[kshitijk4poor]

While hermes pause is engaged, a due heartbeat no longer sends the paused notice to the chat on every 5s poll (it fires once after hermes resume), and scheduled /loop wakeups no longer start agent turns until resume.

Salvages #126905; supersedes the duplicate #127525. No linked issue.

What happened

The heartbeat poller had no pause check, so it injected the due tick on every poll; _hm_estop_gate answered each one with the paused notice and the tick stayed due, so the notice repeated indefinitely (reported as ~10k Telegram messages in 6h and a flood ban). /loop wakeups are injected as internal events, and _hm_estop_gate returns early for internal events, so hermes pause did not stop them at all.

Change

• gateway/run_goals.py _heartbeat_poll_watch: skip while paused, before the tick is claimed, so it fires normally after resume.
• gateway/run_goals.py _loop_wakeup_fire_one: check_paused("loop", logger) after is_due and before fire_tick. is_due only reads state, so the tick stays due for the first scan after resume; the watcher's 15s sleep per scan means no busy loop.
• gateway/run_inbound.py _hm_estop_gate: a heartbeat event that slips past the poller check is dropped silently (""). Real user messages still get the paused notice. Docstring explains why.
• Tests: 2 new in tests/gateway/test_heartbeat_poller.py (contributor); a pause/resume leg folded into the existing test_loop_wakeup_from_secondary_route_fires_through_its_own_bot in tests/gateway/test_multiplex_notice_egress_profile_adapter.py (no new test function).
• contributors/emails/: map entry for the contributor.

Credit

• @dzianisv wrote the heartbeat fix and its two tests in fix(gateway): heartbeats must not resend the paused notice every poll #126905 (first submitter, 2026-09-28). Commit cherry-picked with authorship kept.
• fix(P1): fix(gateway): heartbeats must not resend paused notice every poll #127525 by @creditlabsonarc-rgb (opened 2026-09-29) proposes the same heartbeat fix; fix(gateway): heartbeats must not resend the paused notice every poll #126905 was earlier.
• Follow-up commits by @kshitijk4poor: the /loop pause check and its test leg, the _hm_estop_gate docstring, and the contributor email map.

Verification

• On the rebased head: test_heartbeat_poller.py, test_loop_command.py, test_multiplex_notice_egress_profile_adapter.py, cron/test_estop.py: 37 passed. test_heartbeat_acceptance.py, test_heartbeat_execution_ownership.py, test_heartbeat_watch_lifecycle.py, test_telegram_reply_mode.py: 33 passed. tui_gateway/test_notification_turn_release.py: 6 passed.
• Red on base: both new heartbeat tests fail on main (a heartbeat MessageEvent is received and the paused notice is sent). With the /loop check reverted, the folded test fails with assert ([1], ['tick']) == ([], []).
• Probe, base vs head: a due /loop 5m /status under hermes pause ran on main (handled ['/status'], 1 tick fired) and is held on head ([], 0 ticks), then fires once after resume.
• Mutation: reverting the poller check, the gate's heartbeat drop, or the /loop check each turns a kept test red.
• ruff F401/F811/F821: no new findings vs main. git diff --check clean. Windows-footguns check: no new findings (one existing write_text without encoding= at line 67 of the multiplex test file is on main and untouched here).
• Reviewed over 2 rounds (regression, reuse, quality, efficiency, tests/probe/mutation).

Maintainer notes

• The stack also fixes hermes pause not stopping /loop wakeups: they are internal events, which bypassed _hm_estop_gate. This was found in review and is not part of fix(gateway): heartbeats must not resend the paused notice every poll #126905.
• Other internal-event producers (process-completion notifications, wake, CLI handoff, plugin injection) still bypass the gate, as they do on main. They are one-shot continuations of in-flight work, not recurring ticks, so they are left as they are.
• Pause expiry stays out of scope (feat(estop): keep an allowlisted identity working through a pause, and bound the pause itself #121054).