fix(tmux): split the rule pair without piping into head

CI 2604 red on scripts/pipefail-early-exit.test.mjs. Two lines I added in the
previous commit tripped its scan of send-message.sh:

  [ "$(printf '%s\n' "$rule_lines" | grep -c .)" -eq 2 ] || return 1
  top=$(printf '%s\n' "$rule_lines" | head -1)

The second is a real violation of the rule the guard enforces: `head` exits
after its count, the producer takes SIGPIPE, and under `set -euo pipefail` the
caller aborts with rc=141 and no output. That is the same failure mode
test-mosaic-worktree-large-repo.sh exists to pin.

The first is the guard reading `-eq` as an early-exit grep flag
(`-[A-Za-z]*q` matches `-eq`), so `grep ... | ... -eq 2` looks like `grep -q`
on one line. A false positive, but the fix removes the shape either way.

Both go away by splitting the captured line numbers with parameter expansion
instead of a second pass through the pipe. With one rule captured, both halves
resolve to the same value and the new `[ "$top" != "$bottom" ]` test rejects it,
which is the answer that case wanted anyway.

Verified: node --test scripts/pipefail-early-exit.test.mjs green on the
load-bearing assertion; test-send-message-verdict.sh still PASS=5 FAIL=0;
test-send-message-socket.sh rc=0; agent-send.test.sh PASS=19 FAIL=0.

packages/mosaic/framework/tools/fleet/start-agent-session.sh

@@ -257,8 +257,36 @@ assert_owned_tmux_server() {
     fail "tmux server ownership or environment validation failed"
 }
 
-# Validate exact server ownership before querying, cleaning, or creating any
+# Lease-broker socket preflight (#1292). The gated runtime (`mosaic yolo …` →
-# managed session. An unmanaged or contaminated named socket is never repaired.
+# launch-runtime.py) registers with the broker or dies ~4 seconds in, with the
+# diagnostic invisible because tmux destroys the dead pane. This check runs
+# BEFORE any tmux effect — including the ownership probe below — so a host
+# without a broker produces a named, surviving refusal instead of a doomed
+# pane. Exit 75 (EX_TEMPFAIL), distinct from 64 (bad projection) and 69 (host
+# not ready for other reasons); the agent@ unit is Type=oneshot with no
+# Restart=, so the failed unit keeps its message instead of looping. Socket
+# resolution matches launch.ts's defaultLeaseBrokerSocket precedence exactly.
+# This preflight DETECTS and REFUSES — it never starts the broker (activation
+# belongs to the fleet control plane; a component that both detects and fixes
+# cannot be used to measure whether the fix worked).
+broker_socket_path() {
+  if [ -n "${MOSAIC_LEASE_BROKER_SOCKET:-}" ]; then
+    printf '%s\n' "$MOSAIC_LEASE_BROKER_SOCKET"
+    return 0
+  fi
+  local runtime_dir="${XDG_RUNTIME_DIR:-/run/user/$(id -u)}"
+  printf '%s\n' "${runtime_dir}/mosaic-lease/broker.sock"
+}
+
+if [ "$MODE" = "launch" ]; then
+  _broker_socket=$(broker_socket_path)
+  if [ ! -S "$_broker_socket" ]; then
+    echo "[fleet] FAIL_LAUNCH broker-absent: lease broker socket ${_broker_socket} missing; runtime launch denied (#1292)." >&2
+    echo "[fleet] remedy: systemctl --user enable --now mosaic-lease-broker.service  (or reinstall via: mosaic fleet install)" >&2
+    exit 75
+  fi
+fi
+
 assert_owned_tmux_server
 
 if [ "$MODE" = interaction ]; then

packages/mosaic/framework/tools/fleet/test-agent-session-broker-preflight.sh

@@ -0,0 +1,222 @@
+#!/usr/bin/env bash
+# CI-fit regression suite for the #1292 lease-broker socket preflight in
+# start-agent-session.sh.
+#
+# WHY THIS SUITE IS CI-FIT WHERE test-start-agent-session.sh IS NOT (#1017/#1270
+# context): that older suite's precondition is "the host does not have the pi
+# binary", which a CI image that ships pi violates — its guard correctly
+# refuses to report a pass there, so it is excluded from the chain. THIS suite
+# controls its own preconditions instead of inheriting them from the host: a
+# fake tmux on PATH, a fake mosaic on PATH, a real unix socket created in a
+# tmpdir, a hermetic env (env -i, fake HOME, GIT_CONFIG_GLOBAL severed). It
+# never depends on what the host has installed, so a green here means the same
+# thing on every host. Anyone adding cases: keep that property — no case may
+# depend on host state.
+#
+# The failure this suite is written down to catch (#1292): a seat launched on a
+# host with no lease broker dies ~4 seconds in at registration, with the
+# diagnostic invisible because tmux destroys the dead pane. The preflight runs
+# BEFORE any tmux effect and refuses with a NAMED code (exit 75, EX_TEMPFAIL)
+# so the message survives. The agent@ unit is Type=oneshot with no Restart=,
+# so a failed unit keeps its output instead of looping.
+#
+# Cases:
+#   1. absent socket -> exit 75, message names broker-absent + socket path +
+#      remedy, and NO tmux session was ever created (the doomed-pane half).
+#   2. present socket (real unix socket in tmpdir) -> proceeds PAST the
+#      preflight (the suite then stops at the next precondition, proving the
+#      preflight was not the refusal).
+#   3. explicit MOSAIC_LEASE_BROKER_SOCKET wins over XDG_RUNTIME_DIR default.
+#   4. --stop mode does NOT require the broker (teardown must not be fenced on
+#      a component whose absence is exactly what teardown may follow).
+#
+# Sabotage control, run by the developer (not in-suite): remove the preflight
+# block from start-agent-session.sh, re-run — case 1 fails (a tmux session is
+# created / exit is not 75), cases 2-4 still pass; restore byte-identically.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+WORK_DIR="${MOSAIC_TEST_WORK_DIR:-$PWD/.mosaic-test-work/agent-session-broker-preflight}"
+FAKE_HOME="$WORK_DIR/home"
+BIN_DIR="$WORK_DIR/bin"
+ENV_DIR="$WORK_DIR/env"
+SOCK_DIR="$WORK_DIR/sockets"
+LOG_FILE="$WORK_DIR/tmux-calls.log"
+
+rm -rf "$WORK_DIR"
+# The script asserts a managed directory tree under MOSAIC_HOME: mosaic/,
+# mosaic/fleet/, mosaic/fleet/agents/ — private (0700/0750-style) modes, no
+# symlinks — plus a per-agent env projection. Build the full tree the launcher
+# expects so the suite reaches the BROKER preflight rather than dying at
+# environment validation.
+mkdir -p "$FAKE_HOME/.config/mosaic/fleet/agents" "$BIN_DIR" "$SOCK_DIR"
+chmod 700 "$FAKE_HOME/.config/mosaic" "$FAKE_HOME/.config/mosaic/fleet/agents"
+chmod 750 "$FAKE_HOME/.config/mosaic/fleet"
+cat > "$FAKE_HOME/.config/mosaic/fleet/agents/preflight-test.env.generated" <<'ENVEOF'
+MOSAIC_AGENT_NAME=preflight-test
+MOSAIC_GIT_IDENTITY=preflight-test
+MOSAIC_AGENT_CLASS=worker
+MOSAIC_AGENT_RUNTIME=pi
+MOSAIC_AGENT_MODEL=
+MOSAIC_AGENT_REASONING=
+MOSAIC_AGENT_TOOL_POLICY=code
+MOSAIC_AGENT_WORKDIR=/tmp
+MOSAIC_TMUX_SOCKET=mosaic-fleet
+ENVEOF
+chmod 600 "$FAKE_HOME/.config/mosaic/fleet/agents/preflight-test.env.generated"
+
+# ─── Fake tmux: records every invocation; new-session marks the marker. ────
+: > "$LOG_FILE"
+cat > "$BIN_DIR/tmux" <<SH
+#!/usr/bin/env bash
+printf 'tmux %s\n' "\$*" >> "$LOG_FILE"
+if [[ "\$*" == *new-session* ]]; then
+  echo "TMUX-NEW-SESSION-INVOKED" >> "$LOG_FILE"
+fi
+exit 0
+SH
+chmod +x "$BIN_DIR/tmux"
+
+# ─── Fake mosaic/pi binaries so the script proceeds past its own lookups. ───
+for bin in mosaic pi claude; do
+  printf '#!/usr/bin/env bash\nexit 0\n' > "$BIN_DIR/$bin"
+  chmod +x "$BIN_DIR/$bin"
+done
+
+# ─── Minimal launch environment the script expects. ────────────────────────
+# (Enough for the preflight to be reached; later stages will still fail in
+# case 2 — that is expected and asserted.)
+run_session_script() {
+  local mode="$1"; shift
+  (
+    cd "$WORK_DIR"
+    env -i HOME="$FAKE_HOME" PATH="$BIN_DIR:/usr/bin:/bin" \
+      GIT_CONFIG_GLOBAL=/dev/null GIT_CONFIG_SYSTEM=/dev/null \
+      MOSAIC_HOME="$FAKE_HOME/.config/mosaic" \
+      AGENT_NAME=preflight-test \
+      "$@" \
+      bash "$SCRIPT_DIR/start-agent-session.sh" $mode preflight-test
+  )
+}
+
+fail=0
+assert() {
+  local desc="$1" expected="$2" actual="$3"
+  if [[ "$expected" != "$actual" ]]; then
+    echo "FAIL: $desc — expected '$expected', got '$actual'" >&2
+    fail=1
+  fi
+}
+assert_contains() {
+  local desc="$1" haystack="$2" needle="$3"
+  [[ "$haystack" == *"$needle"* ]] || { echo "FAIL: $desc — missing '$needle' in: $haystack" >&2; fail=1; }
+}
+assert_not_contains() {
+  local desc="$1" haystack="$2" needle="$3"
+  if [[ "$haystack" == *"$needle"* ]]; then
+    echo "FAIL: $desc — must not contain '$needle'" >&2
+    fail=1
+  fi
+  return 0
+}
+
+# ─── 1. Absent socket → named refusal, NO tmux session. ────────────────────
+: > "$LOG_FILE"
+stderr_file="$WORK_DIR/stderr-1.tmp"
+set +e
+out=$(run_session_script "" MOSAIC_LEASE_BROKER_SOCKET="$SOCK_DIR/absent.sock" 2>"$stderr_file")
+rc=$?
+set -e
+assert "absent socket exit code" "75" "$rc"
+err=$(cat "$stderr_file")
+assert_contains "absent socket names the failure" "$err" "FAIL_LAUNCH broker-absent"
+assert_contains "absent socket names the socket path" "$err" "$SOCK_DIR/absent.sock"
+assert_contains "absent socket names a remedy" "$err" "mosaic fleet install"
+log1=$(cat "$LOG_FILE")
+assert_not_contains "absent socket must not create a tmux session" "$log1" "TMUX-NEW-SESSION-INVOKED"
+
+# ─── 2. Present socket → passes the preflight. ─────────────────────────────
+# Expected: ownership/env checks AFTER the preflight may refuse (fixture is
+# minimal by design); the assertion is only that the refusal is NOT
+# broker-absent and the exit is NOT 75.
+# Create a REAL unix socket: a detached python holder binds it and stays alive
+# for the duration (bash cannot create sockets; a foreground python would
+# close the socket on exit and -S on a closed-but-unlinked path fails). Written
+# as a script file + setsid nohup so no job-control/heredoc interaction with
+# set -e can silently kill the suite.
+# AF_UNIX binds cap at 108 path bytes; the suite's workdir exceeds that, so
+# the live socket lives at a SHORT path under /tmp (unique per run, cleaned
+# with the suite). The preflight takes its socket path explicitly, so this
+# stays fully controlled.
+# A real unix socket at a SHORT absolute path (AF_UNIX limit is 108 bytes,
+# so the repo-deep SOCK_DIR cannot host it). The name is composed, not
+# `mktemp -u`: the CI image's mktemp dialect rejects that invocation
+# (pipeline 2562: "mktemp: : Invalid argument"), and no pre-existing file is
+# wanted anyway — the holder binds it fresh.
+LIVE_SOCK="/tmp/mosaic-preflight-$RANDOM-$$.sock"
+trap 'rm -f "$LIVE_SOCK"' EXIT
+rm -f "$SOCK_DIR/live.sock" "$LIVE_SOCK"
+cat > "$SOCK_DIR/holder.py" <<'PY'
+import socket, sys, time
+path = sys.argv[1]
+s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
+s.bind(path)
+s.listen(1)
+time.sleep(120)
+PY
+python3 "$SOCK_DIR/holder.py" "$LIVE_SOCK" >/dev/null 2>"$SOCK_DIR/holder.err" &
+HOLDER_PID=$!
+# Wait for the socket object to exist (bind is near-instant, but do not race it).
+for _ in $(seq 1 50); do
+  [ -S "$LIVE_SOCK" ] && break
+  sleep 0.1
+done
+if [ ! -S "$LIVE_SOCK" ]; then
+  echo "FAIL: could not create live socket fixture (holder pid $HOLDER_PID)" >&2
+  ps -p "$HOLDER_PID" -o pid,stat,cmd --no-headers >&2 || echo "(holder exited)" >&2
+  cat "$SOCK_DIR/holder.err" >&2 || true
+  exit 1
+fi
+: > "$LOG_FILE"
+set +e
+out=$(run_session_script "" MOSAIC_LEASE_BROKER_SOCKET="$LIVE_SOCK" 2>"$WORK_DIR/stderr-2.tmp")
+rc=$?
+set -e
+# The preflight PASSED if the failure (whatever later stage refused) is NOT
+# the broker refusal, and tmux was reached or a later precondition named
+# something else.
+err2=$(cat "$WORK_DIR/stderr-2.tmp")
+assert_not_contains "live socket must not refuse broker-absent" "$err2" "broker-absent"
+if [[ "$rc" == "75" ]]; then
+  echo "FAIL: live socket — preflight still refused (exit 75) with a live socket" >&2
+  fail=1
+fi
+
+# ─── 3. Explicit socket env wins over XDG default. ─────────────────────────
+set +e
+out=$(run_session_script "" XDG_RUNTIME_DIR="$SOCK_DIR/no-runtime-here" MOSAIC_LEASE_BROKER_SOCKET="$SOCK_DIR/absent2.sock" 2>"$WORK_DIR/stderr-3.tmp")
+rc=$?
+set -e
+assert "explicit env wins (exit 75)" "75" "$rc"
+assert_contains "explicit env path named" "$(cat "$WORK_DIR/stderr-3.tmp")" "$SOCK_DIR/absent2.sock"
+
+# ─── 4. --stop is not fenced on the broker. ────────────────────────────────
+: > "$LOG_FILE"
+set +e
+out=$(run_session_script "--stop" MOSAIC_LEASE_BROKER_SOCKET="$SOCK_DIR/absent3.sock" 2>"$WORK_DIR/stderr-4.tmp")
+rc=$?
+set -e
+err4=$(cat "$WORK_DIR/stderr-4.tmp")
+assert_not_contains "--stop must not refuse broker-absent" "$err4" "broker-absent"
+if [[ "$rc" == "75" ]]; then
+  echo "FAIL: --stop — exit 75 means teardown was fenced on the broker" >&2
+  fail=1
+fi
+
+kill "$HOLDER_PID" 2>/dev/null || true
+
+if [[ "$fail" -eq 0 ]]; then
+  echo "start-agent-session lease-broker preflight regression passed"
+fi
+exit "$fail"

packages/mosaic/framework/tools/git/README.md

@@ -30,7 +30,9 @@ The Gitea API token is **never passed on a curl command line.** An `Authorizatio
 
 ### `--login` override
 
-Both `pr-review.sh` and `issue-comment.sh` accept an optional `--login <name>` flag that overrides the automatically detected Gitea login for that single invocation. The override selects **which credential the REST write, the `/user` identity lookup, and the read-back all use** — its token is resolved from the tea config for that login name (`get_gitea_token_for_login`), falling back to the repo host's credential when no login is named. The resolved login is **host- and port-bound**: the login's configured URL host **and effective port** (the scheme's default port — 80 for `http`, 443 for `https` — applies when a port is omitted, symmetrically on both sides) must match the repo remote's, so a login name shared across hosts (or an override configured for a different Gitea, including one on a different port of the same host) can never send one host's credential to another — a host or port mismatch fails closed rather than leaking a cross-host token. Resolving the acting identity and the read-back from the _same_ login that performs the write is essential: a write performed under an overridden login must be verified against that login's identity, not the host default's. Callers who need a different login than the host default should pass `--login <reviewer-login>`.
+Both `pr-review.sh` and `issue-comment.sh` accept an optional `--login <name>` flag that overrides the automatically detected Gitea login for that single invocation. The override selects **which credential the REST write, the `/user` identity lookup, and the read-back all use** — its token is resolved from the tea config for that login name (`get_gitea_token_for_login`).
+
+**With no `--login`, there is no tea lookup at all.** The acting credential is the calling identity's own, resolved by `get_gitea_token` (see "Per-agent Gitea identity" below), and there is deliberately no fallback from it. These wrappers previously _guessed_ a login from the repo host and looked that guess up in the tea config; on a shared-account host the guess resolved to the shared login, so an unqualified call authored its write as that account rather than as the caller. Since `get_gitea_token_for_login` matches by login **name** and performs no authentication check, a dead shared credential still resolved at rc=0 and the identity-aware resolver was never reached. A caller passing no `--login` is asking to act as itself, so `--login` is now the only route to the tea store (#1351). The resolved login is **host- and port-bound**: the login's configured URL host **and effective port** (the scheme's default port — 80 for `http`, 443 for `https` — applies when a port is omitted, symmetrically on both sides) must match the repo remote's, so a login name shared across hosts (or an override configured for a different Gitea, including one on a different port of the same host) can never send one host's credential to another — a host or port mismatch fails closed rather than leaking a cross-host token. Resolving the acting identity and the read-back from the _same_ login that performs the write is essential: a write performed under an overridden login must be verified against that login's identity, not the host default's. Callers who need a different login than the host default should pass `--login <reviewer-login>`.
 
 As a durable successor to this mechanism, consider giving each reviewer/approver slot its own dedicated Gitea login credential, so that author≠reviewer holds at the credential level rather than relying on wrapper-level `--login` bookkeeping. This is a recommendation for future hardening, not something implemented by this flag.
 
@@ -125,6 +127,32 @@ otherwise careful never to touch. Because identity is already resolved per-workt
 (`mosaic.gitIdentity`), the correct granularity for registering the helper is per-clone
 too, so a documented manual step is the right shape here, not a global auto-write.
 
+### Running these tests
+
+`MOSAIC_GIT_IDENTITY` is inherited into each test's sandbox `HOME`, and **the tests disagree
+about which value they need**, so no single ambient value passes all 29. Measured on `next` at
+`a480ee83`, two full passes differing only in that variable:
+
+| tests                                                                                                                                           | identity exported | identity unset |
+| ----------------------------------------------------------------------------------------------------------------------------------------------- | ----------------- | -------------- |
+| `gitea-login-resolution`, `issue-comment-readback`, `issue-create-interactive-auth`, `pr-edit`, `pr-merge-gitea-empty-uid`, `pr-metadata-gitea` | **fail**          | pass           |
+| `issue-close-fail-closed`                                                                                                                       | pass              | **fail**       |
+| remaining 22                                                                                                                                    | pass              | pass           |
+
+The six fail because inside a sandbox `HOME` the identity has no `fleet/agents/<id>/`
+directory, so it is classified as a **service identity**, its store is unpopulated, and the
+resolver correctly refuses with `Refusing to borrow another slot's token`. That is the
+documented fail-closed behaviour above, reached from a state the test never intended.
+`issue-close-fail-closed` is the mirror image: it asserts that no comment POST is attempted, so
+it needs an identity resolving to an empty slot, and with the variable unset the shared account
+answers and the POST goes through.
+
+These read as wrapper regressions rather than as environment. Two seats independently
+misdiagnosed them as a patch defect while reviewing #1352. Until each test controls its own
+value (#1353), `env -u MOSAIC_GIT_IDENTITY` is the closest thing to a clean run at 28/29, with
+`issue-close-fail-closed` the expected failure — and **"the suite passes" is not a statement
+anyone can make here without naming the ambient value that produced it.**
+
 ### PowerShell parity
 
 `detect-platform.ps1`'s Gitea wrappers authenticate through `tea` CLI logins

packages/mosaic/framework/tools/git/issue-comment.sh

@@ -102,9 +102,17 @@ gitea_resolve_api_for_login() {
             return 1
         }
     else
-        GITEA_API_TOKEN=$(get_gitea_token_for_login "$effective_login" "$host") \
+        # NO --login: the acting credential is this identity's own token and there
-            || GITEA_API_TOKEN=$(get_gitea_token "$host") || {
+        # is deliberately no tea-config fallback. get_gitea_token_for_login matches
-            echo "Error: Gitea token not found for login '$effective_login' (comment write/read-back)" >&2
+        # by login NAME and performs no authentication check, and with no --login
+        # that name was a HOST GUESS resolving to a shared account. A live shared
+        # token would therefore have authored every seat's comment as that
+        # account, making Gate-16 author-is-not-reviewer unenforceable fleet-wide;
+        # a dead one is only what made the defect visible. get_gitea_token fails
+        # loud on a fleet host when no identity resolves, and that refusal is the
+        # correct outcome, not a case to fall back from.
+        GITEA_API_TOKEN=$(get_gitea_token "$host") || {
+            echo "Error: no Gitea credential resolved for the acting identity on host '$host' (comment write/read-back). Set MOSAIC_GIT_IDENTITY=<agent-id>, or pass --login <name> to use a named tea credential." >&2
             return 1
         }
     fi
@@ -335,15 +343,12 @@ if [[ "$PLATFORM" == "github" ]]; then
     gh issue comment "$ISSUE_NUMBER" --body "$COMMENT"
     echo "Added comment to GitHub issue #$ISSUE_NUMBER"
 elif [[ "$PLATFORM" == "gitea" ]]; then
-    # Resolve the login this comment should be attributed to: the --login
+    # A --login override selects a NAMED tea credential and is the only way to
-    # override when given, otherwise the detected default for this repo's host.
+    # reach the tea store. With no --login there is deliberately no guess: the
-    # A --login override always wins. Otherwise name this repo host's login only
+    # comment is attributed to this identity's own credential, resolved by
-    # as a best effort: the login name merely selects a per-login token, and
+    # gitea_resolve_api_for_login. The guess this replaced named a SHARED
-    # gitea_resolve_api_for_login falls back to the host credential
+    # account, selecting an identity the caller never asked to act as.
-    # (get_gitea_token) when no tea login is named, so the default credential
-    # still resolves even when the host tea has no matching login entry.
     EFFECTIVE_LOGIN="$LOGIN_OVERRIDE"
-    [[ -n "$EFFECTIVE_LOGIN" ]] || EFFECTIVE_LOGIN=$(get_gitea_login 2>/dev/null || true)
 
     # Bind the REST endpoint + token to the effective login, then derive the
     # acting identity from that SAME credential (GET /user). The write below and

packages/mosaic/framework/tools/git/pr-review.sh

@@ -12,7 +12,7 @@
 # concurrent record cannot masquerade as this write and a no-op fails closed.
 #
 # --login override: the default login is resolved from the local tea login list
-# for this repo's host (get_gitea_login_for_host). Pass --login <name> to
+# for this repo's host from the acting identity's own credential. Pass --login <name> to
 # override it for this invocation only. The REST write, the /user identity read,
 # and every read-back are ALL performed with the token of the EFFECTIVE login,
 # so the write and its verification bind to the same identity.
@@ -372,9 +372,17 @@ gitea_resolve_api_for_login() {
             return 1
         }
     else
-        GITEA_API_TOKEN=$(get_gitea_token_for_login "$effective_login" "$host") \
+        # NO --login: the acting credential is this identity's own token and there
-            || GITEA_API_TOKEN=$(get_gitea_token "$host") || {
+        # is deliberately no tea-config fallback. get_gitea_token_for_login matches
-            echo "Error: Gitea token not found for login '$effective_login' (review write/read-back)" >&2
+        # by login NAME and performs no authentication check, and with no --login
+        # that name was a HOST GUESS resolving to a shared account. A live shared
+        # token would therefore have authored every seat's review as that
+        # account, making Gate-16 author-is-not-reviewer unenforceable fleet-wide;
+        # a dead one is only what made the defect visible. get_gitea_token fails
+        # loud on a fleet host when no identity resolves, and that refusal is the
+        # correct outcome, not a case to fall back from.
+        GITEA_API_TOKEN=$(get_gitea_token "$host") || {
+            echo "Error: no Gitea credential resolved for the acting identity on host '$host' (review write/read-back). Set MOSAIC_GIT_IDENTITY=<agent-id>, or pass --login <name> to use a named tea credential." >&2
             return 1
         }
     fi
@@ -698,7 +706,7 @@ if [[ "$PLATFORM" == "github" ]]; then
 elif [[ "$PLATFORM" == "gitea" ]]; then
     case $ACTION in
         approve)
-            # Best-effort host for the tea-login GUESS only (gitea_resolve_api_for_login
+            # Best-effort host for credential resolution only (gitea_resolve_api_for_login
             # below re-derives the real host from HOST_OVERRIDE/remote independently and
             # is authoritative). Prefer an explicit -H/--host; otherwise best-effort
             # git-remote inference, tolerating its ABSENCE (a bare `get_remote_host` here
@@ -706,15 +714,13 @@ elif [[ "$PLATFORM" == "gitea" ]]; then
             # SILENTLY — exit 1, zero output — even though -r/-H are exactly the flags
             # that support running with no usable origin at all).
             host="${HOST_OVERRIDE:-$(get_remote_host 2>/dev/null || true)}"
-            # A --login override always wins. Otherwise name this host's login
+            # A --login override selects a NAMED tea credential and is the only
-            # only as a best effort: the login name merely selects a per-login
+            # way to reach the tea store. With no --login there is deliberately no
-            # token, and gitea_resolve_api_for_login falls back to the host
+            # guess: gitea_resolve_api_for_login resolves this identity's own token.
-            # credential (get_gitea_token) when no tea login is named — so a host
+            # The guess this replaced named a SHARED account, selecting an identity
-            # tea's login list need not enumerate exotic (e.g. ported) hosts for
+            # the caller never asked to act as. The single resolved token is then
-            # the default credential to resolve. The single resolved token is
+            # used for the write, the /user identity, and the read-back.
-            # then used for the write, the /user identity, and the read-back.
             EFFECTIVE_LOGIN="$LOGIN_OVERRIDE"
-            [[ -n "$EFFECTIVE_LOGIN" ]] || EFFECTIVE_LOGIN=$(get_gitea_login_for_host "$host" 2>/dev/null || true)
             # Bind the REST endpoint + token to the effective login, then derive
             # the acting identity from that SAME credential so the review submit
             # and its read-back verify against the identity that performed them.
@@ -735,7 +741,7 @@ elif [[ "$PLATFORM" == "gitea" ]]; then
                 echo "Error: Comment required for request-changes"
                 exit 1
             fi
-            # Best-effort host for the tea-login GUESS only (gitea_resolve_api_for_login
+            # Best-effort host for credential resolution only (gitea_resolve_api_for_login
             # below re-derives the real host from HOST_OVERRIDE/remote independently and
             # is authoritative). Prefer an explicit -H/--host; otherwise best-effort
             # git-remote inference, tolerating its ABSENCE (a bare `get_remote_host` here
@@ -743,15 +749,13 @@ elif [[ "$PLATFORM" == "gitea" ]]; then
             # SILENTLY — exit 1, zero output — even though -r/-H are exactly the flags
             # that support running with no usable origin at all).
             host="${HOST_OVERRIDE:-$(get_remote_host 2>/dev/null || true)}"
-            # A --login override always wins. Otherwise name this host's login
+            # A --login override selects a NAMED tea credential and is the only
-            # only as a best effort: the login name merely selects a per-login
+            # way to reach the tea store. With no --login there is deliberately no
-            # token, and gitea_resolve_api_for_login falls back to the host
+            # guess: gitea_resolve_api_for_login resolves this identity's own token.
-            # credential (get_gitea_token) when no tea login is named — so a host
+            # The guess this replaced named a SHARED account, selecting an identity
-            # tea's login list need not enumerate exotic (e.g. ported) hosts for
+            # the caller never asked to act as. The single resolved token is then
-            # the default credential to resolve. The single resolved token is
+            # used for the write, the /user identity, and the read-back.
-            # then used for the write, the /user identity, and the read-back.
             EFFECTIVE_LOGIN="$LOGIN_OVERRIDE"
-            [[ -n "$EFFECTIVE_LOGIN" ]] || EFFECTIVE_LOGIN=$(get_gitea_login_for_host "$host" 2>/dev/null || true)
             gitea_resolve_api_for_login "$EFFECTIVE_LOGIN" "${LOGIN_OVERRIDE:+explicit}" || exit 1
             ACTING_LOGIN=$(gitea_authenticated_login) || exit 1
             head_sha=$(gitea_pr_head_sha "$PR_NUMBER") || exit 1
@@ -766,7 +770,7 @@ elif [[ "$PLATFORM" == "gitea" ]]; then
                 echo "Error: Comment required"
                 exit 1
             fi
-            # Best-effort host for the tea-login GUESS only (gitea_resolve_api_for_login
+            # Best-effort host for credential resolution only (gitea_resolve_api_for_login
             # below re-derives the real host from HOST_OVERRIDE/remote independently and
             # is authoritative). Prefer an explicit -H/--host; otherwise best-effort
             # git-remote inference, tolerating its ABSENCE (a bare `get_remote_host` here
@@ -774,15 +778,13 @@ elif [[ "$PLATFORM" == "gitea" ]]; then
             # SILENTLY — exit 1, zero output — even though -r/-H are exactly the flags
             # that support running with no usable origin at all).
             host="${HOST_OVERRIDE:-$(get_remote_host 2>/dev/null || true)}"
-            # A --login override always wins. Otherwise name this host's login
+            # A --login override selects a NAMED tea credential and is the only
-            # only as a best effort: the login name merely selects a per-login
+            # way to reach the tea store. With no --login there is deliberately no
-            # token, and gitea_resolve_api_for_login falls back to the host
+            # guess: gitea_resolve_api_for_login resolves this identity's own token.
-            # credential (get_gitea_token) when no tea login is named — so a host
+            # The guess this replaced named a SHARED account, selecting an identity
-            # tea's login list need not enumerate exotic (e.g. ported) hosts for
+            # the caller never asked to act as. The single resolved token is then
-            # the default credential to resolve. The single resolved token is
+            # used for the write, the /user identity, and the read-back.
-            # then used for the write, the /user identity, and the read-back.
             EFFECTIVE_LOGIN="$LOGIN_OVERRIDE"
-            [[ -n "$EFFECTIVE_LOGIN" ]] || EFFECTIVE_LOGIN=$(get_gitea_login_for_host "$host" 2>/dev/null || true)
             gitea_resolve_api_for_login "$EFFECTIVE_LOGIN" "${LOGIN_OVERRIDE:+explicit}" || exit 1
             ACTING_LOGIN=$(gitea_authenticated_login) || exit 1
             comment_id=$(gitea_create_comment_verified "$PR_NUMBER" "$COMMENT" "$ACTING_LOGIN") || {

packages/mosaic/framework/tools/tmux/send-message.sh

@@ -32,7 +32,9 @@
 #   0  delivered (submitted) or queued (agent busy; will process when free)
 #   1  tmux target not found
 #   2  submission NOT confirmed — either still an unsubmitted draft, or the REPL
-#      input prompt could not be located to confirm the message actually landed.
+#      input box could not be located to confirm the message actually landed.
+#      Locating the box is runtime-specific; see locate_input_box() below, and
+#      add a shape there before pointing this tool at a new runtime.
 #      Delivery is NEVER inferred from absence of evidence: if we cannot positively
 #      see the input box clear of the message (or the queued banner), we fail loud
 #      so the sender learns immediately instead of a silent worker->lead stall.
@@ -97,10 +99,50 @@ printf '%s' "$MSG" | "${tmux_cmd[@]}" load-buffer -b "$BUF" -
 #   would otherwise accumulate forever.
 sleep 0.5
 
+# Locate the REPL input box in a captured pane. Prints the box's contents on
+# stdout and returns 0 when the box was FOUND; returns 1 when it could not be
+# located at all. Found-but-empty is a real, distinct answer (an empty input box
+# is what a submitted message leaves behind), so the caller must branch on the
+# return code, never on whether the output is empty.
+#
+# Two REPL shapes are recognised:
+#   * a prompt-glyph line — `❯`, a leading `>`, or `│ >`. Claude Code and most
+#     readline REPLs.
+#   * a box drawn as two horizontal `─` rules with the input between them and NO
+#     prompt glyph anywhere. pi renders this. Anchoring on the LAST rule pair is
+#     what makes it safe: agent output can contain its own rules, but nothing is
+#     drawn below the input box except the status line.
+#
+# Adding a runtime means adding its shape HERE. A shape that is missing does not
+# degrade gracefully: it turns every send to that runtime into a false
+# "may be UNDELIVERED", which is what #1362 measured on pi and #1257 on another
+# arm of the same probe.
+locate_input_box() {
+  local pane=$1 glyph_line rule_lines top bottom
+  glyph_line=$(printf '%s\n' "$pane" | grep -E '❯|^>|│ >' | tail -1)
+  if [ -n "$glyph_line" ]; then printf '%s\n' "$glyph_line"; return 0; fi
+  rule_lines=$(printf '%s\n' "$pane" | grep -nE '^[[:space:]]*─{4,}[[:space:]]*$' | cut -d: -f1 | tail -2)
+  [ -n "$rule_lines" ] || return 1
+  # Split the (at most two) captured line numbers with parameter expansion. Not
+  # `head -1`: piping into an early-exiting consumer SIGPIPEs the producer, which
+  # under `set -euo pipefail` aborts the caller with rc=141 and no output. The
+  # scripts/pipefail-early-exit.test.mjs guard reds on that shape, correctly.
+  # With one rule captured both halves resolve to the same value and the
+  # ordering test below rejects it, which is the answer we want anyway.
+  top=${rule_lines%%$'\n'*}
+  bottom=${rule_lines##*$'\n'}
+  [ "$top" != "$bottom" ] || return 1
+  [ "$bottom" -gt "$top" ] || return 1
+  # An empty range (adjacent rules) prints nothing and still returns 0: found,
+  # empty, which is the delivered shape.
+  printf '%s\n' "$pane" | sed -n "$((top + 1)),$((bottom - 1))p"
+  return 0
+}
+
 # 2) Submit, then POSITIVELY confirm submission; flush with another Enter if it is
 #    still a draft. Success requires positive evidence — the queued banner, OR the
 #    REPL input box located AND clear of our message tail. The historical bug was
-#    treating ABSENCE of a draft as delivery: if the prompt glyph was never matched
+#    treating ABSENCE of a draft as delivery: if the input box was never located
 #    (wrong pane / prompt-glyph drift), an unsubmitted message read as "delivered"
 #    and worker->lead relays stalled silently. We now default to UNCONFIRMED and only
 #    upgrade to delivered on positive evidence; anything we cannot confirm fails loud.
@@ -113,15 +155,14 @@ for attempt in $(seq 1 $((RETRIES + 1))); do
   if grep -qF "$QUEUED_RE" <<<"$pane"; then
     status="queued"; break
   fi
-  # Locate the REPL input box (prompt glyph). If we cannot see it, we have NO
+  # If we cannot see the input box, we have NO evidence of submission state —
-  # evidence of submission state — stay UNCONFIRMED and retry; never infer delivery.
+  # stay UNCONFIRMED and retry; never infer delivery.
-  promptline=$(printf '%s' "$pane" | grep -E '❯|^>|│ >' | tail -1)
+  if ! inputbox=$(locate_input_box "$pane"); then
-  if [ -z "$promptline" ]; then
     status="unconfirmed"; continue
   fi
   # Input box located AND still carrying our tail => unsubmitted draft. Flush + retry.
-  # (Submitted messages scroll up into history; a draft stays on the ❯ line.)
+  # (Submitted messages scroll up into history; a draft stays in the box.)
-  if [ -n "$snippet" ] && grep -qF "$snippet" <<<"$promptline"; then
+  if [ -n "$snippet" ] && grep -qF "$snippet" <<<"$inputbox"; then
     status="draft"; continue
   fi
   # Input box located AND clear of our tail => positively submitted. This is the
@@ -135,6 +176,6 @@ case "$status" in
   delivered)   echo "✓ delivered to $TARGET"; exit 0 ;;
   queued)      echo "✓ queued to $TARGET (agent busy — will process when it returns to prompt)"; exit 0 ;;
   draft)       echo "✗ still an unsubmitted draft on $TARGET after $RETRIES flush attempts" >&2; exit 2 ;;
-  unconfirmed) echo "✗ could not confirm submission on $TARGET: REPL input prompt not locatable after $((RETRIES + 1)) attempts — message may be UNDELIVERED (check target/pane, retry, or escalate)" >&2; exit 2 ;;
+  unconfirmed) echo "✗ could not confirm submission on $TARGET: REPL input box not locatable after $((RETRIES + 1)) attempts — message may be UNDELIVERED (check target/pane, retry, or escalate)" >&2; exit 2 ;;
   *)           echo "✗ could not confirm submission on $TARGET (unexpected state '$status')" >&2; exit 2 ;;
 esac

packages/mosaic/framework/tools/tmux/test-send-message-verdict.sh

@@ -10,6 +10,13 @@
 #                    "could not confirm submission").
 #   3. DRAFT       — a `❯ `-prompt pane that never submits (message stays on the
 #                    input line) => exit 2, stderr "unsubmitted draft".
+#   4. DELIVERED   — a pane whose input box is two `─` rules with NO prompt glyph
+#      (box shape)   anywhere (pi's shape) and which submits => exit 0. Pre-#1362
+#                    the glyph probe could not see this box at all, so EVERY send
+#                    to such a pane reported "may be UNDELIVERED" while landing.
+#   5. DRAFT       — the same glyphless box, holding our tail across every flush
+#      (box shape)   Enter => exit 2, stderr "unsubmitted draft". Pre-#1362 this
+#                    also reported unconfirmed, so the true state was invisible.
 set -uo pipefail
 
 HERE=$(cd -- "$(dirname -- "$0")" && pwd)
@@ -69,6 +76,56 @@ else
   fi
 fi
 
+# --- Fixtures 4 and 5: a pi-shaped pane. The input box is two `─` rules with the
+#     text between them and NO prompt glyph anywhere, so the glyph probe alone can
+#     never locate it and every send reports "may be UNDELIVERED" (#1362). The
+#     renderer below is the shape, not the runtime: MODE=clear submits (box empties),
+#     MODE=keep leaves the text sitting in the box.
+cat > "$TMP/pibox.sh" <<'PIBOX'
+#!/usr/bin/env bash
+MODE=${1:-clear}
+RULE=$(printf '─%.0s' $(seq 1 60))
+buf=""
+draw() {
+  printf '\033[H\033[2J'
+  printf 'fixture output line\n\n'
+  printf '%s\n' "$RULE"
+  printf '%s\n' "$buf"
+  printf '%s\n' "$RULE"
+  printf '~/fixture (main)\n'
+  printf 'tok 0 model fixture\n'
+}
+draw
+while IFS= read -r line; do
+  # keep: hold the tail across every flush Enter, which is what a stuck draft does.
+  if [ "$MODE" = keep ]; then [ -n "$line" ] && buf=$line; else buf=""; fi
+  draw
+done
+PIBOX
+chmod +x "$TMP/pibox.sh"
+
+tmux -L "$SOCKET" new-session -d -s pibox -c "$TMP" "exec bash '$TMP/pibox.sh' clear"
+sleep 0.3
+out=$("$SEND" -L "$SOCKET" -t "=pibox" -m "pi fixture four delivered ok" 2>"$TMP/e4"); rc=$?
+if [ "$rc" -eq 0 ] && grep -qF "✓ delivered" <<<"$out"; then
+  ok "delivered: glyphless box-drawn REPL that submits => exit 0 ✓ delivered"
+else
+  no "delivered: glyphless box-drawn REPL that submits => exit 0 ✓ delivered" "rc=$rc out=[$out] err=[$(cat "$TMP/e4")]"
+fi
+
+tmux -L "$SOCKET" new-session -d -s piboxdraft -c "$TMP" "exec bash '$TMP/pibox.sh' keep"
+sleep 0.3
+if out=$("$SEND" -L "$SOCKET" -t "=piboxdraft" -r 1 -m "pi fixture five stuck in the box" 2>"$TMP/e5"); then
+  no "draft: glyphless box-drawn pane holding our tail must NOT report success" "expected exit 2, got 0 (out=[$out])"
+else
+  rc=$?
+  if [ "$rc" -eq 2 ] && grep -qF "unsubmitted draft" "$TMP/e5"; then
+    ok "draft: message left in a glyphless box => exit 2 + 'unsubmitted draft'"
+  else
+    no "draft: message left in a glyphless box => exit 2 + stderr" "rc=$rc err=[$(cat "$TMP/e5")]"
+  fi
+fi
+
 echo "---"
 echo "PASS=$PASS FAIL=$FAIL"
 [ "$FAIL" -eq 0 ]

packages/mosaic/package.json

@@ -25,7 +25,7 @@
     "lint": "eslint src",
     "typecheck": "tsc --noEmit",
     "test": "vitest run --passWithNoTests && pnpm run test:framework-shell",
-    "test:framework-shell": "bash framework/tools/quality/scripts/check-test-enumeration.sh && bash framework/tools/quality/scripts/test-check-test-enumeration.sh && python3 framework/tools/quality/scripts/test-framework-drift-check.py && bash framework/tools/quality/scripts/test-framework-drift-doctor.sh && bash framework/systemd/user/test-fleet-units.sh && python3 src/lease-broker/daemon_deadline_unittest.py && python3 src/lease-broker/normative_fragments_unittest.py && python3 src/lease-broker/promotion_binding_unittest.py && python3 src/lease-broker/promotion_trigger_unittest.py && python3 src/lease-broker/receipt_challenge_unittest.py && python3 src/lease-broker/context_recovery_unittest.py && python3 src/lease-broker/recovery_runtime_unittest.py && python3 src/lease-broker/recovery_b1_adversarial_unittest.py && python3 src/lease-broker/receipt_observer_client_unittest.py && python3 src/lease-broker/invariant_r_unittest.py && python3 src/lease-broker/framework_skill_portability_unittest.py && python3 src/lease-broker/revoke_noop_unittest.py && python3 src/mutator-gate/runtime_tools_unittest.py && python3 src/mutator-gate/runtime_launch_guard_unittest.py && python3 src/mutator-gate/version_coupling_unittest.py && python3 framework/tools/lease-broker/check-runtime-launches.py --root ../.. && bash framework/tools/codex/test-pr-diff-context.sh && bash framework/tools/qa/test-deps-preflight.sh && bash framework/tools/git/test-pr-edit.sh && bash framework/tools/git/test-pr-review-gitea-comment.sh && bash framework/tools/git/test-pr-review-repo-host-override.sh && bash framework/tools/git/test-ci-queue-wait-no-status.sh && bash framework/tools/git/test-ci-queue-wait-branch-absent.sh && bash framework/tools/git/test-ci-queue-wait-tristate.sh && bash framework/tools/git/test-ci-queue-wait-github-checks.sh && bash framework/tools/git/test-pr-merge-queue-branch.sh && bash framework/tools/git/test-pr-merge-fork-ci-status.sh && bash framework/tools/git/test-pr-merge-head-pin.sh && bash framework/tools/git/test-pr-merge-message-field.sh && bash framework/tools/git/test-git-credential-mosaic.sh && bash framework/tools/git/test-gitea-token-identity.sh && bash framework/tools/git/test-explain-diagnostic-status-neutral.sh && bash framework/tools/git/test-detect-platform-outside-repo.sh && bash framework/tools/woodpecker/test-terminal-green-contract.sh && bash framework/tools/_scripts/test-install-ordering-guard.sh && bash framework/tools/_scripts/test-mosaic-init-rce.sh && bash framework/tools/tmux/agent-send.test.sh && bash framework/tools/wake/test-wake-store-ack.sh && bash framework/tools/wake/test-wake-store-enqueue-race.sh && bash framework/tools/wake/test-wake-digest-hmac.sh && bash framework/tools/wake/test-wake-digest-quarantine.sh && bash framework/tools/wake/test-wake-detector.sh && bash framework/tools/wake/test-wake-fn-oracle.sh && bash framework/tools/wake/test-wake-reconcile.sh && bash framework/tools/wake/test-wake-beacon.sh && bash framework/tools/wake/test-wake-preimage.sh && bash framework/tools/wake/test-wake-install.sh && bash framework/tools/glpi/test-list-http-status.sh && bash framework/tools/orchestrator/test-board-roll.sh && bash framework/tools/woodpecker/test-ci-wait-exit-matrix.sh && bash framework/tools/_scripts/test-fleet-transport-check.sh && bash framework/tools/_scripts/test-brain-home-check.sh"
+    "test:framework-shell": "bash framework/tools/quality/scripts/check-test-enumeration.sh && bash framework/tools/quality/scripts/test-check-test-enumeration.sh && python3 framework/tools/quality/scripts/test-framework-drift-check.py && bash framework/tools/quality/scripts/test-framework-drift-doctor.sh && bash framework/systemd/user/test-fleet-units.sh && python3 src/lease-broker/daemon_deadline_unittest.py && python3 src/lease-broker/normative_fragments_unittest.py && python3 src/lease-broker/promotion_binding_unittest.py && python3 src/lease-broker/promotion_trigger_unittest.py && python3 src/lease-broker/receipt_challenge_unittest.py && python3 src/lease-broker/context_recovery_unittest.py && python3 src/lease-broker/recovery_runtime_unittest.py && python3 src/lease-broker/recovery_b1_adversarial_unittest.py && python3 src/lease-broker/receipt_observer_client_unittest.py && python3 src/lease-broker/invariant_r_unittest.py && python3 src/lease-broker/framework_skill_portability_unittest.py && python3 src/lease-broker/revoke_noop_unittest.py && python3 src/mutator-gate/runtime_tools_unittest.py && python3 src/mutator-gate/runtime_launch_guard_unittest.py && python3 src/mutator-gate/version_coupling_unittest.py && python3 framework/tools/lease-broker/check-runtime-launches.py --root ../.. && bash framework/tools/codex/test-pr-diff-context.sh && bash framework/tools/qa/test-deps-preflight.sh && bash framework/tools/git/test-pr-edit.sh && bash framework/tools/git/test-pr-review-gitea-comment.sh && bash framework/tools/git/test-pr-review-repo-host-override.sh && bash framework/tools/git/test-ci-queue-wait-no-status.sh && bash framework/tools/git/test-ci-queue-wait-branch-absent.sh && bash framework/tools/git/test-ci-queue-wait-tristate.sh && bash framework/tools/git/test-ci-queue-wait-github-checks.sh && bash framework/tools/git/test-pr-merge-queue-branch.sh && bash framework/tools/git/test-pr-merge-fork-ci-status.sh && bash framework/tools/git/test-pr-merge-head-pin.sh && bash framework/tools/git/test-pr-merge-message-field.sh && bash framework/tools/git/test-git-credential-mosaic.sh && bash framework/tools/git/test-gitea-token-identity.sh && bash framework/tools/git/test-explain-diagnostic-status-neutral.sh && bash framework/tools/git/test-detect-platform-outside-repo.sh && bash framework/tools/woodpecker/test-terminal-green-contract.sh && bash framework/tools/_scripts/test-install-ordering-guard.sh && bash framework/tools/_scripts/test-mosaic-init-rce.sh && bash framework/tools/tmux/agent-send.test.sh && bash framework/tools/wake/test-wake-store-ack.sh && bash framework/tools/wake/test-wake-store-enqueue-race.sh && bash framework/tools/wake/test-wake-digest-hmac.sh && bash framework/tools/wake/test-wake-digest-quarantine.sh && bash framework/tools/wake/test-wake-detector.sh && bash framework/tools/wake/test-wake-fn-oracle.sh && bash framework/tools/wake/test-wake-reconcile.sh && bash framework/tools/wake/test-wake-beacon.sh && bash framework/tools/wake/test-wake-preimage.sh && bash framework/tools/wake/test-wake-install.sh && bash framework/tools/glpi/test-list-http-status.sh && bash framework/tools/orchestrator/test-board-roll.sh && bash framework/tools/woodpecker/test-ci-wait-exit-matrix.sh && bash framework/tools/_scripts/test-fleet-transport-check.sh && bash framework/tools/_scripts/test-brain-home-check.sh && bash framework/tools/fleet/test-agent-session-broker-preflight.sh"
   },
   "dependencies": {
     "@mosaicstack/brain": "workspace:*",

packages/mosaic/src/commands/fleet-place-unit.spec.ts

@@ -0,0 +1,177 @@
+import { lstat, mkdir, mkdtemp, readFile, rm, symlink, writeFile } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { afterEach, describe, expect, it } from 'vitest';
+
+import { placeUnitFile, resolveLeaseBrokerSocketForPreflight } from './fleet.js';
+
+/**
+ * Unit-placement regression harness for #1292.
+ *
+ * The two measured defects this suite pins:
+ *  1. `systemctl enable <name>` does NOT rewrite an existing by-path
+ *     wants-symlink — so placement must remove stale residue explicitly, and
+ *     acceptance asserts on the RESULTING SYMLINK TARGET, never on the enable
+ *     call's argument (asserting the call cannot see where the link ended up).
+ *  2. Node's copyFile FOLLOWS a by-path symlink at the destination and
+ *     overwrites the SEED template. Acceptance asserts on the SEED's bytes
+ *     AND mtime — unchanged — which is the only check that can redden for
+ *     finding 2. The symlink-target assertion catches finding 1; these are
+ *     different defects with different failure modes.
+ *
+ * Fixtures are entirely inside tmpdirs (source template, active systemd dir,
+ * wants dir) — no real host paths are touched by this suite.
+ */
+
+describe('placeUnitFile (#1292 unit placement)', () => {
+  const cleanup: string[] = [];
+  afterEach(async () => {
+    while (cleanup.length > 0) {
+      await rm(cleanup.pop()!, { recursive: true, force: true });
+    }
+  });
+
+  async function fixture() {
+    const root = await mkdtemp(join(tmpdir(), 'place-unit-'));
+    cleanup.push(root);
+    const seedDir = join(root, 'seed');
+    const activeDir = join(root, 'active');
+    await mkdir(seedDir, { recursive: true });
+    await mkdir(activeDir, { recursive: true });
+    const seedTemplate = join(seedDir, 'unit-under-test.service');
+    await writeFile(
+      seedTemplate,
+      '[Unit]\nDescription=seed template\n[Service]\nType=oneshot\nExecStart=/bin/true\n[Install]\nWantedBy=default.target\n',
+    );
+    const activeSource = join(root, 'active-source.service');
+    await writeFile(
+      activeSource,
+      '[Unit]\nDescription=active copy v2\n[Service]\nType=oneshot\nExecStart=/bin/true\n[Install]\nWantedBy=default.target\n',
+    );
+    return { root, seedDir, activeDir, seedTemplate, activeSource };
+  }
+
+  it('places a regular file on a clean host (negative control: no residue anywhere)', async () => {
+    const f = await fixture();
+    const result = await placeUnitFile(f.activeSource, f.activeDir, 'unit-under-test.service');
+    expect(result.unlinkedDestinationSymlink).toBe(false);
+    expect(result.removedStaleWantsSymlink).toBe(false);
+    const info = await lstat(join(f.activeDir, 'unit-under-test.service'));
+    expect(info.isSymbolicLink()).toBe(false);
+    expect(await readFile(join(f.activeDir, 'unit-under-test.service'), 'utf8')).toContain(
+      'active copy v2',
+    );
+    // Seed untouched by construction — but assert it, so the clean-host case
+    // cannot silently regress into seed-mutation.
+    expect(await readFile(f.seedTemplate, 'utf8')).toContain('seed template');
+  });
+
+  it('by-path residue: unlinks destination symlink, places the file, seed bytes AND mtime unchanged (finding 2)', async () => {
+    const f = await fixture();
+    const seedBefore = await readFile(f.seedTemplate, 'utf8');
+    const mtimeBefore = (await lstat(f.seedTemplate)).mtimeMs;
+    // The fomo-lin convention: by-path enable left a symlink AT the unit name
+    // pointing at the seed template, plus a wants-symlink doing the same.
+    await symlink(f.seedTemplate, join(f.activeDir, 'unit-under-test.service'));
+    const wantsDir = join(f.activeDir, 'default.target.wants');
+    await mkdir(wantsDir, { recursive: true });
+    await symlink(f.seedTemplate, join(wantsDir, 'unit-under-test.service'));
+
+    const result = await placeUnitFile(f.activeSource, f.activeDir, 'unit-under-test.service');
+    expect(result.unlinkedDestinationSymlink).toBe(true);
+    expect(result.removedStaleWantsSymlink).toBe(true);
+
+    // FINDING 2's check: the seed is byte-identical and its mtime did not move.
+    expect(await readFile(f.seedTemplate, 'utf8')).toBe(seedBefore);
+    expect((await lstat(f.seedTemplate)).mtimeMs).toBe(mtimeBefore);
+
+    // The destination is now a regular file carrying the ACTIVE content.
+    const destInfo = await lstat(join(f.activeDir, 'unit-under-test.service'));
+    expect(destInfo.isSymbolicLink()).toBe(false);
+    expect(await readFile(join(f.activeDir, 'unit-under-test.service'), 'utf8')).toContain(
+      'active copy v2',
+    );
+  });
+
+  it('by-path residue: no wants-symlink remains pointing at the seed (finding 1 residue cleared)', async () => {
+    const f = await fixture();
+    await symlink(f.seedTemplate, join(f.activeDir, 'unit-under-test.service'));
+    const wantsDir = join(f.activeDir, 'default.target.wants');
+    await mkdir(wantsDir, { recursive: true });
+    await symlink(f.seedTemplate, join(wantsDir, 'unit-under-test.service'));
+
+    await placeUnitFile(f.activeSource, f.activeDir, 'unit-under-test.service');
+
+    // After placement the stale wants link is GONE (enable-by-name recreates
+    // it correctly). A link still present must not point at the seed.
+    try {
+      const link = await lstat(join(wantsDir, 'unit-under-test.service'));
+      if (link.isSymbolicLink()) {
+        const target = await readFile(join(wantsDir, 'unit-under-test.service'), 'utf8').catch(
+          async () => '',
+        );
+        expect(target).not.toContain('seed template');
+      }
+    } catch {
+      // absent wants link — the expected post-placement state
+    }
+  });
+
+  it('idempotence: second placement on a reconciled host is a no-op producing the identical final state', async () => {
+    const f = await fixture();
+    // Reconciled starting state: regular file at the name, wants link to the active copy.
+    await writeFile(
+      join(f.activeDir, 'unit-under-test.service'),
+      await readFile(f.activeSource, 'utf8'),
+    );
+    const wantsDir = join(f.activeDir, 'default.target.wants');
+    await mkdir(wantsDir, { recursive: true });
+    await symlink(
+      join(f.activeDir, 'unit-under-test.service'),
+      join(wantsDir, 'unit-under-test.service'),
+    );
+    const before = await readFile(join(f.activeDir, 'unit-under-test.service'), 'utf8');
+
+    const result = await placeUnitFile(f.activeSource, f.activeDir, 'unit-under-test.service');
+    // No destructive step fired: no unlink, no wants removal.
+    expect(result.unlinkedDestinationSymlink).toBe(false);
+    expect(result.removedStaleWantsSymlink).toBe(false);
+    // Identical final state.
+    expect(await readFile(join(f.activeDir, 'unit-under-test.service'), 'utf8')).toBe(before);
+    const link = await lstat(join(wantsDir, 'unit-under-test.service'));
+    expect(link.isSymbolicLink()).toBe(true);
+  });
+
+  it('double install on by-path residue converges to the identical reconciled state', async () => {
+    const f = await fixture();
+    await symlink(f.seedTemplate, join(f.activeDir, 'unit-under-test.service'));
+    const wantsDir = join(f.activeDir, 'default.target.wants');
+    await mkdir(wantsDir, { recursive: true });
+    await symlink(f.seedTemplate, join(wantsDir, 'unit-under-test.service'));
+
+    await placeUnitFile(f.activeSource, f.activeDir, 'unit-under-test.service');
+    const first = await readFile(join(f.activeDir, 'unit-under-test.service'), 'utf8');
+    const secondRun = await placeUnitFile(f.activeSource, f.activeDir, 'unit-under-test.service');
+    const second = await readFile(join(f.activeDir, 'unit-under-test.service'), 'utf8');
+    expect(secondRun.unlinkedDestinationSymlink).toBe(false);
+    expect(second).toBe(first);
+  });
+});
+
+describe('resolveLeaseBrokerSocketForPreflight (#1292 preflight resolution)', () => {
+  it('explicit MOSAIC_LEASE_BROKER_SOCKET wins', () => {
+    expect(
+      resolveLeaseBrokerSocketForPreflight({ MOSAIC_LEASE_BROKER_SOCKET: '/custom/sock' }, 1000),
+    ).toBe('/custom/sock');
+  });
+  it('XDG_RUNTIME_DIR next', () => {
+    expect(resolveLeaseBrokerSocketForPreflight({ XDG_RUNTIME_DIR: '/run/user/1001' }, 1000)).toBe(
+      '/run/user/1001/mosaic-lease/broker.sock',
+    );
+  });
+  it('falls back to /run/user/<uid>', () => {
+    expect(resolveLeaseBrokerSocketForPreflight({}, 1002)).toBe(
+      '/run/user/1002/mosaic-lease/broker.sock',
+    );
+  });
+});

packages/mosaic/src/commands/fleet-reconciler-command.spec.ts

@@ -73,6 +73,10 @@ function program(
     runner,
     reconcileDeps: {
       homeDirectory: '/home/mosaic',
+      // Deterministic broker presence: without a seam the reconciler probes the
+      // REAL host socket (#1297 F3), making every CLI start test answer the
+      // host's broker state instead of its own property.
+      checkBrokerSocket: async () => true,
       readHolderIdentity: async () => '11111111-1111-4111-8111-111111111111',
       validateRoster: async () => undefined,
       prepareProjections: async () => [{ agentName: 'coder0' }],

packages/mosaic/src/commands/fleet.spec.ts

@@ -1,11 +1,24 @@
-import { chmod, lstat, mkdir, mkdtemp, readFile, rm, stat, writeFile } from 'node:fs/promises';
+import {
+  chmod,
+  lstat,
+  mkdir,
+  mkdtemp,
+  readFile,
+  readlink,
+  rm,
+  stat,
+  symlink,
+  writeFile,
+} from 'node:fs/promises';
 import { tmpdir } from 'node:os';
 import { dirname, join, resolve } from 'node:path';
+import { createServer } from 'node:net';
 import { Command } from 'commander';
 import { afterEach, describe, expect, it, vi } from 'vitest';
 import {
   acquireRestartLock,
   addAgentToRoster,
+  brokerSocketPresent,
   buildAgentSendCommand,
   buildAgentWatchAttachCommand,
   buildAgentWatchCommand,
@@ -42,6 +55,7 @@ import {
   parseSystemdShow,
   parseTmuxListPanes,
   parseTmuxListSessions,
+  placeUnitFile,
   registerFleetCommand,
   removeAgentFromRoster,
   resolveFleetPaths,
@@ -50,6 +64,7 @@ import {
   RESTART_LOCK_STALE_MS,
   RUNTIME_ACCEPTABLE_COMMANDS,
   serializeRosterToYaml,
+  UnitPlacementError,
   VERIFY_DEFAULT_TIMEOUT_MS,
   VERIFY_POLL_INTERVAL_MS,
   type AgentPsRow,
@@ -836,13 +851,25 @@ describe('fleet command construction', () => {
     };
     const program = new Command();
     program.exitOverride();
-    registerFleetCommand(program, { runner, mosaicHome: home });
+    // #1292: inject a present broker socket so the preflight passes and this
+    // spec keeps testing its ORIGINAL property (holder-before-agent ordering).
+    // The preflight's own refusal behavior has dedicated specs below.
+    registerFleetCommand(program, {
+      runner,
+      mosaicHome: home,
+      checkBrokerSocket: async () => true,
+    });
 
     try {
       await program.parseAsync(['node', 'mosaic', 'fleet', 'start']);
       await program.parseAsync(['node', 'mosaic', 'fleet', 'stop']);
 
       expect(calls).toEqual([
+        // #1292: fleet start enables + starts the broker FIRST (enable is
+        // idempotent; the unit exists after install), re-checking the socket
+        // before any holder/agent lifecycle effect.
+        ['systemctl', '--user', 'enable', 'mosaic-lease-broker.service'],
+        ['systemctl', '--user', 'start', 'mosaic-lease-broker.service'],
         ['systemctl', '--user', 'start', 'mosaic-tmux-holder.service'],
         ['systemctl', '--user', 'start', '[email protected]'],
         ['systemctl', '--user', 'stop', '[email protected]'],
@@ -853,6 +880,92 @@ describe('fleet command construction', () => {
     }
   });
 
+  it('fleet start refuses with a named error when the broker socket does not appear (#1292)', async () => {
+    const home = await tempDir();
+    const rosterPath = join(home, 'fleet', 'roster.yaml');
+    await mkdir(join(home, 'fleet'), { recursive: true });
+    await writeFile(
+      rosterPath,
+      ['version: 1', 'transport: tmux', 'agents:', '  - name: coder0', '    runtime: codex'].join(
+        '\n',
+      ),
+    );
+    const calls: string[][] = [];
+    const runner: CommandRunner = async (command, args) => {
+      calls.push([command, ...args]);
+      return { stdout: '', stderr: '', exitCode: 0 };
+    };
+    const program = new Command();
+    program.exitOverride();
+    const errors: string[] = [];
+    const origError = console.error;
+    console.error = (...args: unknown[]) => {
+      errors.push(args.join(' '));
+    };
+    registerFleetCommand(program, {
+      runner,
+      mosaicHome: home,
+      checkBrokerSocket: async () => false,
+    });
+    try {
+      await program.parseAsync(['node', 'mosaic', 'fleet', 'start']);
+      // Refused: no holder/agent starts were issued after the broker attempt.
+      expect(calls).toEqual([
+        ['systemctl', '--user', 'enable', 'mosaic-lease-broker.service'],
+        ['systemctl', '--user', 'start', 'mosaic-lease-broker.service'],
+      ]);
+      expect(errors.join('\n')).toContain('broker-absent');
+      expect(errors.join('\n')).toContain('mosaic fleet install');
+    } finally {
+      console.error = origError;
+      await rm(home, { recursive: true, force: true });
+    }
+  });
+
+  it('fleet start re-probes the broker on the SECOND invocation — no ActiveState trust (#1292 sticky half)', async () => {
+    const home = await tempDir();
+    const rosterPath = join(home, 'fleet', 'roster.yaml');
+    await mkdir(join(home, 'fleet'), { recursive: true });
+    await writeFile(
+      rosterPath,
+      ['version: 1', 'transport: tmux', 'agents:', '  - name: coder0', '    runtime: codex'].join(
+        '\n',
+      ),
+    );
+    const calls: string[][] = [];
+    const runner: CommandRunner = async (command, args) => {
+      calls.push([command, ...args]);
+      return { stdout: '', stderr: '', exitCode: 0 };
+    };
+    const program = new Command();
+    program.exitOverride();
+    // Broker socket NEVER appears — the second start must refuse exactly like
+    // the first; RemainAfterExit-style stale unit state changes nothing
+    // because the check is the socket, not systemctl.
+    registerFleetCommand(program, {
+      runner,
+      mosaicHome: home,
+      checkBrokerSocket: async () => false,
+    });
+    const errors: string[] = [];
+    const origError = console.error;
+    console.error = (...args: unknown[]) => {
+      errors.push(args.join(' '));
+    };
+    try {
+      await program.parseAsync(['node', 'mosaic', 'fleet', 'start']);
+      await program.parseAsync(['node', 'mosaic', 'fleet', 'start']);
+      // Two invocations, each refusing after its own broker attempt:
+      expect(
+        calls.filter((c) => c.join(' ') === 'systemctl --user start [email protected]'),
+      ).toHaveLength(0);
+      expect(errors.filter((e) => e.includes('broker-absent')).length).toBeGreaterThanOrEqual(2);
+    } finally {
+      console.error = origError;
+      await rm(home, { recursive: true, force: true });
+    }
+  });
+
   it('waits for an in-flight restart to clear before relaunching (re-entry guard)', async () => {
     const home = await tempDir();
     const rosterPath = join(home, 'fleet', 'roster.yaml');
@@ -2066,8 +2179,19 @@ describe('fleet install — auto-enable units for boot-survival', () => {
 
     await enableFleetUnits(runner, minimalRoster, {});
 
+    expect(calls).toContainEqual(['systemctl', '--user', 'enable', 'mosaic-lease-broker.service']);
     expect(calls).toContainEqual(['systemctl', '--user', 'enable', 'mosaic-tmux-holder.service']);
     expect(calls).toContainEqual(['systemctl', '--user', 'enable', '[email protected]']);
+    // The broker must be enabled BEFORE the holder and agents: a start of any
+    // gated runtime without the broker is exactly the #1292 4-second death.
+    const brokerIndex = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user enable mosaic-lease-broker.service',
+    );
+    const holderIndex = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user enable mosaic-tmux-holder.service',
+    );
+    expect(brokerIndex).toBeGreaterThanOrEqual(0);
+    expect(brokerIndex).toBeLessThan(holderIndex);
   });
 
   it('install still succeeds when systemctl enable returns non-zero (non-fatal)', async () => {
@@ -4362,3 +4486,70 @@ describe('fleet ps — heartbeat path resolution', () => {
     );
   });
 });
+
+describe('#1297 review: the real broker probe, exercised without any seam', () => {
+  it('brokerSocketPresent answers a REAL unix socket via stat().isSocket() (access(S_IFSOCK) threw ERR_OUT_OF_RANGE)', async () => {
+    const dir = await tempDir();
+    const sockPath = join(dir, 'broker.sock');
+    const server = createServer();
+    await new Promise<void>((resolve) => {
+      server.listen(sockPath, resolve);
+    });
+    try {
+      // A live unix socket answers true through the REAL probe — no seam.
+      expect(await brokerSocketPresent({}, { MOSAIC_LEASE_BROKER_SOCKET: sockPath })).toBe(true);
+      // Discrimination is by file type: a regular file that EXISTS is not a
+      // socket. The old implementation could not reach either verdict — it
+      // threw ERR_OUT_OF_RANGE (node >= 24) and the catch answered false.
+      const notASocket = join(dir, 'not-a-sock');
+      await writeFile(notASocket, 'x');
+      expect(await brokerSocketPresent({}, { MOSAIC_LEASE_BROKER_SOCKET: notASocket })).toBe(false);
+      // Absent path: false, not a throw.
+      expect(
+        await brokerSocketPresent({}, { MOSAIC_LEASE_BROKER_SOCKET: join(dir, 'gone.sock') }),
+      ).toBe(false);
+    } finally {
+      await new Promise<void>((resolve) => {
+        server.close(() => resolve());
+      });
+    }
+    await rm(dir, { recursive: true, force: true });
+  });
+
+  // EACCES-based unlink failure requires a non-root uid: root bypasses
+  // directory mode bits (CAP_DAC_OVERRIDE), so the abort path cannot be
+  // triggered this way under CI's root runner. Skipped there, exercised on
+  // every non-root dev host.
+  const itUnlessRoot =
+    typeof process.getuid === 'function' && process.getuid() === 0 ? it.skip : it;
+  itUnlessRoot(
+    'placeUnitFile aborts with UnitPlacementError when unlink fails — never copies through a live symlink',
+    async () => {
+      const dir = await tempDir();
+      const unitDir = join(dir, 'systemd', 'user');
+      await mkdir(unitDir, { recursive: true });
+      // By-path residue: destination is a symlink pointing somewhere else.
+      const residueTarget = join(dir, 'residue-target');
+      await writeFile(residueTarget, 'RESIDUE-BYTES');
+      const destination = join(unitDir, 'x.service');
+      await symlink(residueTarget, destination);
+      const source = join(dir, 'seed.service');
+      await writeFile(source, 'UNIT-BYTES');
+      // Read-only unit dir: unlink now fails EACCES (test runs as the owner,
+      // not root, so mode bits are enforced).
+      await chmod(unitDir, 0o500);
+      try {
+        await expect(placeUnitFile(source, unitDir, 'x.service')).rejects.toThrow(
+          UnitPlacementError,
+        );
+      } finally {
+        await chmod(unitDir, 0o700);
+      }
+      // The copy-through never happened: residue bytes intact, destination
+      // still the symlink (abort, not overwrite-through).
+      expect(await readFile(residueTarget, 'utf8')).toBe('RESIDUE-BYTES');
+      expect(await readlink(destination)).toBe(residueTarget);
+      await rm(dir, { recursive: true, force: true });
+    },
+  );
+});

packages/mosaic/src/commands/fleet.ts

@@ -1,11 +1,13 @@
-import { constants } from 'node:fs';
+import { constants, type Stats } from 'node:fs';
 import {
   access,
   chmod,
   copyFile,
+  lstat,
   mkdir,
   open,
   readFile,
+  readlink,
   stat,
   unlink,
   writeFile,
@@ -90,6 +92,8 @@ export type SleepFn = (ms: number) => Promise<void>;
 
 export interface FleetCommandDeps {
   runner?: CommandRunner;
+  /** Test seam for the #1292 fleet-start broker preflight (socket presence). */
+  checkBrokerSocket?: (path: string) => Promise<boolean> | boolean;
   /** Injectable interactive runner for commands needing inherited TTY (e.g., `tmux attach`). */
   interactiveRunner?: InteractiveRunner;
   /**
@@ -815,6 +819,135 @@ export function buildSystemdEnableCommand(unit: string): string[] {
   return ['systemctl', '--user', 'enable', unit];
 }
 
+/**
+ * Place a unit file into the ACTIVE systemd user directory, never through a
+ * symlink (#1292, measured 2026-08-17).
+ *
+ * ⚠ SET-INDEPENDENCE (fomo-lin, 2026-08-17): the set of unit names carrying
+ * by-path residue and the set of unit names this install copies are
+ * INDEPENDENT. Until 0.0.50 they were disjoint only by accident of which
+ * units the install happened to name — fomo-lin survived copy-through solely
+ * because its one by-path symlink (the broker) was the one unit the install
+ * did NOT copy. Adding the broker to the copy set made the intersection
+ * non-empty on the first run. Whoever adds a fifth unit to the placement
+ * list inherits this helper and its unlink step; do not place units with a
+ * bare copyFile.
+ *
+ * A host provisioned by the enable-by-path convention carries a symlink AT
+ * the unit-name path in ~/.config/systemd/user/ pointing at the shipped
+ * template under ~/.config/mosaic/systemd/user/. Node's copyFile FOLLOWS
+ * that link and overwrites the SEED template instead of placing the active
+ * unit (verified with fs.copyFile on a throwaway systemd user instance) —
+ * silent, rc=0, and it mutates the directory every later reseed reads from.
+ * The same measurement showed `systemctl enable <name>` does NOT rewrite an
+ * existing by-path wants-symlink, so reconciliation must be explicit.
+ *
+ * Placement therefore: if the destination is a symlink, unlink it first
+ * (unlink → copy — copy-then-unlink would mutate the seed and then destroy
+ * the evidence that it did); then copy. Also removes a stale
+ * `default.target.wants/<name>` symlink that points outside the active
+ * directory (readlink — NOT readFile, which follows the link and returns the
+ * target's CONTENT), so the subsequent enable-by-name recreates it against
+ * the active copy. Idempotent: on a clean or already-reconciled destination
+ * every step is a no-op (the copy rewrites identical bytes).
+ *
+ * Returns what was done, for assertions and install reporting.
+ */
+export interface PlaceUnitResult {
+  readonly unit: string;
+  readonly destination: string;
+  /** A symlink at the unit-name path was unlinked (by-path residue). */
+  readonly unlinkedDestinationSymlink: boolean;
+  /** A stale wants-symlink pointing outside the active dir was removed. */
+  readonly removedStaleWantsSymlink: boolean;
+}
+
+/**
+ * placeUnitFile failed. Thrown BEFORE any copy: no destination bytes were
+ * written, so a residue target cannot have been clobbered by a copy-through
+ * (#1297 review F2).
+ */
+export class UnitPlacementError extends Error {
+  constructor(
+    readonly unit: string,
+    message: string,
+  ) {
+    super(message);
+    this.name = UnitPlacementError.name;
+  }
+}
+
+function isErrnoException(error: unknown): error is NodeJS.ErrnoException {
+  return error instanceof Error && 'code' in error && typeof error.code === 'string';
+}
+
+export async function placeUnitFile(
+  source: string,
+  systemdUserDir: string,
+  unit: string,
+): Promise<PlaceUnitResult> {
+  const destination = join(systemdUserDir, unit);
+  let unlinkedDestinationSymlink = false;
+  // Destination-absent and unlink-FAILED are different outcomes and must not
+  // share a catch (#1297 review F2): a swallowed unlink error used to fall
+  // through to copyFile through the still-live symlink, silently reintroducing
+  // the exact copy-through this helper exists to prevent.
+  let destinationInfo: Stats | undefined;
+  try {
+    destinationInfo = await lstat(destination);
+  } catch (error) {
+    if (!isErrnoException(error) || error.code !== 'ENOENT') {
+      throw new UnitPlacementError(
+        unit,
+        `cannot inspect destination ${destination}: ${error instanceof Error ? error.message : String(error)}`,
+      );
+    }
+    // ENOENT: absent destination — nothing to unlink, copy below is safe.
+  }
+  if (destinationInfo?.isSymbolicLink()) {
+    try {
+      await unlink(destination);
+    } catch (error) {
+      // Abort BEFORE the copy: proceeding would run copyFile through the
+      // still-live symlink and overwrite the residue target's bytes.
+      throw new UnitPlacementError(
+        unit,
+        `cannot unlink destination symlink ${destination}: ${error instanceof Error ? error.message : String(error)}`,
+      );
+    }
+    unlinkedDestinationSymlink = true;
+  }
+  await copyFile(source, destination);
+
+  let removedStaleWantsSymlink = false;
+  const wantsLink = join(systemdUserDir, 'default.target.wants', unit);
+  try {
+    const wantsInfo = await lstat(wantsLink);
+    if (wantsInfo.isSymbolicLink()) {
+      // readlink — NOT readFile: readFile FOLLOWS the link and returns the
+      // target file's CONTENT, which is not the question being asked.
+      let target: string | undefined;
+      try {
+        target = await readlink(wantsLink);
+      } catch {
+        target = undefined;
+      }
+      // Normalize (systemctl writes absolute targets; a relative one resolves
+      // against the wants dir). A wants-symlink pointing anywhere other than
+      // the active copy (the by-path convention points at the seed template)
+      // survives enable-by-name unchanged — remove it so enable recreates it.
+      if (target !== undefined && resolve(dirname(wantsLink), target) !== destination) {
+        await unlink(wantsLink);
+        removedStaleWantsSymlink = true;
+      }
+    }
+  } catch {
+    // absent wants link — nothing to reconcile
+  }
+
+  return { unit, destination, unlinkedDestinationSymlink, removedStaleWantsSymlink };
+}
+
 /**
  * Returns the systemctl --user disable command for a given unit.
  * Used by `fleet remove` so a removed agent's enabled unit cannot resurrect on
@@ -849,6 +982,22 @@ export async function enableFleetUnits(
     let succeeded = 0;
     let failed = 0;
 
+    // The lease broker ships with the fleet and every gated runtime needs it
+    // (#1292): seats die at lease registration without it, and no documented
+    // path ever enabled it. Enabled first — alongside the holder — and the
+    // unit must have been placed by installFleet's placeUnitFile step.
+    const brokerResult = await runner(
+      ...splitCommand(buildSystemdEnableCommand('mosaic-lease-broker.service')),
+    );
+    if (brokerResult.exitCode === 0) {
+      succeeded++;
+    } else {
+      failed++;
+      process.stderr.write(
+        `Warning: could not enable mosaic-lease-broker.service: ${brokerResult.stderr || brokerResult.stdout || 'non-zero exit'}\n`,
+      );
+    }
+
     const holderResult = await runner(
       ...splitCommand(buildSystemdEnableCommand('mosaic-tmux-holder.service')),
     );
@@ -1545,7 +1694,7 @@ export function registerFleetCommand(program: Command, deps: FleetCommandDeps =
     .description('Install local fleet tools and user systemd units')
     .option('--no-enable', 'Skip enabling units for boot-survival')
     .action(async (opts: { enable?: boolean }) => {
-      await installFleet(cmd, frameworkRoot);
+      await installFleet(cmd, frameworkRoot, runner);
       // Unit enablement needs agent names only, so it reads either version.
       const roster = await loadRosterReadModel(cmd);
       await enableFleetUnits(runner, roster, opts);
@@ -1556,7 +1705,7 @@ export function registerFleetCommand(program: Command, deps: FleetCommandDeps =
     .description('Install local fleet tools and user systemd units')
     .option('--no-enable', 'Skip enabling units for boot-survival')
     .action(async (opts: { enable?: boolean }) => {
-      await installFleet(cmd, frameworkRoot);
+      await installFleet(cmd, frameworkRoot, runner);
       // Unit enablement needs agent names only, so it reads either version.
       const roster = await loadRosterReadModel(cmd);
       await enableFleetUnits(runner, roster, opts);
@@ -1609,6 +1758,37 @@ export function registerFleetCommand(program: Command, deps: FleetCommandDeps =
             );
             return;
           }
+          if (action === 'start') {
+            // Broker preflight (#1292), re-probed on EVERY invocation: a
+            // gated runtime started without a live lease broker dies ~4s in
+            // while the unit reports active (RemainAfterExit) — enabling +
+            // starting here and then RE-CHECKING the socket refuses loudly
+            // instead of reporting rc0 over a doomed start. This is the
+            // second-start check as much as the first: it never trusts unit
+            // ActiveState.
+            await runChecked(runner, [
+              'systemctl',
+              '--user',
+              'enable',
+              'mosaic-lease-broker.service',
+            ]);
+            await runChecked(runner, [
+              'systemctl',
+              '--user',
+              'start',
+              'mosaic-lease-broker.service',
+            ]);
+            if (!(await brokerSocketPresent(deps))) {
+              console.error(
+                '[fleet] broker-absent: lease broker socket did not appear after enable+start (#1292).',
+              );
+              console.error(
+                '[fleet] remedy: mosaic fleet install (it reconciles either enable convention)',
+              );
+              process.exitCode = 1;
+              return;
+            }
+          }
           if (action === 'restart') {
             // Serialize the holder+agents teardown/relaunch behind the restart lock
             // so a re-entrant restart waits for clean shutdown before relaunching,
@@ -2367,7 +2547,11 @@ export function registerFleetAgentCommands(
     });
 }
 
-async function installFleet(cmd: Command, frameworkRoot: string): Promise<void> {
+async function installFleet(
+  cmd: Command,
+  frameworkRoot: string,
+  runner: CommandRunner,
+): Promise<void> {
   const activePaths = resolveFleetPaths(cmd.opts<{ mosaicHome: string }>().mosaicHome);
   assertDefaultMosaicHomeForSystemd(activePaths.mosaicHome);
   // Read model first: every file this function places is roster-independent, and
@@ -2419,18 +2603,40 @@ async function installFleet(cmd: Command, frameworkRoot: string): Promise<void>
   for (const toolPath of executableToolPaths) {
     await chmod(toolPath, 0o755);
   }
-  await copyFile(
+  // Unit placement (#1292): every unit goes through placeUnitFile — never a
-    join(frameworkRoot, 'systemd', 'user', 'mosaic-tmux-holder.service'),
+  // bare copyFile — so a by-path-enable symlink at the destination is
-    join(activePaths.systemdUserDir, 'mosaic-tmux-holder.service'),
+  // unlinked rather than written through (copy-through would silently
+  // overwrite the SEED template, measured 2026-08-17). The lease broker unit
+  // is placed here too: previously the install named three units and omitted
+  // the broker entirely, which is why no documented path ever enabled it.
+  const placedUnits = await Promise.all(
+    [
+      'mosaic-tmux-holder.service',
+      '[email protected]',
+      '[email protected]',
+      'mosaic-lease-broker.service',
+    ].map((unit) =>
+      placeUnitFile(join(frameworkRoot, 'systemd', 'user', unit), activePaths.systemdUserDir, unit),
+    ),
   );
-  await copyFile(
+  const reconciled = placedUnits.filter(
-    join(frameworkRoot, 'systemd', 'user', '[email protected]'),
+    (result) => result.unlinkedDestinationSymlink || result.removedStaleWantsSymlink,
-    join(activePaths.systemdUserDir, '[email protected]'),
-  );
-  await copyFile(
-    join(frameworkRoot, 'systemd', 'user', '[email protected]'),
-    join(activePaths.systemdUserDir, '[email protected]'),
   );
+  if (reconciled.length > 0) {
+    console.log(
+      `Reconciled ${reconciled.length} unit placement(s) from by-path enable residue: ${reconciled.map((r) => r.unit).join(', ')}`,
+    );
+  }
+  // systemd will not see a replaced unit file without a reload; do it once
+  // after all placements, before any enable call below. runCommand never
+  // rejects (it resolves exitCode 127 on spawn error), so a plain await with
+  // an exitCode check matches the rest of this file's systemctl handling.
+  const reloadResult = await runner(...splitCommand(['systemctl', '--user', 'daemon-reload']));
+  if (reloadResult.exitCode !== 0) {
+    process.stderr.write(
+      `Warning: systemctl --user daemon-reload after unit placement failed (non-systemd host?): ${reloadResult.stderr || reloadResult.stdout || 'non-zero exit'}\n`,
+    );
+  }
 
   // On roster v2 the reconciler owns the generated env: `apply` writes it and
   // `regen` rebuilds it, both from projectRosterV2AgentGeneratedEnv. Writing it
@@ -2627,6 +2833,45 @@ function splitCommand(command: string[]): [string, string[]] {
   return [bin, args];
 }
 
+/**
+ * Lease-broker socket presence for the fleet-start preflight (#1292).
+ * Resolution precedence matches launch.ts's defaultLeaseBrokerSocket and
+ * start-agent-session.sh's broker_socket_path: explicit
+ * MOSAIC_LEASE_BROKER_SOCKET, else $XDG_RUNTIME_DIR/mosaic-lease/broker.sock,
+ * else /run/user/<uid>/mosaic-lease/broker.sock. Pure filesystem check — this
+ * deliberately does NOT consult systemd state: a unit can be active
+ * (RemainAfterExit) with no live socket, and the socket is the thing the
+ * gated runtime connects to. Injectable via deps for tests.
+ */
+export function resolveLeaseBrokerSocketForPreflight(
+  env: NodeJS.ProcessEnv = process.env,
+  uid: number = typeof process.getuid === 'function' ? process.getuid() : 0,
+): string {
+  if (env['MOSAIC_LEASE_BROKER_SOCKET']) return env['MOSAIC_LEASE_BROKER_SOCKET'];
+  const runtimeDir = env['XDG_RUNTIME_DIR'] ?? `/run/user/${uid}`;
+  return join(runtimeDir, 'mosaic-lease', 'broker.sock');
+}
+
+export async function brokerSocketPresent(
+  deps: FleetCommandDeps,
+  env: NodeJS.ProcessEnv = process.env,
+): Promise<boolean> {
+  const check = deps.checkBrokerSocket;
+  const socketPath = resolveLeaseBrokerSocketForPreflight(env);
+  if (check) return check(socketPath);
+  // S_IFSOCK (0xC000) is a file-TYPE constant, not an access() mode (0-7):
+  // access(path, S_IFSOCK) throws ERR_OUT_OF_RANGE on node >= 24 (measured on
+  // v24.18.0, #1297 review F1) and cannot succeed on any node — the old catch
+  // swallowed the throw, so this probe could NEVER return true and every
+  // un-seamed call reported the broker absent. stat() + isSocket() is the real
+  // check and matches the bash side's [ -S ].
+  try {
+    return (await stat(socketPath)).isSocket();
+  } catch {
+    return false;
+  }
+}
+
 /** All supported fleet profile names. */
 export type FleetProfile =
   | 'general'

packages/mosaic/src/commands/lease-doctor-check.ts

@@ -205,6 +205,12 @@ export async function runLeaseEnforcementDoctorCheck(
     message:
       `Lease-enforcement hooks (${matchedMarkers.join(', ')}) are wired in ~/.claude/settings.json, but ${reasons.join(' and ')}. ` +
       'Every gated tool call will fail closed and BRICK this agent (see #869). ' +
-      'Remediate by activating the lease-broker supervisor (systemd unit + socket) or by removing the enforcement hooks from ~/.claude/settings.json.',
+      // #1292: one remedy, correct under BOTH enable conventions (by-path on
+      // the seed template, and copy-then-enable in the active dir). Written
+      // from the 2026-08-17 symlink measurement: `systemctl enable` by name
+      // does NOT rewrite an existing by-path wants-symlink, so teaching a
+      // manual systemctl line here could leave a host with two competing
+      // wants links. fleet install reconciles either shape.
+      'Remedy: run `mosaic fleet install` (it reconciles either enable convention), or remove the enforcement hooks from ~/.claude/settings.json.',
   };
 }

packages/mosaic/src/fleet/fleet-reconciler.acceptance.spec.ts

@@ -169,6 +169,16 @@ function reconcileDeps(host: FakeLifecycleHost): FleetReconcileDeps {
     applyProjection: async () => undefined,
     readRoster: async () => host.roster,
     acquireMutationLock: async () => async () => undefined,
+    // Hermetic broker observation (#1297 F3): without this, the plan probes
+    // the REAL host filesystem, so the "stable JSON" fixtures answered true
+    // on any machine with a live lease broker and false elsewhere. Pointing
+    // both paths at fixtures that do not exist pins socketPresent:false and
+    // unitInstalled:false on every host, which is what these fixtures assert.
+    brokerSocketEnv: {
+      MOSAIC_LEASE_BROKER_SOCKET: '/nonexistent/mosaic-lease/broker.sock',
+      XDG_CONFIG_HOME: '/nonexistent/mosaic-config',
+      XDG_RUNTIME_DIR: '/nonexistent/run',
+    },
   };
 }
 
@@ -459,6 +469,7 @@ describe('FCM-M3-002 reconciler lifecycle acceptance', (): void => {
       plan: {
         generation: 7,
         holder: 'owned',
+        broker: { unitInstalled: false, socketPresent: false },
         agents: [
           {
             name: 'coder0',

packages/mosaic/src/fleet/fleet-reconciler.spec.ts

@@ -1,6 +1,7 @@
 import { chmod, mkdir, mkdtemp, readFile, rm, symlink, writeFile } from 'node:fs/promises';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
+import { createServer } from 'node:net';
 import { afterEach, describe, expect, it } from 'vitest';
 import {
   acquirePrivateReconcileLock,
@@ -92,6 +93,179 @@ async function run(command: FleetReconcileCommand, overrides: Partial<FleetRecon
 }
 
 describe('fleet roster-owned reconciler', (): void => {
+  // ── #1292: broker as first-class plan member + broker-first start ordering ──
+
+  it('reports broker unit and socket state in the plan (socket is the signal, not unit state)', async (): Promise<void> => {
+    const result = await run('status', {
+      statPath: async () => true,
+      checkBrokerSocket: async () => true,
+    });
+    expect(result.plan.broker).toEqual({ unitInstalled: true, socketPresent: true });
+  });
+
+  it('reports a dead broker as socketPresent=false even when the unit is installed (enabled-but-dead is the #1292 shape)', async (): Promise<void> => {
+    const result = await run('status', {
+      statPath: async () => true,
+      checkBrokerSocket: async () => false,
+    });
+    expect(result.plan.broker).toEqual({ unitInstalled: true, socketPresent: false });
+  });
+
+  it('probes the REAL filesystem when no seam is injected — live socket and unit report healthy, absent paths report absent (#1297 F3)', async (): Promise<void> => {
+    const dir = await mkdtemp(join(tmpdir(), 'mosaic-broker-probe-'));
+    cleanup = dir;
+    const configHome = join(dir, 'config');
+    const unitDir = join(configHome, 'systemd', 'user');
+    await mkdir(unitDir, { recursive: true });
+    await writeFile(join(unitDir, 'mosaic-lease-broker.service'), '[Unit]\n');
+    const sockPath = join(dir, 'broker.sock');
+    const server = createServer();
+    await new Promise<void>((resolve) => {
+      server.listen(sockPath, resolve);
+    });
+    try {
+      const result = await run('status', {
+        brokerSocketEnv: {
+          MOSAIC_LEASE_BROKER_SOCKET: sockPath,
+          XDG_CONFIG_HOME: configHome,
+          XDG_RUNTIME_DIR: dir,
+        },
+      });
+      expect(result.plan.broker).toEqual({ unitInstalled: true, socketPresent: true });
+      // Absent paths through the SAME seam-less path answer false — this is
+      // the half the old default got right; healthy is the half it got wrong.
+      const absent = await run('status', {
+        brokerSocketEnv: {
+          MOSAIC_LEASE_BROKER_SOCKET: join(dir, 'gone.sock'),
+          XDG_CONFIG_HOME: join(dir, 'gone-config'),
+        },
+      });
+      expect(absent.plan.broker).toEqual({ unitInstalled: false, socketPresent: false });
+    } finally {
+      await new Promise<void>((resolve) => {
+        server.close(() => resolve());
+      });
+    }
+  });
+
+  it('command start refuses with a named error when the broker socket does not appear after enable+start (#1297 F3)', async (): Promise<void> => {
+    const calls: string[][] = [];
+    await expect(
+      run('start', {
+        checkBrokerSocket: async () => false,
+        runner: async (command, args) => {
+          calls.push([command, ...args]);
+          if (command === 'tmux' && args.includes('list-sessions')) {
+            return { stdout: '_holder\ncoder0\n', stderr: '', exitCode: 0 };
+          }
+          if (command === 'tmux' && args.includes('show-environment')) {
+            return {
+              stdout:
+                'HOME=/home/mosaic\nMOSAIC_FLEET_OWNER=11111111-1111-4111-8111-111111111111\nMOSAIC_TMUX_HOLDER=_holder\nMOSAIC_TMUX_SOCKET=mosaic-fleet\nPATH=/usr/bin:/bin\nPWD=/home/mosaic\n',
+              stderr: '',
+              exitCode: 0,
+            };
+          }
+          return { stdout: '', stderr: '', exitCode: 0 };
+        },
+      }),
+    ).rejects.toThrow(/broker-absent/);
+    // Refused: broker enable+start attempted, no holder/agent unit touched.
+    const agentStarts = calls.filter(
+      (c) => c.join(' ') === 'systemctl --user start [email protected]',
+    );
+    expect(agentStarts).toHaveLength(0);
+  });
+
+  it('command start enables and starts the broker BEFORE the holder and any agent unit', async (): Promise<void> => {
+    const calls: string[][] = [];
+    const result = await run('start', {
+      // Deterministic broker presence: without the seam this test answers the
+      // HOST's broker state (passes on a machine with a live broker, refuses
+      // on CI), not the ordering property it exists for (#1297 follow-up).
+      checkBrokerSocket: async () => true,
+      runner: async (command, args) => {
+        calls.push([command, ...args]);
+        if (command === 'tmux' && args.includes('list-sessions')) {
+          return { stdout: '_holder\ncoder0\n', stderr: '', exitCode: 0 };
+        }
+        if (command === 'tmux' && args.includes('show-environment')) {
+          return {
+            stdout:
+              'HOME=/home/mosaic\nMOSAIC_FLEET_OWNER=11111111-1111-4111-8111-111111111111\nMOSAIC_TMUX_HOLDER=_holder\nMOSAIC_TMUX_SOCKET=mosaic-fleet\nPATH=/usr/bin:/bin\nPWD=/home/mosaic\n',
+            stderr: '',
+            exitCode: 0,
+          };
+        }
+        return { stdout: '', stderr: '', exitCode: 0 };
+      },
+    });
+    expect(result.lifecycle).toBe('complete');
+    const brokerEnable = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user enable mosaic-lease-broker.service',
+    );
+    const brokerStart = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user start mosaic-lease-broker.service',
+    );
+    const holderStart = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user start mosaic-tmux-holder.service',
+    );
+    const agentStart = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user start [email protected]',
+    );
+    expect(brokerEnable).toBeGreaterThanOrEqual(0);
+    expect(brokerStart).toBeGreaterThan(brokerEnable);
+    // Holder start may be absent (holder 'owned' in this fixture); if present it must follow the broker.
+    if (holderStart >= 0) expect(holderStart).toBeGreaterThan(brokerStart);
+    expect(agentStart).toBeGreaterThan(brokerStart);
+  });
+
+  it('apply with a running desired agent also enables and starts the broker first', async (): Promise<void> => {
+    const calls: string[][] = [];
+    const runningRoster: FleetRosterV2 = {
+      ...roster,
+      agents: roster.agents.map((agent) =>
+        agent.name === 'coder0'
+          ? { ...agent, lifecycle: { enabled: true, desiredState: 'running' as const } }
+          : agent,
+      ),
+    };
+    const result = await executeFleetReconcile({
+      roster: runningRoster,
+      command: 'apply',
+      expectedGeneration: 7,
+      deps: deps({
+        readRoster: async () => runningRoster,
+        // Deterministic broker presence (see start-ordering test note).
+        checkBrokerSocket: async () => true,
+        runner: async (command, args) => {
+          calls.push([command, ...args]);
+          if (command === 'tmux' && args.includes('list-sessions')) {
+            return { stdout: '_holder\n', stderr: '', exitCode: 0 };
+          }
+          if (command === 'tmux' && args.includes('show-environment')) {
+            return {
+              stdout:
+                'HOME=/home/mosaic\nMOSAIC_FLEET_OWNER=11111111-1111-4111-8111-111111111111\nMOSAIC_TMUX_HOLDER=_holder\nMOSAIC_TMUX_SOCKET=mosaic-fleet\nPATH=/usr/bin:/bin\nPWD=/home/mosaic\n',
+              stderr: '',
+              exitCode: 0,
+            };
+          }
+          return { stdout: '', stderr: '', exitCode: 0 };
+        },
+      }),
+    });
+    expect(result.applied).toBe(true);
+    const brokerStart = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user start mosaic-lease-broker.service',
+    );
+    const agentStart = calls.findIndex(
+      (c) => c.join(' ') === 'systemctl --user start [email protected]',
+    );
+    expect(brokerStart).toBeGreaterThanOrEqual(0);
+    expect(agentStart).toBeGreaterThan(brokerStart);
+  });
+
   it('fails closed on a symlinked fleet ancestor without touching its target', async (): Promise<void> => {
     const home = await lockHome();
     const fleet = join(home, 'fleet');
@@ -375,6 +549,8 @@ describe('fleet roster-owned reconciler', (): void => {
       expectedGeneration: 7,
       deps: deps({
         readRoster: async () => runningRoster,
+        // Deterministic broker presence (see start-ordering test note).
+        checkBrokerSocket: async () => true,
         runner: async (command, args) => {
           calls.push([command, ...args]);
           if (command === 'tmux' && args.includes('list-sessions')) {

packages/mosaic/src/fleet/fleet-reconciler.ts

@@ -1,5 +1,5 @@
 import { constants } from 'node:fs';
-import { lstat, open, readFile, unlink, type FileHandle } from 'node:fs/promises';
+import { lstat, open, readFile, stat, unlink, type FileHandle } from 'node:fs/promises';
 import { randomUUID } from 'node:crypto';
 import { homedir } from 'node:os';
 import { join } from 'node:path';
@@ -44,6 +44,10 @@ export interface FleetReconcileDeps {
   readonly overrideDir?: string;
   readonly homeDirectory?: string;
   readonly readHolderIdentity?: () => Promise<string>;
+  /** Test/observation seams for the lease-broker plan member (#1292). */
+  readonly statPath?: (path: string) => Promise<boolean> | boolean;
+  readonly checkBrokerSocket?: (path: string) => Promise<boolean> | boolean;
+  readonly brokerSocketEnv?: NodeJS.ProcessEnv;
   readonly validateRoster?: (roster: FleetRosterV2) => Promise<void>;
   readonly prepareProjections?: (roster: FleetRosterV2) => Promise<readonly unknown[]>;
   readonly applyProjection?: (prepared: unknown) => Promise<unknown>;
@@ -75,6 +79,17 @@ export interface FleetReconcileObservedAgent {
 export interface FleetReconcilePlan {
   readonly generation: number;
   readonly holder: 'owned' | 'missing' | 'ownership-mismatch';
+  /**
+   * Lease broker observation (#1292): every gated runtime registers with the
+   * broker or dies ~4s in — a broker not in the plan cannot be reported as
+   * drifted, which made "broker died an hour ago" and "broker fine"
+   * produce identical output. `unitInstalled` = unit file present in the
+   * active dir; `socketPresent` = live broker at the resolved socket path.
+   */
+  readonly broker: {
+    readonly unitInstalled: boolean;
+    readonly socketPresent: boolean;
+  };
   readonly agents: readonly FleetReconcileObservedAgent[];
   readonly unmanagedSessions: readonly string[];
 }
@@ -246,7 +261,17 @@ export async function executeFleetReconcile(
             lifecycle: 'complete',
             plan,
           };
-        } catch {
+        } catch (error: unknown) {
+          // A named lifecycle precondition (broker-absent after enable+start,
+          // #1297 F3) must surface as itself — converting it to the generic
+          // recoverable result would hide the diagnosis and report a clean
+          // refusal where a loud one is the point.
+          if (
+            error instanceof FleetReconcileError &&
+            error.code === 'lifecycle-precondition-failed'
+          ) {
+            throw error;
+          }
           result = {
             applied: false,
             authoritativeRoster: 'unchanged',
@@ -315,6 +340,63 @@ function isObservational(command: FleetReconcileCommand): boolean {
   return command === 'plan' || command === 'status' || command === 'verify' || command === 'doctor';
 }
 
+/**
+ * Observe the lease broker for the plan (#1292). Unit presence via systemctl
+ * is-system-running is NOT the signal — a unit can be enabled-but-dead. The
+ * authoritative signal is the socket the gated runtimes connect to, matching
+ * broker-supervisor.ts's `checkBrokerSupervisorHealth` (healthy ===
+ * socketPresent). Injectable so tests drive every branch without a broker.
+ */
+function resolveBrokerSocketPath(env: NodeJS.ProcessEnv): string {
+  const uid = typeof process.getuid === 'function' ? process.getuid() : 0;
+  const runtimeDir = env['XDG_RUNTIME_DIR'] ?? `/run/user/${uid}`;
+  return env['MOSAIC_LEASE_BROKER_SOCKET'] ?? join(runtimeDir, 'mosaic-lease', 'broker.sock');
+}
+
+/**
+ * Probe the broker socket. Seams take precedence, but with no seam injected
+ * the REAL stat().isSocket() runs (#1297 review F3): production passes no
+ * seams, and defaulting to false made plan/status/doctor report a healthy
+ * broker as absent — a dead broker was indistinguishable from noise.
+ */
+async function brokerSocketPresent(
+  deps: FleetReconcileDeps,
+  env: NodeJS.ProcessEnv,
+): Promise<boolean> {
+  const socketPath = resolveBrokerSocketPath(env);
+  const check = deps.checkBrokerSocket;
+  if (check) return check(socketPath);
+  try {
+    return (await stat(socketPath)).isSocket();
+  } catch {
+    return false;
+  }
+}
+
+async function observeBroker(deps: FleetReconcileDeps): Promise<FleetReconcilePlan['broker']> {
+  const homeDirectory = deps.homeDirectory ?? homedir();
+  const env = (deps.brokerSocketEnv ?? process.env) as NodeJS.ProcessEnv;
+  const configHome = env['XDG_CONFIG_HOME'] ?? join(homeDirectory, '.config');
+  const unitPath = join(configHome, 'systemd', 'user', 'mosaic-lease-broker.service');
+  const statPath = deps.statPath;
+  let unitInstalled = false;
+  let socketPresent = false;
+  try {
+    // Same principle as the socket probe: no seam → look at the real
+    // filesystem. A unit file placed by installFleet (or a by-path residue
+    // symlink resolving to it) satisfies stat().isFile().
+    unitInstalled = statPath ? await statPath(unitPath) : (await stat(unitPath)).isFile();
+  } catch {
+    unitInstalled = false;
+  }
+  try {
+    socketPresent = await brokerSocketPresent(deps, env);
+  } catch {
+    socketPresent = false;
+  }
+  return { unitInstalled, socketPresent };
+}
+
 async function observeFleet(
   roster: FleetRosterV2,
   deps: FleetReconcileDeps,
@@ -325,10 +407,12 @@ async function observeFleet(
     '-F',
     '#{session_name}',
   ]);
+  const broker = await observeBroker(deps);
   if (sessionsResult.exitCode !== 0) {
     return {
       generation: roster.generation,
       holder: 'missing',
+      broker,
       agents: await observeAgents(roster, deps, new Set<string>()),
       unmanagedSessions: [],
     };
@@ -351,6 +435,7 @@ async function observeFleet(
   return {
     generation: roster.generation,
     holder,
+    broker,
     agents: await observeAgents(roster, deps, sessions),
     unmanagedSessions: Object.freeze(unmanagedSessions.sort()),
   };
@@ -507,6 +592,17 @@ async function executeExplicitLifecycle(
   plan: FleetReconcilePlan,
   agents: readonly FleetRosterV2Agent[],
 ): Promise<FleetReconcileResult> {
+  const lifecycleApplyFailed = (): FleetReconcileResult => ({
+    applied: false,
+    authoritativeRoster: 'unchanged',
+    projections: 'not-applied',
+    lifecycle: 'incomplete',
+    plan,
+    recovery: {
+      code: 'lifecycle-apply-failed',
+      action: 'rerun-after-inspecting-owned-resources',
+    },
+  });
   if (request.command === 'start') {
     for (const agent of agents) {
       if (!agent.lifecycle.enabled) {
@@ -517,6 +613,40 @@ async function executeExplicitLifecycle(
       }
     }
   }
+  // Broker FIRST (#1292): a gated runtime started without a running lease
+  // broker dies ~4 seconds in at registration — enable the unit (install
+  // places it) and start it before any holder/agent lifecycle effect.
+  try {
+    if (request.command === 'start') {
+      await runChecked(request.deps, 'systemctl', [
+        '--user',
+        'enable',
+        'mosaic-lease-broker.service',
+      ]);
+      await runChecked(request.deps, 'systemctl', [
+        '--user',
+        'start',
+        'mosaic-lease-broker.service',
+      ]);
+    }
+  } catch {
+    return lifecycleApplyFailed();
+  }
+  if (request.command === 'start') {
+    // Socket re-check after start, as a NAMED precondition (#1297 review
+    // F3) — the same protection the v1 path in commands/fleet.ts has had all
+    // along: the unit reporting active is not the signal; the socket is.
+    // Deliberately outside the try/catch above: a swallowed FleetReconcileError
+    // here read as a generic recoverable failure, hiding the named refusal.
+    // Runs BEFORE any holder/agent unit is touched so nothing doomed starts.
+    const env = (request.deps.brokerSocketEnv ?? process.env) as NodeJS.ProcessEnv;
+    if (!(await brokerSocketPresent(request.deps, env))) {
+      throw new FleetReconcileError(
+        'lifecycle-precondition-failed',
+        'broker-absent: lease broker socket did not appear after enable+start (#1292; #1297 F3). Remedy: mosaic fleet install.',
+      );
+    }
+  }
   try {
     if (request.command === 'start' && plan.holder === 'missing') {
       await runChecked(request.deps, 'systemctl', [
@@ -533,17 +663,7 @@ async function executeExplicitLifecycle(
       ]);
     }
   } catch {
-    return {
+    return lifecycleApplyFailed();
-      applied: false,
-      authoritativeRoster: 'unchanged',
-      projections: 'not-applied',
-      lifecycle: 'incomplete',
-      plan,
-      recovery: {
-        code: 'lifecycle-apply-failed',
-        action: 'rerun-after-inspecting-owned-resources',
-      },
-    };
   }
   return {
     applied: true,
@@ -563,6 +683,22 @@ async function applyDesiredLifecycle(
     (agent: FleetRosterV2Agent): boolean =>
       agent.lifecycle.enabled && agent.lifecycle.desiredState === 'running',
   );
+  // Broker before any running agent, same ordering and reason as the
+  // command-driven path above (#1292).
+  if (needsRunningAgent) {
+    await runChecked(deps, 'systemctl', ['--user', 'enable', 'mosaic-lease-broker.service']);
+    await runChecked(deps, 'systemctl', ['--user', 'start', 'mosaic-lease-broker.service']);
+    // Same socket re-check as the explicit start path (#1297 F3): apply with
+    // running desired agents starts gated runtimes too, and a broker that
+    // starts but never binds dooms them the same way.
+    const env = (deps.brokerSocketEnv ?? process.env) as NodeJS.ProcessEnv;
+    if (!(await brokerSocketPresent(deps, env))) {
+      throw new FleetReconcileError(
+        'lifecycle-precondition-failed',
+        'broker-absent: lease broker socket did not appear after enable+start (#1292; #1297 F3). Remedy: mosaic fleet install.',
+      );
+    }
+  }
   if (needsRunningAgent && plan.holder === 'missing') {
     await runChecked(deps, 'systemctl', ['--user', 'start', 'mosaic-tmux-holder.service']);
   }