personal memory agent
activity_state cold-start fix: default vocabulary + always_on meeting + post-hook validation
Three changes from CPO design opinion (activity-state-cold-start-design):
1. Mark meeting as always_on in DEFAULT_ACTIVITIES — meetings are as
universal as email and messaging, auto-included for any facet.
2. Return all 16 defaults when no activities.jsonl exists — gives
unconfigured facets full detection vocabulary out of the box.
Once a user configures explicit activities, those take over.
3. Validate activity IDs in post_process() — drops LLM output entries
whose activity field doesn't match the configured vocabulary,
preventing hallucinated activity types (e.g., facet name used as
activity ID) from polluting data. Logs warning on drops.
Also fixes add_activity_to_facet to check raw JSONL (not merged
defaults) when detecting duplicates, so adding activities still works
correctly with the cold-start default vocabulary.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+321
-24
+31
-15
muse/activity_state.py
+31
-15
muse/activity_state.py
···
529
529
logger.warning("activity_state output is not an array")
530
530
return None
531
531
532
+
# Extract facet from output path
533
+
facet = _extract_facet_from_output_path(output_path)
534
+
535
+
# Validate activity IDs against configured vocabulary
536
+
if facet:
537
+
from think.activities import get_facet_activities
538
+
539
+
valid_ids = {a["id"] for a in get_facet_activities(facet)}
540
+
original_count = len(items)
541
+
items = [item for item in items if item.get("activity", "") in valid_ids]
542
+
dropped = original_count - len(items)
543
+
if dropped:
544
+
logger.warning(
545
+
"Dropped %d activity entries with unrecognized activity IDs for facet %s",
546
+
dropped,
547
+
facet,
548
+
)
549
+
532
550
# Load previous state for since resolution
533
551
prev_active: list[dict] = []
534
552
prev_ended: list[dict] = []
535
-
if day:
536
-
facet = _extract_facet_from_output_path(output_path)
537
-
if facet:
538
-
previous_segment = find_previous_segment(day, segment, stream=stream)
539
-
if previous_segment:
540
-
prev_state, _ = load_previous_state(
541
-
day, previous_segment, facet, stream=stream
542
-
)
543
-
if prev_state:
544
-
prev_active = [
545
-
item for item in prev_state if item.get("state") == "active"
546
-
]
547
-
prev_ended = [
548
-
item for item in prev_state if item.get("state") == "ended"
549
-
]
553
+
if day and facet:
554
+
previous_segment = find_previous_segment(day, segment, stream=stream)
555
+
if previous_segment:
556
+
prev_state, _ = load_previous_state(
557
+
day, previous_segment, facet, stream=stream
558
+
)
559
+
if prev_state:
560
+
prev_active = [
561
+
item for item in prev_state if item.get("state") == "active"
562
+
]
563
+
prev_ended = [
564
+
item for item in prev_state if item.get("state") == "ended"
565
+
]
550
566
551
567
# Track which previous items have been claimed to avoid double-matching
552
568
claimed: set[int] = set()
+69
-3
tests/test_activities.py
+69
-3
tests/test_activities.py
···
105
105
def test_facet_activities_empty():
106
106
"""Test loading activities from a facet with no activities file.
107
107
108
-
Even with no activities.jsonl, always-on defaults are included.
108
+
With no activities.jsonl, all defaults are returned as the vocabulary.
109
109
"""
110
110
from think.activities import DEFAULT_ACTIVITIES, get_facet_activities
111
111
112
112
activities = get_facet_activities("personal")
113
113
assert isinstance(activities, list)
114
114
115
-
# Should contain exactly the always-on defaults
115
+
# Should contain all defaults (full vocabulary for unconfigured facets)
116
+
all_default_ids = {a["id"] for a in DEFAULT_ACTIVITIES}
117
+
assert {a["id"] for a in activities} == all_default_ids
118
+
119
+
120
+
def test_meeting_is_always_on():
121
+
"""Test that meeting is marked always_on in DEFAULT_ACTIVITIES."""
122
+
from think.activities import DEFAULT_ACTIVITIES
123
+
116
124
always_on_ids = {a["id"] for a in DEFAULT_ACTIVITIES if a.get("always_on")}
117
-
assert {a["id"] for a in activities} == always_on_ids
125
+
assert "meeting" in always_on_ids
126
+
assert "email" in always_on_ids
127
+
assert "messaging" in always_on_ids
128
+
129
+
130
+
def test_unconfigured_facet_returns_all_defaults():
131
+
"""Test that a facet with no activities.jsonl gets all 16 defaults."""
132
+
from think.activities import DEFAULT_ACTIVITIES, get_facet_activities
133
+
134
+
with tempfile.TemporaryDirectory() as tmpdir:
135
+
original_path = os.environ.get("JOURNAL_PATH")
136
+
os.environ["JOURNAL_PATH"] = tmpdir
137
+
138
+
facet_path = Path(tmpdir) / "facets" / "new_facet"
139
+
facet_path.mkdir(parents=True)
140
+
141
+
try:
142
+
activities = get_facet_activities("new_facet")
143
+
assert len(activities) == len(DEFAULT_ACTIVITIES)
144
+
145
+
activity_ids = {a["id"] for a in activities}
146
+
default_ids = {a["id"] for a in DEFAULT_ACTIVITIES}
147
+
assert activity_ids == default_ids
148
+
149
+
# All should be marked as not custom
150
+
for activity in activities:
151
+
assert activity.get("custom") is False
152
+
153
+
finally:
154
+
if original_path:
155
+
os.environ["JOURNAL_PATH"] = original_path
156
+
157
+
158
+
def test_configured_facet_includes_meeting_always_on():
159
+
"""Test that a facet with explicit activities auto-includes meeting."""
160
+
from think.activities import get_facet_activities, save_facet_activities
161
+
162
+
with tempfile.TemporaryDirectory() as tmpdir:
163
+
original_path = os.environ.get("JOURNAL_PATH")
164
+
os.environ["JOURNAL_PATH"] = tmpdir
165
+
166
+
facet_path = Path(tmpdir) / "facets" / "work"
167
+
facet_path.mkdir(parents=True)
168
+
169
+
try:
170
+
# Save only coding — meeting, email, messaging should auto-include
171
+
save_facet_activities("work", [{"id": "coding"}])
172
+
173
+
activities = get_facet_activities("work")
174
+
activity_ids = {a["id"] for a in activities}
175
+
176
+
assert "coding" in activity_ids
177
+
assert "meeting" in activity_ids
178
+
assert "email" in activity_ids
179
+
assert "messaging" in activity_ids
180
+
181
+
finally:
182
+
if original_path:
183
+
os.environ["JOURNAL_PATH"] = original_path
118
184
119
185
120
186
def test_facet_activities_roundtrip():
+204
tests/test_activity_state.py
+204
tests/test_activity_state.py
···
1392
1392
items = json.loads(result)
1393
1393
assert len(items) == 1
1394
1394
assert items[0]["id"] == "coding_143000_300"
1395
+
1396
+
1397
+
class TestActivityIdValidation:
1398
+
"""Tests for post-hook activity ID validation against configured vocabulary."""
1399
+
1400
+
def test_drops_unrecognized_activity_ids(self):
1401
+
"""Post-hook drops LLM output entries with activity IDs not in config."""
1402
+
from unittest.mock import patch
1403
+
1404
+
from muse.activity_state import post_process
1405
+
1406
+
with tempfile.TemporaryDirectory() as tmpdir:
1407
+
original_path = os.environ.get("JOURNAL_PATH")
1408
+
os.environ["JOURNAL_PATH"] = tmpdir
1409
+
1410
+
try:
1411
+
# Create facet with only coding and meeting configured
1412
+
facet_dir = Path(tmpdir) / "facets" / "work" / "activities"
1413
+
facet_dir.mkdir(parents=True)
1414
+
(facet_dir / "activities.jsonl").write_text(
1415
+
'{"id": "coding"}\n{"id": "meeting"}'
1416
+
)
1417
+
1418
+
llm_output = json.dumps(
1419
+
[
1420
+
{
1421
+
"activity": "coding",
1422
+
"state": "new",
1423
+
"description": "Writing code",
1424
+
"level": "high",
1425
+
},
1426
+
{
1427
+
"activity": "Technical Work",
1428
+
"state": "new",
1429
+
"description": "Hallucinated from facet name",
1430
+
"level": "medium",
1431
+
},
1432
+
]
1433
+
)
1434
+
1435
+
context = {
1436
+
"segment": "143000_300",
1437
+
"output_path": f"{tmpdir}/20260130/143000_300/agents/work/activity_state.json",
1438
+
}
1439
+
1440
+
with patch("muse.activity_state.callosum_send"):
1441
+
result = post_process(llm_output, context)
1442
+
1443
+
items = json.loads(result)
1444
+
assert len(items) == 1
1445
+
assert items[0]["activity"] == "coding"
1446
+
1447
+
finally:
1448
+
if original_path:
1449
+
os.environ["JOURNAL_PATH"] = original_path
1450
+
1451
+
def test_logs_warning_on_dropped_activities(self, caplog):
1452
+
"""Post-hook logs a warning when dropping unrecognized activity IDs."""
1453
+
import logging
1454
+
from unittest.mock import patch
1455
+
1456
+
from muse.activity_state import post_process
1457
+
1458
+
with tempfile.TemporaryDirectory() as tmpdir:
1459
+
original_path = os.environ.get("JOURNAL_PATH")
1460
+
os.environ["JOURNAL_PATH"] = tmpdir
1461
+
1462
+
try:
1463
+
facet_dir = Path(tmpdir) / "facets" / "work" / "activities"
1464
+
facet_dir.mkdir(parents=True)
1465
+
(facet_dir / "activities.jsonl").write_text('{"id": "coding"}')
1466
+
1467
+
llm_output = json.dumps(
1468
+
[
1469
+
{
1470
+
"activity": "meetings",
1471
+
"state": "new",
1472
+
"description": "Hallucinated",
1473
+
"level": "medium",
1474
+
},
1475
+
]
1476
+
)
1477
+
1478
+
context = {
1479
+
"segment": "143000_300",
1480
+
"output_path": f"{tmpdir}/20260130/143000_300/agents/work/activity_state.json",
1481
+
}
1482
+
1483
+
with caplog.at_level(logging.WARNING, logger="muse.activity_state"):
1484
+
with patch("muse.activity_state.callosum_send"):
1485
+
post_process(llm_output, context)
1486
+
1487
+
assert "Dropped 1 activity entries" in caplog.text
1488
+
assert "work" in caplog.text
1489
+
1490
+
finally:
1491
+
if original_path:
1492
+
os.environ["JOURNAL_PATH"] = original_path
1493
+
1494
+
def test_valid_activity_ids_pass_through(self):
1495
+
"""Post-hook preserves entries with valid activity IDs."""
1496
+
from unittest.mock import patch
1497
+
1498
+
from muse.activity_state import post_process
1499
+
1500
+
with tempfile.TemporaryDirectory() as tmpdir:
1501
+
original_path = os.environ.get("JOURNAL_PATH")
1502
+
os.environ["JOURNAL_PATH"] = tmpdir
1503
+
1504
+
try:
1505
+
facet_dir = Path(tmpdir) / "facets" / "work" / "activities"
1506
+
facet_dir.mkdir(parents=True)
1507
+
(facet_dir / "activities.jsonl").write_text(
1508
+
'{"id": "coding"}\n{"id": "meeting"}'
1509
+
)
1510
+
1511
+
llm_output = json.dumps(
1512
+
[
1513
+
{
1514
+
"activity": "coding",
1515
+
"state": "new",
1516
+
"description": "Writing code",
1517
+
"level": "high",
1518
+
},
1519
+
{
1520
+
"activity": "meeting",
1521
+
"state": "new",
1522
+
"description": "Standup",
1523
+
"level": "medium",
1524
+
},
1525
+
{
1526
+
"activity": "email",
1527
+
"state": "new",
1528
+
"description": "Checking inbox",
1529
+
"level": "low",
1530
+
},
1531
+
]
1532
+
)
1533
+
1534
+
context = {
1535
+
"segment": "143000_300",
1536
+
"output_path": f"{tmpdir}/20260130/143000_300/agents/work/activity_state.json",
1537
+
}
1538
+
1539
+
with patch("muse.activity_state.callosum_send"):
1540
+
result = post_process(llm_output, context)
1541
+
1542
+
items = json.loads(result)
1543
+
# All three are valid: coding + meeting explicit, email always_on
1544
+
assert len(items) == 3
1545
+
activity_ids = {item["activity"] for item in items}
1546
+
assert activity_ids == {"coding", "meeting", "email"}
1547
+
1548
+
finally:
1549
+
if original_path:
1550
+
os.environ["JOURNAL_PATH"] = original_path
1551
+
1552
+
def test_unconfigured_facet_allows_all_defaults(self):
1553
+
"""Post-hook allows all default activity IDs for unconfigured facets."""
1554
+
from unittest.mock import patch
1555
+
1556
+
from muse.activity_state import post_process
1557
+
1558
+
with tempfile.TemporaryDirectory() as tmpdir:
1559
+
original_path = os.environ.get("JOURNAL_PATH")
1560
+
os.environ["JOURNAL_PATH"] = tmpdir
1561
+
1562
+
try:
1563
+
# Create facet dir but no activities.jsonl
1564
+
facet_dir = Path(tmpdir) / "facets" / "new_facet"
1565
+
facet_dir.mkdir(parents=True)
1566
+
1567
+
llm_output = json.dumps(
1568
+
[
1569
+
{
1570
+
"activity": "meeting",
1571
+
"state": "new",
1572
+
"description": "Team sync",
1573
+
"level": "high",
1574
+
},
1575
+
{
1576
+
"activity": "coding",
1577
+
"state": "new",
1578
+
"description": "Writing code",
1579
+
"level": "medium",
1580
+
},
1581
+
]
1582
+
)
1583
+
1584
+
context = {
1585
+
"segment": "143000_300",
1586
+
"output_path": f"{tmpdir}/20260130/143000_300/agents/new_facet/activity_state.json",
1587
+
}
1588
+
1589
+
with patch("muse.activity_state.callosum_send"):
1590
+
result = post_process(llm_output, context)
1591
+
1592
+
items = json.loads(result)
1593
+
# Both are valid defaults
1594
+
assert len(items) == 2
1595
+
1596
+
finally:
1597
+
if original_path:
1598
+
os.environ["JOURNAL_PATH"] = original_path
+17
-6
think/activities.py
+17
-6
think/activities.py
···
34
34
"name": "Meetings",
35
35
"description": "Video calls, in-person meetings, and conferences",
36
36
"icon": "📅",
37
+
"always_on": True,
37
38
"instructions": (
38
39
"Levels: high=actively speaking/presenting, medium=listening attentively,"
39
40
" low=muted or multitasking during call."
···
298
299
# Load facet-specific activities
299
300
facet_activities = _load_activities_jsonl(facet)
300
301
302
+
# If no explicit activities configured, use all defaults as the vocabulary
303
+
if not facet_activities:
304
+
result = []
305
+
for default in DEFAULT_ACTIVITIES:
306
+
activity = dict(default)
307
+
activity["custom"] = False
308
+
activity.setdefault("priority", "normal")
309
+
result.append(activity)
310
+
return result
311
+
301
312
seen_ids: set[str] = set()
302
313
result = []
303
314
for fa in facet_activities:
···
469
480
Returns:
470
481
The added activity dict
471
482
"""
472
-
# Check if already attached
473
-
existing = get_facet_activities(facet)
474
-
for activity in existing:
475
-
if activity.get("id") == activity_id:
476
-
# Already attached - return existing
477
-
return activity
483
+
# Check if already explicitly attached (in JSONL, not just defaults)
484
+
existing_raw = _load_activities_jsonl(facet)
485
+
for entry in existing_raw:
486
+
if entry.get("id") == activity_id:
487
+
# Already attached - return full activity with defaults merged
488
+
return get_activity_by_id(facet, activity_id) or entry
478
489
479
490
# Build new activity entry
480
491
defaults_by_id = {a["id"]: a for a in DEFAULT_ACTIVITIES}