emrg: isolate stop_daemon in test_stop_all_process_residual_still_aborts (⛔ host red line) by argszero · Pull Request #889 · argszero/emrg

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>

argszero · 2026-08-20T08:42:52Z

Fix a ⛔ host-red-line violation (host directive 2026-08-18T22:58): running the full pytest suite killed the real local emrgd twice (2026-08-20 16:18/16:25) because test_stop_all_process_residual_still_aborts called _stop_all.stop_all() without isolating stop_daemon — stop_daemon() really connects to ws://127.0.0.1:56031 and sends shutdown to the running daemon.

Fix

tests/test_stop_all.py line 677 test: add

monkeypatch.setattr(_stop_all, stop_daemon, lambda: None)

with a ⛔ red-line comment explaining why (matches the file's existing isolation convention — the other two stop_all() tests already do this).

Audit (rant acceptance item 2)

All 3 _stop_all.stop_all() test call sites (lines 656 / 681 / 968) now isolate stop_daemon
test_scheduler.pysched.stop_all() = TaskScheduler method (stops handler coroutines, NOT the daemon process) — safe
test_installer_stop.py = static source analysis, never executes stop paths — safe

Verification

pytest tests/test_stop_all.py::TestIndependentLockProbe → 10 passed, 0.28s, no shutdown triggered
Full uv run pytest tests/ → 980 passed + 1 skipped, daemon stays alive after the suite
import + CLI OK

…rts (⛔ host red line) The test called _stop_all.stop_all() without monkeypatching stop_daemon, which REALLY connects to ws://127.0.0.1:56031 and shuts down the running emrgd — observed twice on 2026-08-20 16:18/16:25 (emrgd.log 'shutdown requested by client (source=stop_all)') while running the full pytest suite. The other two stop_all() tests in the file already isolate stop_daemon; this one was the missed spot. Fix: add monkeypatch.setattr(_stop_all, 'stop_daemon', lambda: None) with a ⛔ red-line comment, matching the file's convention. Full-suite audit confirms no other un-isolated stop_all()/stop_daemon() real-execution tests remain (test_installer_stop.py is static source analysis; scheduler stop_all() stops handler coroutines, not the daemon).

argszero

✅ LGTM — cycle 1 (single-file ⛔ red-line fix: stop_daemon isolated in the missed test, matching file convention; full-suite audit confirms no other un-isolated stop paths; pytest 980 green with daemon alive; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 2 (independent re-review: single-file change adds stop_daemon monkeypatch isolation to the missed test, matching the same-file convention; ⛔ red-line comment documents the 2026-08-20 16:18/16:25 real-daemon kill evidence; test + test-windows CI PASS)

argszero

✅ LGTM — cycle 3 (re-review: head 4b4201c unchanged, single-file ⛔ red-line isolation fix consistent with file convention; test + test-windows CI PASS, mergeable CLEAN)

…ails loudly (⛔ red line backstop) (#890) The 2026-08-20 16:18/16:25 incidents killed the live emrgd twice because tests/test_stop_all.py:677 lacked stop_daemon isolation. PR #889 fixed the specific leak; this adds the definitive anti-regression layer: an autouse conftest fixture replaces emrg._stop_all.stop_daemon with a raiser so any future test that triggers a real stop path fails with AssertionError instead of killing the daemon. A canary test verifies the guard stays installed. Agent.md pytest count synced 981 -> 982. Co-authored-by: EMRG Evolution <emrg@argszero.dev>

Co-authored-by: EMRG Evolution <emrg@argszero.dev>