fix: wake native away mode on captain-relevant events

[ruby-dlee]

Intent

Fix the Claude away-mode overnight escalation wedge in one single-concern PR, with a postmortem committed before the implementation. Preserve bash triage and batching so routine wakes remain token-free, but deliver captain-relevant needs-decision, ready-to-merge, and failure events by completing Claude's native tracked background task instead of typing into the composer. Keep state/.wake-queue as the lossless backlog, remove pane busy/composer guards from native delivery, retain honest legacy compatibility semantics, update the /afk skill and AGENTS.md section 8, and prove in isolated regression tests that native delivery succeeds under the old false-defer conditions. Hold the PR for captain merge and do not merge it.

What Changed

• Route captain-relevant native AFK escalations through tracked background-task completion while preserving token-free Bash triage, batching, and the lossless wake queue.
• Keep terminal-backed injection compatibility, but bypass pane-target, busy, and composer guards for native delivery and support re-arming without leaving away mode.
• Add regression coverage for the former false-defer conditions and document native/compatibility semantics alongside the injection-wedge postmortem.

Risk Assessment

✅ Low: Captain, the change is focused, preserves the durable wake backlog and legacy delivery path, and conforms to the stated native-delivery requirements without source-verifiable defects.

Testing

The supplied full baseline already passed; focused regressions and an isolated three-event runtime scenario also passed with CLI transcripts captured, and final cleanup left the worktree clean. No screenshot was applicable to this non-UI shell behavior.

Evidence: All escalation classes batched through native delivery

^D��ROUTINE: tracked task remained parked; no completion notification
COMPLETION:
  afk: starting supervise daemon in foreground; keep this command as a tracked background session
  afk-reap-wake: Supervisor escalate (3 event(s)): needs.status: needs-decision: choose the safe rollout | ready.status: done: PR https://github.com/Ruby-Labs/relvino/pull/4242 checks green | failure.status: failed: deployment smoke test exited 2 (pre-read); drain state/.wake-queue, handle the batch, then restart the away daemon as a native tracked background task
DURABLE WAKE BACKLOG:
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/routine.status
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/routine.status
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/needs.status
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/needs.status
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/ready.status
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/ready.status
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/failure.status
  - kind=signal payload=signal: /Users/dongkeun/.no-mistakes/worktrees/b174997b7206/01KY50TWT58BM1EWXNFSQH98MD/.afk-native-all-events.4jEJRE/failure.status
PERSISTED STATUS RECORDS REFERENCED BY THE BACKLOG:
  - needs.status: needs-decision: choose the safe rollout
  - ready.status: done: PR https://github.com/Ruby-Labs/relvino/pull/4242 checks green
  - failure.status: failed: deployment smoke test exited 2
POST-DELIVERY: away mode remains active; legacy injection-wedge marker absent

Evidence: Focused false-defer regression transcript

^D��ok - routine wake stays in bash and keeps the native task parked
ok - captain-relevant wake completes natively despite old pane-guard false-defer inputs
ok - completed native task can be re-armed without leaving away mode
all native afk reap-wake tests passed

Evidence: Reproducible multi-event evidence harness

#!/usr/bin/env bash
set -euo pipefail
export FM_GATE_REFUSE_BYPASS=1

ROOT=${1:?usage: native-reap-wake-all-events.sh <repo-root>}
START="$ROOT/bin/fm-afk-start.sh"
STATE_DIR=
DAEMON_PID=

cleanup() {
  if [ -n "${DAEMON_PID:-}" ] && kill -0 "$DAEMON_PID" 2>/dev/null; then
    kill "$DAEMON_PID" 2>/dev/null || true
    wait "$DAEMON_PID" 2>/dev/null || true
  fi
  [ -z "${STATE_DIR:-}" ] || rm -rf "$STATE_DIR" 2>/dev/null || true
}
trap cleanup EXIT

wait_for_seen() {
  local marker=$1 expected=$2
  for _ in $(seq 1 120); do
    if [ -f "$marker" ] && grep -F "$expected" "$marker" >/dev/null 2>&1; then
      return 0
    fi
    sleep 0.1
  done
  echo "error: watcher did not classify '$expected'" >&2
  return 1
}

STATE_DIR=$(mktemp -d "$ROOT/.afk-native-all-events.XXXXXX")
: > "$STATE_DIR/.afk"
printf 'none\t-\tnative\n' > "$STATE_DIR/.afk-daemon-terminal"

FM_STATE_OVERRIDE="$STATE_DIR" \
FM_AFK_STATE_PREPARED=1 \
FM_ESCALATE_BATCH_SECS=10 \
FM_HOUSEKEEPING_TICK=1 \
FM_POLL=1 \
FM_SIGNAL_GRACE=1 \
FM_HEARTBEAT=999999 \
FM_CHECK_INTERVAL=999999 \
FM_BUSY_REGEX='.' \
FM_SUPERVISOR_BACKEND=unsupported \
FM_SUPERVISOR_TARGET=missing-pane \
"$START" > "$STATE_DIR/daemon.out" 2> "$STATE_DIR/daemon.err" &
DAEMON_PID=$!

for _ in $(seq 1 60); do
  [ -f "$STATE_DIR/.supervise-daemon.pid" ] && break
  sleep 0.1
done
if [ ! -f "$STATE_DIR/.supervise-daemon.pid" ]; then
  echo "error: native away daemon did not start" >&2
  sed 's/^/  /' "$STATE_DIR/daemon.err" >&2
  exit 1
fi

printf 'working: routine progress\n' > "$STATE_DIR/routine.status"
sleep 3
kill -0 "$DAEMON_PID"
[ "$(grep -c 'afk-reap-wake:' "$STATE_DIR/daemon.out" 2>/dev/null || true)" -eq 0 ]
echo "ROUTINE: tracked task remained parked; no completion notification"

needs='needs-decision: choose the safe rollout'
ready='done: PR https://github.com/Ruby-Labs/relvino/pull/4242 checks green'
failed='failed: deployment smoke test exited 2'

printf '%s\n' "$needs" > "$STATE_DIR/needs.status"
wait_for_seen "$STATE_DIR/.subsuper-seen-status-needs" "$needs"
printf '%s\n' "$ready" > "$STATE_DIR/ready.status"
wait_for_seen "$STATE_DIR/.subsuper-seen-status-ready" "$ready"
printf '%s\n' "$failed" > "$STATE_DIR/failure.status"
wait_for_seen "$STATE_DIR/.subsuper-seen-status-failure" "$failed"

for _ in $(seq 1 240); do
  kill -0 "$DAEMON_PID" 2>/dev/null || break
  sleep 0.1
done
if kill -0 "$DAEMON_PID" 2>/dev/null; then
  echo "error: batched native delivery did not complete" >&2
  exit 1
fi
wait "$DAEMON_PID"
DAEMON_PID=

grep -F 'Supervisor escalate (3 event(s))' "$STATE_DIR/daemon.out" >/dev/null
grep -F "$needs" "$STATE_DIR/daemon.out" >/dev/null
grep -F "$ready" "$STATE_DIR/daemon.out" >/dev/null
grep -F "$failed" "$STATE_DIR/daemon.out" >/dev/null
grep -F 'needs.status' "$STATE_DIR/.wake-queue" >/dev/null
grep -F 'ready.status' "$STATE_DIR/.wake-queue" >/dev/null
grep -F 'failure.status' "$STATE_DIR/.wake-queue" >/dev/null
[ -e "$STATE_DIR/.afk" ]
[ ! -e "$STATE_DIR/.subsuper-inject-wedged" ]

echo "COMPLETION:"
sed 's/^/  /' "$STATE_DIR/daemon.out"
echo "DURABLE WAKE BACKLOG:"
awk -F '\t' '{printf "  - kind=%s payload=%s\n", $3, $5}' "$STATE_DIR/.wake-queue"
echo "PERSISTED STATUS RECORDS REFERENCED BY THE BACKLOG:"
for status_file in needs.status ready.status failure.status; do
  printf '  - %s: ' "$status_file"
  cat "$STATE_DIR/$status_file"
done
echo "POST-DELIVERY: away mode remains active; legacy injection-wedge marker absent"

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

✅ **Test** - passed

✅ No issues found.

• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; uv run --directory tools/agent-fleet --locked pytest || rc=1; uv run --directory tools/agent-fleet --locked python -m compileall -q src || rc=1; exit "$rc"
• Pre-supplied successful baseline: command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; uv run --directory tools/agent-fleet --locked pytest || rc=1; uv run --directory tools/agent-fleet --locked python -m compileall -q src || rc=1; exit "$rc"
• bash tests/fm-afk-reap-wake-e2e.test.sh captured via script
• bash tests/fm-daemon.test.sh
• bash …/native-reap-wake-all-events.sh <worktree> captured via script
• git diff-tree --no-commit-id --name-only -r bda8705 and ordered base-to-target git log
• Final git status --short --untracked-files=all and transient test-directory scan

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.