode storage root: {@code ~/.local/share/opencode} (the XDG data dir) —
+     * always FLEETD's OWN {@code user.home}, whichever OS user runs the daemon.
+     *
+     * <p><b>fleetd #219 site 2 — a decision, not a patch.</b> Under {@code memberHerdrSocket:} the
+     * member pane runs as a <em>different</em> OS user, and opencode writes {@code opencode.db}
+     * under <em>that</em> user's {@code $HOME}, not fleetd's. Scanning fleetd's own {@code
+     * user.home} is therefore looking in the wrong place — a wrong-LOCATION failure, not a
+     * wrong-PERMISSION one like site 1, and it fails quietly: {@link
+     * SessionAwareHandle#agentSessionId()} would keep returning {@code null} forever, which reads
+     * as "opencode does not support resume" rather than "fleetd looked in the wrong home." fleetd
+     * #209 is the reason that silence is unacceptable.
+     *
+     * <p>Three ways to close the gap were weighed:
+     * <ol>
+     *   <li><b>Make the member's home configurable.</b> Correct in principle, but this ticket's
+     *       scope is the two existing call sites, not a new config key — {@code memberHerdrSocket}
+     *       already carries the second herdr's socket path, not its user's home, and inventing a
+     *       parallel key here without also wiring it through discovery's actual callers is a
+     *       half-shipped feature (the exact shape CB-596/CB-611 warn against).</li>
+     *   <li><b>Derive it</b> (e.g. from {@code worktreeRoot}'s owner, or {@code getent passwd}).
+     *       Rejected: nothing in this codebase resolves a Unix username to a home directory today,
+     *       and guessing wrong would silently point discovery at a THIRD wrong location — worse
+     *       than the current gap, because it would look like it should work.</li>
+     *   <li><b>Declare discovery unavailable</b> under {@code memberHerdrSocket}, and say so once,
+     *       loudly, instead of scanning a directory that structurally cannot hold the answer.</li>
+     * </ol>
+     *
+     * <p>Option 3 is taken — the one this ticket says to default to when unsure. {@link
+     * OpenCodeLauncher#spawn} routes {@link SessionAwareHandle#agentSessionId()} through {@link
+     * HerdrPeerLauncher#memberHerdrSocketConfigured()} before ever calling {@link
+     * OpenCodeSessionDiscovery#sessionIdForDirectory}, so under {@code memberHerdrSocket} the
+     * database at this root is never even opened, and one WARN per launcher instance names the gap
+     * instead of the {@code null} return reading as "unsupported." Capability advertising is
+     * unaffected: {@link #capabilities()} always includes {@code SESSION_RESUME}, since {@code
+     * memberHerdrSocket} absent (today's only live mode) is unchanged by this decision.
+     */
     private static Path defaultDiscoveryRoot() {
         return Path.of(System.getProperty("user.home"), ".local", "share", "opencode");
     }
@@ -331,7 +369,7 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
      */
     private Path writeConfig(FleetConfig.Profile cfg, String charterText, String cwd) {
         try {
-            Path dir = Files.createTempDirectory(configRoot, "fleetd-opencode-");
+            Path dir = Files.createTempDirectory(configParentDir(), "fleetd-opencode-");
             dir.toFile().deleteOnExit();
 
             ObjectNode root = JSON.createObjectNode();
@@ -402,6 +440,14 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
             // carries operator-supplied values (URL, model id, api key), so escaping must be real.
             Files.writeString(cfgFile, JSON.writerWithDefaultPrettyPrinter().writeValueAsString(root));
             cfgFile.toFile().deleteOnExit();
+            if (memberHerdrSocketConfigured()) {
+                // fleetd #219: the same "different OS user" gap fleetd #213 closed for the ZDOTDIR
+                // scrub — share read-only with worktreeGroup rather than leaving the directory under
+                // fleetd's own 0700 java.io.tmpdir, where the member's OS user could not even
+                // traverse it. memberGroup() cannot be null here: configParentDir() above already
+                // refused this spawn if either worktreeRoot or worktreeGroup was missing.
+                EnvAllowListScrub.shareWithGroup(dir, memberGroup());
+            }
             return cfgFile;
         } catch (IOException e) {
             throw new UncheckedIOException(
@@ -409,6 +455,57 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
         }
     }
 
+    /**
+     * fleetd #219 site 1: where {@link #writeConfig} creates its per-spawn directory.
+     *
+     * <ul>
+     *   <li>{@code memberHerdrSocket} ABSENT (today's only mode): byte-identical to before this
+     *       fix — always {@link #configRoot} (defaults to {@code java.io.tmpdir}, fleetd's own
+     *       process).</li>
+     *   <li>{@code memberHerdrSocket} PRESENT: {@code java.io.tmpdir} is fleetd's own per-user temp
+     *       dir (mode {@code 0700} on macOS) — the member pane runs as a DIFFERENT OS user under
+     *       this config key and cannot even traverse it, so the directory holding {@code
+     *       opencode.json} (which tells the member where the bridge MCP is) and the member charter
+     *       would be unreadable to the very process it is written for. The directory instead goes
+     *       under {@code worktreeRoot}, shared read-only with {@code worktreeGroup} via {@link
+     *       EnvAllowListScrub#shareWithGroup} — the SAME mechanism fleetd #213 built for the ZDOTDIR
+     *       scrub, reused here rather than duplicated (see {@link
+     *       HerdrPeerLauncher#memberScrubParentDir()}).</li>
+     * </ul>
+     *
+     * <p><b>Unlike the ZDOTDIR scrub, a missing {@code worktreeRoot}/{@code worktreeGroup} here
+     * REFUSES the spawn instead of degrading.</b> The ZDOTDIR scrub is a credential CONTROL: a
+     * degraded control (CB-596's sentinel overlay) is still worth having. This config file is not a
+     * control — it is the ONLY way the member learns where the bridge MCP lives. Writing it
+     * somewhere the member cannot read would not degrade anything; it would spawn a member that
+     * occupies a pane and never becomes deliverable, since {@code fleet_send} waits ~60s on the
+     * readiness gate and then fails with nothing pointing at a temp directory as the cause. Refusing
+     * up front, with a message that names the missing config key, is the honest failure — an
+     * undeliverable member is not a working spawn either way, so nothing is lost by refusing loudly
+     * instead of failing silently later.
+     *
+     * @throws IllegalStateException when {@code memberHerdrSocket} is configured but {@code
+     *                                worktreeRoot} and/or {@code worktreeGroup} is not
+     */
+    private Path configParentDir() {
+        if (!memberHerdrSocketConfigured()) {
+            return configRoot;
+        }
+        Path root = memberScrubParentDir();
+        String group = memberGroup();
+        if (root == null || group == null) {
+            throw new IllegalStateException("memberHerdrSocket is configured, so opencode's config "
+                    + "directory (opencode.json + member charter) must be placed where the member's "
+                    + "OS user can read it — worktreeRoot, shared via worktreeGroup — but "
+                    + (root == null ? "worktreeRoot" : "worktreeGroup") + " is not configured. "
+                    + "Refusing to spawn rather than write a config the member cannot read: that "
+                    + "member would occupy a pane and never become deliverable, with nothing "
+                    + "pointing at the real cause. Configure both worktreeRoot and worktreeGroup to "
+                    + "enable opencode member spawns under memberHerdrSocket.");
+        }
+        return root;
+    }
+
     /**
      * Declare a custom OpenAI-compatible provider so the worker talks to a pinned endpoint (a local
      * vLLM, say) instead of opencode's default gateway (CB-508).
@@ -517,11 +614,16 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
         return afterScheme.contains("/") ? trimmed : trimmed + "/v1";
     }
 
+    /** One WARN per launcher instance for the fleetd #219 site-2 discovery-unavailable gap. */
+    private final AtomicBoolean discoveryUnavailableWarned =
+            new AtomicBoolean();
+
     /** Add lazy on-disk session discovery to the base handle. */
     @Override
     public PeerHandle spawn(SpawnRequest req) {
         PeerHandle inner = super.spawn(req);
-        return new SessionAwareHandle(inner, discovery, effectiveCwd(req));
+        return new SessionAwareHandle(inner, discovery, effectiveCwd(req),
+                this::memberHerdrSocketConfigured, discoveryUnavailableWarned);
     }
 
     /**
@@ -536,11 +638,17 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
         private final PeerHandle delegate;
         private final OpenCodeSessionDiscovery discovery;
         private final String cwd;
+        private final BooleanSupplier discoveryUnavailable;
+        private final AtomicBoolean discoveryUnavailableWarned;
 
-        SessionAwareHandle(PeerHandle delegate, OpenCodeSessionDiscovery discovery, String cwd) {
+        SessionAwareHandle(PeerHandle delegate, OpenCodeSessionDiscovery discovery, String cwd,
+                            BooleanSupplier discoveryUnavailable,
+                            AtomicBoolean discoveryUnavailableWarned) {
             this.delegate = delegate;
             this.discovery = discovery;
             this.cwd = cwd;
+            this.discoveryUnavailable = discoveryUnavailable;
+            this.discoveryUnavailableWarned = discoveryUnavailableWarned;
         }
 
         @Override
@@ -565,6 +673,22 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
 
         @Override
         public String agentSessionId() {
+            // fleetd #219 site 2: under memberHerdrSocket the member pane runs as a different OS
+            // user, so opencode.db lives under THAT user's $HOME, not the one discoveryRoot was
+            // built from (see OpenCodeLauncher#defaultDiscoveryRoot's javadoc for the full
+            // reasoning). Scanning fleetd's own $HOME under that config would only ever find "no
+            // row" and read as "resume unsupported" — declare it unavailable instead, once, loudly.
+            if (discoveryUnavailable.getAsBoolean()) {
+                if (discoveryUnavailableWarned.compareAndSet(false, true)) {
+                    log.warn("opencode session discovery unavailable: memberHerdrSocket is "
+                            + "configured, so opencode's on-disk session database lives under the "
+                            + "MEMBER's own $HOME, not fleetd's ({}) — agentSessionId will stay null "
+                            + "for every opencode member under this config, and SESSION_RESUME "
+                            + "cannot be honored (fleetd #209/#219).",
+                            System.getProperty("user.home"));
+                }
+                return null;
+            }
             // Lazy + retried, never a spawn-time blocker: opencode writes the session record only
             // when the session is first persisted, so null here is the correct interim answer and
             // the caller re-calls later (each call re-scans, picking up a record that has since

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

@@ -682,6 +682,12 @@ class CompletionResolverTest {
                 "classified BACKEND_ERROR from the raw scrape even though the trimmed block was empty");
         assertTrue(waiter.getNow(null).text().contains("API Error: 400 invalid request body"),
                 "the failure carries the matched line: " + waiter.getNow(null).text());
+        assertTrue(waiter.getNow(null).text().contains("--- pane tail ---"),
+                "fleetd#164: the failure must carry the pane, not only the matched line — with an "
+                        + "empty trimmed block the raw scrape is the only copy of what the member said: "
+                        + waiter.getNow(null).text());
+        assertTrue(waiter.getNow(null).text().contains("╭"),
+                "the carried pane is the raw scrape, chrome included: " + waiter.getNow(null).text());
     }
 
     @Test

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

@@ -61,8 +61,75 @@ class ClaudeCodeLauncherTest {
         assertTrue(args.stream().noneMatch(a -> a.contains("\"bridge\"")),
                 "the mount is named fleet since CB-632 — a member addresses its tools as "
                         + "mcp__fleet__*, and CLAUDE.md's role-detection ladder names that prefix");
-        assertTrue(args.contains("--append-system-prompt"));
-        assertTrue(args.stream().anyMatch(a -> a.contains("fleet_reply")), "reply charter present");
+        // fleetd #220: the charter travels as a FILE, never inline — charter prose on the command
+        // line overruns the byte cap of the pane line herdr types it into.
+        assertTrue(args.contains("--append-system-prompt-file"));
+        assertFalse(args.contains("--append-system-prompt"),
+                "the inline flag would put ~800 bytes of prose on the pane command line");
+        String charterFile = args.get(args.indexOf("--append-system-prompt-file") + 1);
+        assertTrue(readFile(charterFile).contains("fleet_reply"), "reply charter present in the file");
+    }
+
+    /** Read a charter file the launcher wrote, failing the test rather than the build on an IO error. */
+    private static String readFile(String path) {
+        try {
+            return java.nio.file.Files.readString(java.nio.file.Path.of(path));
+        } catch (java.io.IOException e) {
+            throw new AssertionError("charter file " + path + " is not readable", e);
+        }
+    }
+
+    /**
+     * fleetd #220 regression. herdr TYPES the launch command into the pane, and the pty line buffer
+     * holds 1024 bytes — past that the tail is dropped with no error anywhere, so the backend exits
+     * on a mangled argument and the spawn dies as an unexplained readiness timeout. That is what
+     * happened when #214 added --session-id to a command already 978 bytes long: --autocompact
+     * 250000 arrived as --autocompact 25. This asserts the whole assembled command still fits, with
+     * the flags a real spawn carries (MCP mount, charter, model, autocompact, session id).
+     */
+    @Test
+    void theAssembledLaunchCommandFitsThePaneLineLimit() {
+        FakeHerdr herdr = new FakeHerdr();
+        // Shaped like the live sonnet profile, because the bug is a SUM: the charter alone fits,
+        // and so does every flag alone. Only model + autocompact + session id on top of the charter
+        // crossed the cap, which is why nothing caught it until a member failed to spawn.
+        FleetConfig.Profile cfg = new FleetConfig.Profile(
+                "sonnet", null, "claude-sonnet-5", null, null,
+                List.of("claude"), "tab", "fleetd-workers", "w #{n}", "http://127.0.0.1:8765/mcp",
+                null, null, null, null, null, null, null, null, true, null, null, null, null, null,
+                250000);
+        new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
+                new SubscriptionGuard(Set.of()), Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null)
+                .spawn();
+
+        List<String> args = spawnedArgs(herdr);
+        int bytes = "claude".length();
+        for (String arg : args) {
+            bytes += arg.getBytes(java.nio.charset.StandardCharsets.UTF_8).length + 3;
+        }
+        assertTrue(bytes <= HerdrPeerLauncher.PANE_COMMAND_BYTE_LIMIT,
+                "the launch command must fit the pane line: " + bytes + " bytes vs limit "
+                        + HerdrPeerLauncher.PANE_COMMAND_BYTE_LIMIT + " — args " + args);
+    }
+
+    /**
+     * fleetd #220: the guard refuses a command that cannot fit, instead of letting the pty drop the
+     * tail. The refusal must name the size and the argument to blame — a spawn that fails with
+     * "did not reach injectable state" tells the operator nothing, which is the whole reason this
+     * bug took a live pane scrape to find.
+     */
+    @Test
+    void anOverlongLaunchCommandIsRefusedWithTheSizeAndTheCulprit() {
+        FakeHerdr herdr = new FakeHerdr();
+        String huge = "x".repeat(1500);
+        ClaudeCodeLauncher launcher = service(herdr, List.of("claude", huge), null);
+
+        PeerUnreachableException refused = assertThrows(PeerUnreachableException.class, launcher::spawn);
+
+        assertTrue(refused.getMessage().contains("1024"), "names the limit: " + refused.getMessage());
+        assertTrue(refused.getMessage().contains("1500"), "names the culprit's size: " + refused.getMessage());
+        assertFalse(herdr.called("agent.start"),
+                "nothing may be started — a truncated command is worse than no spawn");
     }
 
     // CB-634: a profile with ideMcpUrl set mounts the IDE Index MCP as a second server and pins
@@ -262,10 +329,16 @@ class ClaudeCodeLauncherTest {
 
         Map<?, ?> start = (Map<?, ?>) herdr.lastCall("agent.start").params();
         assertEquals("claude", start.get("kind"), "herdr launches the canonical executable by kind");
-        // CB-533: the shared fixture pins model "coder", so the model flag is the whole args list.
         // What this test guards is that argv[0] is NOT repeated — herdr supplies it from `kind`.
-        assertEquals(List.of("--model", "coder"), start.get("args"),
-                "the configured executable is not repeated in args");
+        // fleetd #214: a plain spawn now also carries fleetd's own minted session id.
+        List<String> args = spawnedArgs(herdr);
+        assertFalse(args.contains("claude"), "the configured executable is not repeated in args");
+        int flag = args.indexOf("--session-id");
+        assertTrue(flag >= 0, "a plain spawn mints a session id: " + args);
+        assertDoesNotThrow(() -> UUID.fromString(args.get(flag + 1)), "the minted id is a valid UUID");
+        assertEquals(List.of("--model", "coder"),
+                List.of(args.get(args.size() - 2), args.get(args.size() - 1)),
+                "the CB-533 model flag still trails the launch flags: " + args);
     }
 
     @Test
@@ -278,7 +351,15 @@ class ClaudeCodeLauncherTest {
         assertFalse(args.contains("--append-system-prompt"), "no reply charter without mcpUrl");
         // CB-533: the model flag is independent of the MCP mount — pinning the model is not part of
         // "mount the bridge", so an unmounted worker still runs the model its profile names.
-        assertEquals(List.of("--verbose", "--model", "coder"), args,
+        // fleetd #214: a plain spawn now carries fleetd's own minted session id; drop the two
+        // mint elements when checking the rest of the argv.
+        int flag = args.indexOf("--session-id");
+        assertTrue(flag >= 0, "a plain spawn mints a session id even without a bridge mount: " + args);
+        assertDoesNotThrow(() -> UUID.fromString(args.get(flag + 1)), "the minted id is a valid UUID");
+        List<String> rest = new java.util.ArrayList<>(args);
+        rest.remove(flag + 1);
+        rest.remove(flag);
+        assertEquals(List.of("--verbose", "--model", "coder"), rest,
                 "the operator's own args are preserved, in order, ahead of the model flag");
     }
 
@@ -649,7 +730,7 @@ class ClaudeCodeLauncherTest {
         assertEquals("/work/proj", cwd, "effectiveCwd via SpawnRequest must match the three-arg resolution");
     }
 
-    // --- CB-547a: durable session identity (mint / resume / no-identity legacy) -----------------
+    // --- CB-547a / fleetd #214: durable session identity (always mint / resume) -----------------
 
     @Test
     void freshSpawnMintsASessionIdAndPassesTheName() {
@@ -684,18 +765,28 @@ class ClaudeCodeLauncherTest {
     }
 
     @Test
-    void noIdentitySpawnKeepsTheLegacyArgvAndCarriesNoSessionHandle() {
+    void plainSpawnMintsASessionIdSoEveryMemberIsResumable() {
+        // fleetd #214: a plain spawn passes no sessionName and no resumeSessionId, yet the member
+        // must still be resumable — the id is minted unconditionally, and it is the ONLY resume
+        // handle a claude-code member has (unlike opencode, nothing resolves it after the launch).
+        // The binary requires a valid UUID (checked against claude 2.1.252: a non-UUID is refused
+        // at argument parsing with "Invalid session ID. Must be a valid UUID.").
         FakeHerdr herdr = new FakeHerdr();
         ClaudeCodeLauncher svc = service(herdr, List.of("ccs", "ltms-local"), null);
 
         PeerHandle handle = svc.spawn(new SpawnRequest("ltms-local", null, null));
 
         List<String> args = spawnedArgs(herdr);
-        assertFalse(args.contains("--session-id"), "no identity → no --session-id");
-        assertFalse(args.contains("-n"), "no identity → no -n");
-        assertFalse(args.contains("-r"), "no identity → no -r");
-        assertNull(handle.agentSessionId(), "no identity → no resume handle");
-        assertNull(handle.sessionName(), "no identity → no logical name");
+        int flag = args.indexOf("--session-id");
+        assertTrue(flag >= 0 && flag + 1 < args.size(),
+                "--session-id is minted even when no identity is requested: " + args);
+        String minted = args.get(flag + 1);
+        assertDoesNotThrow(() -> UUID.fromString(minted), "--session-id is a valid UUID: " + minted);
+        assertEquals(minted, handle.agentSessionId(),
+                "the resume handle is the minted id, so every member is resumable from fleet_list");
+        assertFalse(args.contains("-n"), "no sessionName was requested → no -n");
+        assertFalse(args.contains("-r"), "no resume was requested → no -r");
+        assertNull(handle.sessionName(), "no sessionName was requested → no logical name");
     }
 
     @Test

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

@@ -12,20 +12,26 @@ import dev.ltms.fleet.peer.Capability;
 import dev.ltms.fleet.peer.MemberRole;
 import dev.ltms.fleet.peer.SpawnRequest;
 import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.io.TempDir;
 import org.slf4j.LoggerFactory;
 
+import java.io.IOException;
 import java.nio.file.Files;
 import java.nio.file.Path;
+import java.nio.file.attribute.PosixFileAttributeView;
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
 import java.util.Set;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.function.Function;
 import java.util.function.Supplier;
 
 import static org.junit.jupiter.api.Assertions.assertEquals;
 import static org.junit.jupiter.api.Assertions.assertFalse;
 import static org.junit.jupiter.api.Assertions.assertNotNull;
 import static org.junit.jupiter.api.Assertions.assertTrue;
+import static org.junit.jupiter.api.Assumptions.assumeTrue;
 
 /**
  * CB-633: proves the allow-list scrub is actually WIRED INTO the spawn path — not merely that its
@@ -119,6 +125,18 @@ class HerdrPeerLauncherAllowListWiringTest {
                 FleetConfig.MemberCredentials.POLICY_ALLOW_LIST, allow, List.of(), null);
     }
 
+    /**
+     * Same as {@link #allowList()} but with an operator-configured {@code known:} list, so the
+     * fallback overlay (the non-zsh / no-memberLoginShell path) has something visible to shadow —
+     * {@link FleetConfig.MemberCredentials#blockedSet()} is {@code known - allow}, so an empty
+     * {@code known} (what {@link #allowList()} uses) blocks nothing and a fallback test would have
+     * no sentinel entry to assert on.
+     */
+    private static Supplier<FleetConfig.MemberCredentials> allowListWithKnown(List<String> known) {
+        return () -> new FleetConfig.MemberCredentials(
+                FleetConfig.MemberCredentials.POLICY_ALLOW_LIST, List.of(), known, null);
+    }
+
     /**
      * CB-633 follow-up: a name that lives ONLY in {@code memberCredentials.allow:} — no profile
      * mentions it — must survive the scrub the real spawn path generates. Calling {@code
@@ -371,6 +389,137 @@ class HerdrPeerLauncherAllowListWiringTest {
                 "the log must never contain an env var VALUE, only its NAME — got: " + messages);
     }
 
+    /**
+     * fleetd #213 defect 1, acceptance criterion 1: {@code memberHerdrSocket} configured and {@code
+     * memberLoginShell} configured as non-zsh must fall back to the sentinel overlay exactly like a
+     * non-zsh {@code $SHELL} does today — and the "generated ZDOTDIR" INFO must not appear, since no
+     * scrub actually runs. The WiringLauncher's own {@code env("SHELL")} is deliberately set to
+     * {@code /bin/zsh} — the OPPOSITE of what {@code memberLoginShell} says — so a launcher that
+     * (incorrectly) fell back to fleetd's own {@code $SHELL} here would wrongly pass the gate and
+     * fail this test.
+     */
+    @Test
+    void memberHerdrSocketWithNonZshMemberLoginShellFallsBackToTheOverlay() {
+        FakeHerdr herdr = new FakeHerdr();
+        WiringLauncher launcher = new WiringLauncher(herdr, allowListWithKnown(List.of("SOME_TOKEN")),
+                "/bin/zsh", null,
+                () -> configWithMemberHerdrSocket("/tmp/other-user-herdr.sock", "/bin/bash"));
+
+        List<String> messages = spawnAndCaptureLogs(launcher);
+
+        assertEquals("blocked-by-fleetd-cb596-see-gitea-issue-82", launcher.env.get("SOME_TOKEN"),
+                "a non-zsh memberLoginShell must fall back to the CB-596 sentinel overlay, exactly "
+                        + "like a non-zsh $SHELL does when memberHerdrSocket is absent");
+        assertFalse(launcher.env.containsKey("ZDOTDIR"),
+                "no scrub directory may be generated when the configured member login shell is not zsh");
+        assertFalse(messages.stream().anyMatch(m -> m.contains("generated ZDOTDIR")),
+                "the 'generated ZDOTDIR' INFO must not appear when the scrub never runs — got: " + messages);
+    }
+
+    /**
+     * fleetd #213 defect 1, acceptance criterion 2: {@code memberHerdrSocket} configured and NO
+     * {@code memberLoginShell} configured must fall back exactly like criterion 1 above — AND
+     * fleetd's own {@code $SHELL} must never even be consulted (not merely "not decisive"). The
+     * fixture's {@code env} function reports {@code /bin/zsh} for {@code SHELL} — a value that would
+     * WRONGLY pass the zsh gate if the fix regressed to reading it — while flagging whether it was
+     * ever asked for at all, so this test fails loudly on either kind of regression.
+     */
+    @Test
+    void memberHerdrSocketWithNoMemberLoginShellFallsBackAndNeverConsultsFleetdsOwnShell() {
+        FakeHerdr herdr = new FakeHerdr();
+        AtomicBoolean shellQueried = new AtomicBoolean(false);
+        Function<String, String> env = name -> {
+            if ("SHELL".equals(name)) {
+                shellQueried.set(true);
+                return "/bin/zsh"; // would wrongly pass the zsh gate if this ever leaked through
+            }
+            return null;
+        };
+        WiringLauncher launcher = new WiringLauncher(herdr, allowListWithKnown(List.of("SOME_TOKEN")), env,
+                () -> configWithMemberHerdrSocket("/tmp/other-user-herdr.sock", null));
+
+        List<String> messages = spawnAndCaptureLogs(launcher);
+
+        assertFalse(shellQueried.get(), "fleetd's own $SHELL must never be consulted once "
+                + "memberHerdrSocket is configured — only memberLoginShell: may decide the gate");
+        assertEquals("blocked-by-fleetd-cb596-see-gitea-issue-82", launcher.env.get("SOME_TOKEN"),
+                "no memberLoginShell configured must fall back to the sentinel overlay, same as a "
+                        + "configured non-zsh shell");
+        assertFalse(messages.stream().anyMatch(m -> m.contains("generated ZDOTDIR")),
+                "no scrub may run without a configured memberLoginShell — got: " + messages);
+    }
+
+    /**
+     * fleetd #213 defect 2, acceptance criterion 3: with {@code memberHerdrSocket} configured, a
+     * zsh {@code memberLoginShell}, and {@code worktreeRoot}/{@code worktreeGroup} both configured,
+     * the generated scrub directory must live under {@code worktreeRoot} — NEVER under {@code
+     * java.io.tmpdir}, which is fleetd's own 0700 temp dir and unreadable by the member's different
+     * OS user. {@code worktreeGroup} is set to the CURRENT process's own primary group so {@code
+     * EnvAllowListScrub}'s group-sharing step resolves on whatever host runs this test, rather than
+     * hardcoding a group name that may not exist here.
+     */
+    @Test
+    void memberHerdrSocketWithZshMemberLoginShellPutsTheScrubOutsideJavaIoTmpdir(@TempDir Path worktreeRoot)
+            throws IOException {
+        String group = currentUserGroup();
+        FakeHerdr herdr = new FakeHerdr();
+        WiringLauncher launcher = new WiringLauncher(herdr, allowList(), "/bin/bash-should-be-ignored", null,
+                () -> configWithMemberHerdrSocketRootAndGroup("/tmp/other-user-herdr.sock", "/bin/zsh",
+                        worktreeRoot.toString(), group));
+
+        launcher.spawn(new SpawnRequest("test", null, null, null, null, MemberRole.DEV));
+
+        String zdotdir = launcher.env.get("ZDOTDIR");
+        assertNotNull(zdotdir, "a zsh memberLoginShell with worktreeRoot+worktreeGroup configured "
+                + "must still generate a ZDOTDIR");
+        Path dir = Path.of(zdotdir);
+        // The immediate PARENT is asserted (not merely startsWith(java.io.tmpdir)), because a
+        // JUnit @TempDir is itself carved out of the JVM's java.io.tmpdir — startsWith alone would
+        // pass by coincidence of the test fixture, not because the launcher used worktreeRoot.
+        assertEquals(worktreeRoot.toAbsolutePath().normalize(), dir.getParent(),
+                "the generated scrub directory's parent must be the configured worktreeRoot, not "
+                        + "System.getProperty(\"java.io.tmpdir\") — got parent " + dir.getParent());
+    }
+
+    /**
+     * fleetd #213, acceptance criterion 4: with {@code memberHerdrSocket} absent (today's only
+     * mode), behaviour must be byte-identical to before this fix — fleetd's own {@code $SHELL}
+     * still decides the gate (proven here, not merely assumed, by flagging the lookup), and the
+     * scrub still lands under {@code java.io.tmpdir}.
+     */
+    @Test
+    void memberHerdrSocketAbsentStillConsultsFleetdsOwnShellAndBehavesAsBefore() {
+        FakeHerdr herdr = new FakeHerdr();
+        AtomicBoolean shellQueried = new AtomicBoolean(false);
+        Function<String, String> env = name -> {
+            if ("SHELL".equals(name)) {
+                shellQueried.set(true);
+                return "/bin/zsh";
+            }
+            return null;
+        };
+        WiringLauncher launcher = new WiringLauncher(herdr, allowList(), env, null); // memberHerdrSocket absent
+
+        launcher.spawn(new SpawnRequest("test", null, null, null, null, MemberRole.DEV));
+
+        assertTrue(shellQueried.get(), "with memberHerdrSocket absent, fleetd's own $SHELL must "
+                + "still decide the zsh gate, unchanged from before this fix");
+        String zdotdir = launcher.env.get("ZDOTDIR");
+        assertNotNull(zdotdir, "SHELL=/bin/zsh with memberHerdrSocket absent must still generate a "
+                + "ZDOTDIR, as before this fix");
+        Path dir = Path.of(zdotdir);
+        assertTrue(dir.startsWith(Path.of(System.getProperty("java.io.tmpdir"))),
+                "with memberHerdrSocket absent the scrub directory must still be generated under "
+                        + "java.io.tmpdir, unchanged from before this fix: " + dir);
+    }
+
+    /** The current process's own primary group — resolvable on whatever host runs this test. */
+    private static String currentUserGroup() throws IOException {
+        PosixFileAttributeView view = Files.getFileAttributeView(Path.of("."), PosixFileAttributeView.class);
+        assumeTrue(view != null, "this host's filesystem does not support POSIX group ownership");
+        return view.readAttributes().group().getName();
+    }
+
     /** Spawn once through the real launcher path, capturing every INFO+ line this class logs. */
     private static List<String> spawnAndCaptureLogs(HerdrPeerLauncher launcher) {
         Logger logger = (Logger) LoggerFactory.getLogger(HerdrPeerLauncher.class);
@@ -446,6 +595,19 @@ class HerdrPeerLauncherAllowListWiringTest {
                     0, () -> 0L, () -> { }, null, creds, hostEnvNames, config);
         }
 
+        /**
+         * Full control over the {@code env} lookup, bypassing the {@code shell}/{@code
+         * extraEnvValues} convenience above entirely — fleetd #213's "fleetd's own $SHELL must
+         * never be consulted" tests need to OBSERVE whether {@code SHELL} was ever looked up, which
+         * a plain value substitution cannot do.
+         */
+        WiringLauncher(FakeHerdr herdr, Supplier<FleetConfig.MemberCredentials> creds,
+                       Function<String, String> env, Supplier<FleetConfig> config) {
+            super("test", new AgentControl(herdr), new WorkspaceControl(herdr),
+                    Map.of("test", profile()), "test",
+                    env, 0, () -> 0L, () -> { }, null, creds, null, config);
+        }
+
         @Override
         protected Launch buildLaunch(FleetConfig.Profile cfg, LaunchSpec spec) {
             Map<String, String> launchEnv = baseEnv(cfg);
@@ -477,6 +639,72 @@ class HerdrPeerLauncherAllowListWiringTest {
                 null, null, null, null, null, null, null, null, null, null, null).withDefaults();
     }
 
+    /**
+     * fleetd #213: as {@link #configWithMemberHerdrSocket(String)}, plus the {@code
+     * memberLoginShell:} the member's OS user actually runs — the config key {@link
+     * HerdrPeerLauncher#applyEnvironmentAllowListPolicy} must consult instead of fleetd's own
+     * {@code $SHELL} once {@code memberHerdrSocket} is configured.
+     */
+    private static FleetConfig configWithMemberHerdrSocket(String memberHerdrSocket, String memberLoginShell) {
+        return new FleetConfig(
+                null,               // bind
+                null,               // herdrSocket
+                memberHerdrSocket,  // memberHerdrSocket
+                Map.of(),           // profiles
+                null,               // guard
+                null,               // worktreeRoot
+                null,               // lifecycle
+                null,               // spawnReadyTimeoutMs
+                null,               // spawnReadyPollMs
+                null,               // broker
+                null,               // primary
+                null,               // fleet
+                null,               // leadHeartbeat
+                null,               // health
+                null,               // placement
+                null,               // auth
+                null,               // configReload
+                null,               // quarantineCooldownSeconds
+                null,               // memberCredentials
+                null,               // coordinator
+                null,               // worktreeGroup
+                memberLoginShell    // memberLoginShell
+        ).withDefaults();
+    }
+
+    /**
+     * fleetd #213: as above, plus {@code worktreeRoot:}/{@code worktreeGroup:} — both required for
+     * the ZDOTDIR scrub to run at all once {@code memberHerdrSocket} is configured; either missing
+     * falls back to the sentinel overlay, same as a non-zsh {@code memberLoginShell}.
+     */
+    private static FleetConfig configWithMemberHerdrSocketRootAndGroup(String memberHerdrSocket,
+            String memberLoginShell, String worktreeRoot, String worktreeGroup) {
+        return new FleetConfig(
+                null,               // bind
+                null,               // herdrSocket
+                memberHerdrSocket,  // memberHerdrSocket
+                Map.of(),           // profiles
+                null,               // guard
+                worktreeRoot,       // worktreeRoot
+                null,               // lifecycle
+                null,               // spawnReadyTimeoutMs
+                null,               // spawnReadyPollMs
+                null,               // broker
+                null,               // primary
+                null,               // fleet
+                null,               // leadHeartbeat
+                null,               // health
+                null,               // placement
+                null,               // auth
+                null,               // configReload
+                null,               // quarantineCooldownSeconds
+                null,               // memberCredentials
+                null,               // coordinator
+                worktreeGroup,      // worktreeGroup
+                memberLoginShell    // memberLoginShell
+        ).withDefaults();
+    }
+
     /** The generated directory is a temp directory; make sure the test does not leave a pile. */
     @Test
     void theGeneratedDirectoryIsRemovedWhenThePaneIsStopped() {

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

@@ -1,5 +1,9 @@
 package dev.ltms.fleet.member;
 
+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.fasterxml.jackson.databind.JsonNode;
 import com.fasterxml.jackson.databind.ObjectMapper;
 import dev.ltms.fleet.config.FleetConfig;
@@ -14,9 +18,13 @@ import dev.ltms.fleet.peer.PeerUnreachableException;
 import dev.ltms.fleet.peer.SpawnRequest;
 import org.junit.jupiter.api.Test;
 import org.junit.jupiter.api.io.TempDir;
+import org.slf4j.LoggerFactory;
 
+import java.io.IOException;
 import java.nio.file.Files;
 import java.nio.file.Path;
+import java.nio.file.attribute.PosixFileAttributeView;
+import java.nio.file.attribute.PosixFilePermissions;
 import java.util.List;
 import java.util.Map;
 import java.util.concurrent.ExecutorService;
@@ -25,6 +33,7 @@ import java.util.concurrent.Future;
 import java.util.function.Supplier;
 
 import static org.junit.jupiter.api.Assertions.*;
+import static org.junit.jupiter.api.Assumptions.assumeTrue;
 
 /**
  * The opencode adapter's launch build: a file-based MCP mount + reply-charter instructions (no
@@ -610,4 +619,196 @@ class OpenCodeLauncherTest {
         assertTrue(json.path("mcp").path("intellij").isMissingNode(),
                 "no IDE server when ideMcpUrl is unset");
     }
+
+    // --- fleetd #219: config root + discovery root under memberHerdrSocket ------------------------
+
+    /** A config with {@code memberHerdrSocket:} set, and optionally {@code worktreeRoot:}/{@code worktreeGroup:}. */
+    private static FleetConfig configWithMemberHerdrSocket(String worktreeRoot, String worktreeGroup) {
+        return new FleetConfig(
+                null,                    // bind
+                null,                    // herdrSocket
+                "/tmp/other-user.sock",  // memberHerdrSocket
+                Map.of(),                // profiles
+                null,                    // guard
+                worktreeRoot,            // worktreeRoot
+                null,                    // lifecycle
+                null,                    // spawnReadyTimeoutMs
+                null,                    // spawnReadyPollMs
+                null,                    // broker
+                null,                    // primary
+                null,                    // fleet
+                null,                    // leadHeartbeat
+                null,                    // health
+                null,                    // placement
+                null,                    // auth
+                null,                    // configReload
+                null,                    // quarantineCooldownSeconds
+                null,                    // memberCredentials
+                null,                    // coordinator
+                worktreeGroup,           // worktreeGroup
+                null                     // memberLoginShell
+        ).withDefaults();
+    }
+
+    private static OpenCodeLauncher serviceWithConfig(FakeHerdr herdr, Path configRoot, Path discoveryRoot,
+                                                        FleetConfig.Profile cfg, Supplier<FleetConfig> config) {
+        return new OpenCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
+                Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null,
+                0, System::currentTimeMillis, () -> { }, configRoot, discoveryRoot, null, null, config);
+    }
+
+    /** The current process's own primary group — resolvable on whatever host runs this test. */
+    private static String currentUserGroup() throws IOException {
+        PosixFileAttributeView view = Files.getFileAttributeView(Path.of("."), PosixFileAttributeView.class);
+        assumeTrue(view != null, "this host's filesystem does not support POSIX group ownership");
+        return view.readAttributes().group().getName();
+    }
+
+    /**
+     * fleetd #219 site 1, acceptance criterion 1: with {@code memberHerdrSocket} configured and both
+     * {@code worktreeRoot}/{@code worktreeGroup} set, the generated {@code opencode.json} directory
+     * lives under {@code worktreeRoot} — NEVER under the injected {@code configRoot} (standing in for
+     * {@code java.io.tmpdir}, fleetd's own 0700 temp dir, unreadable by the member's different OS
+     * user) — and is shared read-only with the group via the SAME mechanism (fleetd #213's {@link
+     * EnvAllowListScrub#shareWithGroup}) the ZDOTDIR scrub uses.
+     */
+    @Test
+    void memberHerdrSocketWithWorktreeRootAndGroupPutsConfigDirUnderWorktreeRootAndSharesIt(
+            @TempDir Path configRoot, @TempDir Path worktreeRoot) throws Exception {
+        String group = currentUserGroup();
+        FakeHerdr herdr = new FakeHerdr();
+        serviceWithConfig(herdr, configRoot, configRoot,
+                opencodeCfg("google/gemini-2.5-pro", "http://127.0.0.1:8765/mcp", null),
+                () -> configWithMemberHerdrSocket(worktreeRoot.toString(), group)).spawn();
+
+        String cfgPath = startEnv(herdr).get("OPENCODE_CONFIG");
+        assertNotNull(cfgPath, "the profile still needs a config file");
+        Path cfgFile = Path.of(cfgPath);
+        Path dir = cfgFile.getParent();
+        assertEquals(worktreeRoot.toAbsolutePath().normalize(), dir.getParent(),
+                "the generated directory's parent must be worktreeRoot, not the injected configRoot "
+                        + "standing in for java.io.tmpdir — got parent " + dir.getParent());
+        assertFalse(dir.startsWith(configRoot),
+                "the generated directory must NOT be created under configRoot when memberHerdrSocket "
+                        + "is configured: " + dir);
+
+        assertEquals("rwxr-x---", PosixFilePermissions.toString(Files.getPosixFilePermissions(dir)),
+                "the directory must be group-traversable+readable, owner-only writable");
+        assertEquals("rw-r-----", PosixFilePermissions.toString(Files.getPosixFilePermissions(cfgFile)),
+                "opencode.json must be group-readable, never group-writable");
+    }
+
+    /**
+     * fleetd #219 site 1, acceptance criterion 2: with {@code memberHerdrSocket} configured but
+     * NEITHER {@code worktreeRoot} nor {@code worktreeGroup} set, the launcher must refuse the spawn
+     * rather than write a config under {@code java.io.tmpdir} the member cannot read — that member
+     * would occupy a pane and never become deliverable, with nothing pointing at the real cause.
+     */
+    @Test
+    void memberHerdrSocketWithoutWorktreeRootOrGroupRefusesTheSpawn(@TempDir Path configRoot) {
+        FakeHerdr herdr = new FakeHerdr();
+        OpenCodeLauncher launcher = serviceWithConfig(herdr, configRoot, configRoot,
+                opencodeCfg("google/gemini-2.5-pro", "http://127.0.0.1:8765/mcp", null),
+                () -> configWithMemberHerdrSocket(null, null));
+
+        IllegalStateException ex = assertThrows(IllegalStateException.class,
+                () -> launcher.spawn(new SpawnRequest(null, null, null)),
+                "a missing worktreeRoot/worktreeGroup must refuse the spawn, not write an unreadable config");
+        assertTrue(ex.getMessage().contains("worktreeRoot"),
+                "the refusal must name the missing config key — got: " + ex.getMessage());
+        assertFalse(herdr.called("tab.create"),
+                "the spawn must be refused BEFORE any pane is created — got calls: " + herdr.calls);
+    }
+
+    /**
+     * fleetd #219 site 1, acceptance criterion 2 (the other missing half): {@code worktreeRoot} set
+     * but {@code worktreeGroup} missing must ALSO refuse — either one alone is not enough to
+     * guarantee the member's OS user can read the directory.
+     */
+    @Test
+    void memberHerdrSocketWithWorktreeRootButNoGroupRefusesTheSpawn(
+            @TempDir Path configRoot, @TempDir Path worktreeRoot) {
+        FakeHerdr herdr = new FakeHerdr();
+        OpenCodeLauncher launcher = serviceWithConfig(herdr, configRoot, configRoot,
+                opencodeCfg("google/gemini-2.5-pro", "http://127.0.0.1:8765/mcp", null),
+                () -> configWithMemberHerdrSocket(worktreeRoot.toString(), null));
+
+        IllegalStateException ex = assertThrows(IllegalStateException.class,
+                () -> launcher.spawn(new SpawnRequest(null, null, null)));
+        assertTrue(ex.getMessage().contains("worktreeGroup"),
+                "worktreeRoot alone is not enough — got: " + ex.getMessage());
+    }
+
+    /**
+     * fleetd #219 site 1, acceptance criterion 3: with {@code memberHerdrSocket} ABSENT — even when
+     * a live, non-null {@code config} supplier is threaded through (not merely {@code config == null},
+     * which every other test in this file already exercises) — the generated directory must still
+     * land directly under the injected {@code configRoot}, byte-identical to before this fix.
+     */
+    @Test
+    void memberHerdrSocketAbsentStaysUnderConfigRootEvenWithALiveConfigSupplier(@TempDir Path configRoot)
+            throws Exception {
+        FakeHerdr herdr = new FakeHerdr();
+        FleetConfig config = new FleetConfig(null, null, null, Map.of(), null, null, null, null, null,
+                null, null, null, null, null, null, null, null, null, null, null, null, null).withDefaults();
+        serviceWithConfig(herdr, configRoot, configRoot,
+                opencodeCfg("google/gemini-2.5-pro", "http://127.0.0.1:8765/mcp", null), () -> config)
+                .spawn();
+
+        String cfgPath = startEnv(herdr).get("OPENCODE_CONFIG");
+        assertNotNull(cfgPath);
+        assertTrue(Path.of(cfgPath).startsWith(configRoot),
+                "with memberHerdrSocket absent, the config directory must still be created directly "
+                        + "under configRoot, unchanged from before this fix");
+    }
+
+    /**
+     * fleetd #219 site 2: under {@code memberHerdrSocket}, opencode session discovery must be
+     * declared unavailable rather than silently scanning fleetd's own {@code discoveryRoot} — which,
+     * under this config key, is NOT where the member's opencode actually writes its session
+     * database. This test proves the gate is real, not merely "no record yet": a matching record IS
+     * written to {@code discoveryRoot} (the exact fixture {@link
+     * #theHandleDiscoversTheSessionIdForTheWorkersCwdOnlyAfterItAppears} proves discovery would
+     * otherwise find), and {@code agentSessionId()} must still return {@code null} — proving the
+     * gate, not a coincidental absence of data, is what produced the null. One WARN is also logged,
+     * exactly once even across repeated calls.
+     */
+    @Test
+    void discoveryIsUnavailableUnderMemberHerdrSocketEvenWhenARecordExists(
+            @TempDir Path configRoot, @TempDir Path worktreeRoot, @TempDir Path discRoot) throws Exception {
+        String group = currentUserGroup();
+        OpenCodeSessionDiscoveryTest.writeRecord(discRoot, "ses_should_be_hidden", "/work/dir", 1000L);
+
+        FakeHerdr herdr = new FakeHerdr();
+        OpenCodeLauncher launcher = new OpenCodeLauncher(new AgentControl(herdr),
+                new WorkspaceControl(herdr), Map.of("gemini", opencodeCfg(null, null, null)),
+                "gemini", _ -> null, 0, System::currentTimeMillis, () -> { }, configRoot, discRoot,
+                null, null, () -> configWithMemberHerdrSocket(worktreeRoot.toString(), group));
+
+        Logger logger = (Logger) LoggerFactory.getLogger(OpenCodeLauncher.class);
+        Level original = logger.getLevel();
+        logger.setLevel(Level.INFO);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        logger.addAppender(appender);
+        PeerHandle handle;
+        try {
+            handle = launcher.spawn(new SpawnRequest(null, "/work/dir", null));
+            assertNull(handle.agentSessionId(),
+                    "memberHerdrSocket configured: discovery must stay unavailable even though a "
+                            + "matching record exists in discoveryRoot");
+            assertNull(handle.agentSessionId(), "the gate must hold on a second call too");
+        } finally {
+            logger.detachAppender(appender);
+            logger.setLevel(original);
+        }
+        List<String> warnings = appender.list.stream()
+                .filter(e -> e.getLevel() == Level.WARN)
+                .map(ILoggingEvent::getFormattedMessage)
+                .toList();
+        assertEquals(1, warnings.size(),
+                "exactly one WARN across two agentSessionId() calls — got: " + warnings);
+        assertTrue(warnings.get(0).contains("memberHerdrSocket"),
+                "the WARN must name memberHerdrSocket as the reason — got: " + warnings.get(0));
+    }
 }

fleetd/src/test/java/dev/ltms/fleet/session/SessionManagerTest.java

@@ -887,15 +887,18 @@ class SessionManagerTest {
     }
 
     @Test
-    void acquireWithNeitherSessionFieldLeavesAgentSessionIdNull() {
+    void acquireWithNeitherSessionFieldStillMintsAnAgentSessionId() {
+        // fleetd #214: the claude-code launcher mints a session id for EVERY spawn, so a member is
+        // resumable even when the spawn asked for no session identity.
         FakeHerdr herdr = new FakeHerdr();
         SessionManager sessions = sessionManager(herdr);
 
         MemberSession s = sessions.acquire("ltms-local", null, null, null);
 
-        assertNull(s.agentSessionId(), "no identity requested — unchanged from before CB-584");
-        assertFalse(SessionManager.rosterView(s, null).containsKey("agentSessionId"),
-                "a null id is omitted from the roster, like charterSha256 for a receipt-less session");
+        assertNotNull(s.agentSessionId(),
+                "fleetd #214: a plain spawn mints a session id, so every member is resumable");
+        assertTrue(SessionManager.rosterView(s, null).containsKey("agentSessionId"),
+                "the minted id is in the roster, so fleet_list advertises every member's resume handle");
     }
 
     @Test