personal memory agent
speakers: wipe command also clears legacy agents/ label paths
`sol call speakers wipe` globbed only chronicle/*/*/*/talents/ for
speaker_labels.json + speaker_corrections.json. In practice every
historical segment stores those files under agents/ — talents/ is the
destination for new writes from apps/speakers/attribution.py. A real
dry-run on jer's journal reported 0 labels and 0 corrections when the
actual counts are 3360 + 96.
Cover both paths so the wipe catches historical and new-write artifacts.
Test fixture now creates both layouts and asserts counts of 2 in each
category.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
+34
-18
+14
-10
apps/speakers/tests/test_wipe.py
+14
-10
apps/speakers/tests/test_wipe.py
···
27
27
28
28
targets = [
29
29
journal / "chronicle/20240101/test/120000_300/mic_audio.npz",
30
-
journal / "chronicle/20240101/test/120000_300/talents/speaker_labels.json",
31
-
journal / "chronicle/20240101/test/120000_300/talents/speaker_corrections.json",
30
+
# labels + corrections live under both agents/ (historical) and talents/
31
+
# (current writes); the wipe must catch both.
32
+
journal / "chronicle/20240101/test/120000_300/agents/speaker_labels.json",
33
+
journal / "chronicle/20240101/test/120000_300/agents/speaker_corrections.json",
34
+
journal / "chronicle/20240102/test/120000_300/talents/speaker_labels.json",
35
+
journal / "chronicle/20240102/test/120000_300/talents/speaker_corrections.json",
32
36
journal / "entities/alice/voiceprints.npz",
33
37
journal / "entities/alice/owner_centroid.npz",
34
38
journal / "entities/bob/owner_centroid.npz",
···
49
53
report = wipe_speaker_artifacts(dry_run=True)
50
54
51
55
assert report.segment_embeddings.count == 1
52
-
assert report.speaker_labels.count == 1
53
-
assert report.speaker_corrections.count == 1
56
+
assert report.speaker_labels.count == 2 # agents/ + talents/
57
+
assert report.speaker_corrections.count == 2 # agents/ + talents/
54
58
assert report.entity_voiceprints.count == 1
55
59
assert report.owner_centroids.count == 2
56
60
assert report.owner_candidate.count == 1
57
-
assert report.total_files == 7
58
-
assert report.total_bytes == 7 * len(FILE_BYTES)
61
+
assert report.total_files == 9
62
+
assert report.total_bytes == 9 * len(FILE_BYTES)
59
63
60
64
for path in targets:
61
65
assert path.exists()
···
67
71
report = wipe_speaker_artifacts(dry_run=False)
68
72
69
73
assert report.segment_embeddings.count == 1
70
-
assert report.speaker_labels.count == 1
71
-
assert report.speaker_corrections.count == 1
74
+
assert report.speaker_labels.count == 2 # agents/ + talents/
75
+
assert report.speaker_corrections.count == 2 # agents/ + talents/
72
76
assert report.entity_voiceprints.count == 1
73
77
assert report.owner_centroids.count == 2
74
78
assert report.owner_candidate.count == 1
75
-
assert report.total_files == 7
76
-
assert report.total_bytes == 7 * len(FILE_BYTES)
79
+
assert report.total_files == 9
80
+
assert report.total_bytes == 9 * len(FILE_BYTES)
77
81
78
82
for path in targets:
79
83
assert not path.exists()
+20
-8
apps/speakers/wipe.py
+20
-8
apps/speakers/wipe.py
···
38
38
journal = Path(get_journal())
39
39
report = WipeReport()
40
40
41
+
# Labels and corrections currently live under agents/ for most historical
42
+
# segments; the attribution module now writes new files under talents/.
43
+
# Cover both paths so no resemblyzer-space artifacts leak through the wipe.
44
+
label_patterns = [
45
+
"chronicle/*/*/*/agents/speaker_labels.json",
46
+
"chronicle/*/*/*/talents/speaker_labels.json",
47
+
]
48
+
correction_patterns = [
49
+
"chronicle/*/*/*/agents/speaker_corrections.json",
50
+
"chronicle/*/*/*/talents/speaker_corrections.json",
51
+
]
52
+
53
+
def _expand(patterns: list[str]) -> list[Path]:
54
+
paths: list[Path] = []
55
+
for pattern in patterns:
56
+
paths.extend(journal.glob(pattern))
57
+
return sorted(paths)
58
+
41
59
categories: list[tuple[WipeCategory, list[Path]]] = [
42
60
(report.segment_embeddings, sorted(journal.glob("chronicle/*/*/*/*.npz"))),
43
-
(
44
-
report.speaker_labels,
45
-
sorted(journal.glob("chronicle/*/*/*/talents/speaker_labels.json")),
46
-
),
47
-
(
48
-
report.speaker_corrections,
49
-
sorted(journal.glob("chronicle/*/*/*/talents/speaker_corrections.json")),
50
-
),
61
+
(report.speaker_labels, _expand(label_patterns)),
62
+
(report.speaker_corrections, _expand(correction_patterns)),
51
63
(report.entity_voiceprints, sorted(journal.glob("entities/*/voiceprints.npz"))),
52
64
(report.owner_centroids, sorted(journal.glob("entities/*/owner_centroid.npz"))),
53
65
(report.owner_candidate, [journal / "awareness" / "owner_candidate.npz"]),