personal memory agent
feat(schedule): hydrate facet enum at runtime + drop dead density/content_type defaults
Mirror the sense runtime-enum hydration pattern for schedule's `facet` and
add a drift detector for the activity enum against DEFAULT_ACTIVITIES.
Strip impossible degraded-payload fallbacks on density/content_type at the
sense consumer sites and the matching tests.
+46
-69
+3
-1
talent/schedule.md
+3
-1
talent/schedule.md
···
14
14
15
15
$daily_preamble
16
16
17
+
$facets
18
+
17
19
# Future Schedule Extraction
18
20
19
21
**Input:** A markdown file containing chronologically ordered transcripts of a workday plus the screen agent's output for the same day. Calendar views, meeting invitations, scheduling UIs, and project-management interfaces are captured in the screen content; verbal mentions of future plans appear in transcripts.
···
95
97
96
98
- **`participation_confidence`** — `0.0`–`1.0` — overall confidence in the participation list for this item. Lower when many attendees are inferred rather than confirmed.
97
99
98
-
- **`facet`** — Facet ID from the configured facets context. Required. If no facet fits cleanly, skip the item rather than miscategorizing.
100
+
- **`facet`** — Required. The facet ID slug; MUST be one of the configured facets listed in the input. Pick the closest configured facet for the scheduled item; if multiple plausibly fit, choose the one most central to the event's purpose. Do not invent slugs.
99
101
100
102
- **`cancelled`** — Boolean. `true` when the screen shows this item as cancelled (strikethrough, "Cancelled" label, declined, greyed out). `false` otherwise.
101
103
+1
-1
talent/schedule.schema.json
+1
-1
talent/schedule.schema.json
+38
-17
tests/test_schedule_schema.py
+38
-17
tests/test_schedule_schema.py
···
6
6
7
7
from jsonschema import Draft202012Validator
8
8
9
-
from think.talent import get_talent
9
+
from think.activities import DEFAULT_ACTIVITIES
10
+
from think.talent import (
11
+
RUNTIME_FACETS_SENTINEL,
12
+
get_talent,
13
+
hydrate_runtime_enums,
14
+
)
10
15
11
16
TALENT_DIR = Path(__file__).resolve().parents[1] / "talent"
12
17
PARTICIPATION_ENTRY_SCHEMA_PATH = TALENT_DIR / "participation_entry.schema.json"
13
18
SCHEDULE_SCHEMA_PATH = TALENT_DIR / "schedule.schema.json"
14
-
15
-
SCHEDULE_ACTIVITY_ENUM = {
16
-
"meeting",
17
-
"call",
18
-
"deadline",
19
-
"appointment",
20
-
"event",
21
-
"travel",
22
-
"reminder",
23
-
"errand",
24
-
"celebration",
25
-
"doctor_appointment",
26
-
}
27
19
28
20
SCHEDULE_REQUIRED_FIELDS = {
29
21
"activity",
···
50
42
return schema
51
43
52
44
45
+
def _expected_schedule_activity_ids() -> set[str]:
46
+
# Why: `meeting` is emitted by both the schedule talent and sense; the
47
+
# other 9 are schedule-only (their instructions carry the marker).
48
+
schedule_only = {
49
+
a["id"]
50
+
for a in DEFAULT_ACTIVITIES
51
+
if "Scheduled events emitted by talent/schedule.md"
52
+
in a.get("instructions", "")
53
+
}
54
+
return {"meeting"} | schedule_only
55
+
56
+
53
57
def _sample_schedule_payloads() -> list[list[dict]]:
54
58
return [
55
59
[
···
145
149
assert get_talent("schedule")["json_schema"] == _load_schedule_schema()
146
150
147
151
152
+
def test_schedule_schema_facet_uses_runtime_sentinel_constant():
153
+
schema = _load_schedule_schema()
154
+
facet_schema = schema["items"]["properties"]["facet"]
155
+
156
+
assert facet_schema["enum"] == [RUNTIME_FACETS_SENTINEL]
157
+
158
+
159
+
def test_schedule_activity_enum_matches_default_activities_drift_detector():
160
+
schema = _load_schedule_schema()
161
+
item_schema = schema["items"]
162
+
163
+
assert set(item_schema["properties"]["activity"]["enum"]) == (
164
+
_expected_schedule_activity_ids()
165
+
)
166
+
167
+
148
168
def test_schedule_participation_entry_diverges_from_shared_fragment():
149
169
"""Schedule omits entity_id because the hook fills it via find_matching_entity."""
150
170
schedule_schema = _load_schedule_schema()
···
175
195
176
196
assert schedule_schema["type"] == "array"
177
197
assert set(item_schema["required"]) == SCHEDULE_REQUIRED_FIELDS
178
-
assert set(properties["activity"]["enum"]) == SCHEDULE_ACTIVITY_ENUM
198
+
assert set(properties["activity"]["enum"]) == _expected_schedule_activity_ids()
179
199
assert (
180
200
participation_items["properties"]["role"]["enum"]
181
201
== fragment["properties"]["role"]["enum"]
···
189
209
assert properties["cancelled"]["type"] == "boolean"
190
210
191
211
192
-
def test_schedule_hook_fixtures_validate_against_schema():
193
-
validator = Draft202012Validator(_load_schedule_schema())
212
+
def test_schedule_hook_fixtures_validate_against_schema(monkeypatch):
213
+
monkeypatch.setattr("think.talent.get_facets", lambda: {"work": {}})
214
+
validator = Draft202012Validator(hydrate_runtime_enums(_load_schedule_schema()))
194
215
195
216
for payload in _sample_schedule_payloads():
196
217
assert list(validator.iter_errors(payload)) == []
-46
tests/test_sense_splitter.py
-46
tests/test_sense_splitter.py
···
171
171
172
172
173
173
class TestEdgeCases:
174
-
def test_missing_optional_fields(self, tmp_path):
175
-
from think.sense_splitter import write_sense_outputs
176
-
177
-
seg_dir = Path(tmp_path) / "20260304" / "default" / "090000_300"
178
-
179
-
# Advisory schema validation means the splitter must still tolerate degraded input.
180
-
write_sense_outputs({}, seg_dir)
181
-
182
-
agents_dir = seg_dir / "talents"
183
-
assert (agents_dir / "activity.md").exists()
184
-
assert (agents_dir / "facets.json").exists()
185
-
assert (agents_dir / "density.json").exists()
186
-
assert (agents_dir / "sense.json").exists()
187
-
assert (agents_dir / "activity.md").read_text(encoding="utf-8") == ""
188
-
assert (
189
-
json.loads((agents_dir / "facets.json").read_text(encoding="utf-8")) == []
190
-
)
191
-
density = json.loads((agents_dir / "density.json").read_text(encoding="utf-8"))
192
-
assert density["classification"] == "active"
193
-
194
-
def test_null_fields(self, tmp_path):
195
-
from think.sense_splitter import write_sense_outputs
196
-
197
-
seg_dir = Path(tmp_path) / "20260304" / "default" / "090000_300"
198
-
sense_json = {
199
-
"density": None,
200
-
"content_type": None,
201
-
"activity_summary": None,
202
-
"entities": None,
203
-
"facets": None,
204
-
"meeting_detected": None,
205
-
"speakers": None,
206
-
}
207
-
208
-
# Advisory schema validation means the splitter must still tolerate degraded input.
209
-
write_sense_outputs(sense_json, seg_dir)
210
-
211
-
agents_dir = seg_dir / "talents"
212
-
assert (agents_dir / "activity.md").read_text(encoding="utf-8") == ""
213
-
assert (
214
-
json.loads((agents_dir / "facets.json").read_text(encoding="utf-8")) == []
215
-
)
216
-
density = json.loads((agents_dir / "density.json").read_text(encoding="utf-8"))
217
-
assert density["classification"] == "active"
218
-
assert not (agents_dir / "speakers.json").exists()
219
-
220
174
def test_empty_activity_summary(self, tmp_path):
221
175
from think.sense_splitter import write_sense_outputs
222
176
+2
-2
think/activity_state_machine.py
+2
-2
think/activity_state_machine.py
···
140
140
if self._should_reset(segment_key, day, previous_segment_key):
141
141
changes.extend(self._end_all(segment_key, "ended_gap"))
142
142
143
-
density = sense_output.get("density") or "active"
144
-
content_type = sense_output.get("content_type") or "idle"
143
+
density = sense_output["density"]
144
+
content_type = sense_output["content_type"]
145
145
activity_summary = sense_output.get("activity_summary") or ""
146
146
raw_entities = sense_output.get("entities") or []
147
147
entity_names = [
+1
-1
think/sense_splitter.py
+1
-1
think/sense_splitter.py
···
30
30
"""Write unified Sense output into per-agent files."""
31
31
agents_dir = seg_dir / "talents"
32
32
33
-
density = sense_json.get("density") or "active"
33
+
density = sense_json["density"]
34
34
activity_summary = sense_json.get("activity_summary") or ""
35
35
entities = sense_json.get("entities") or []
36
36
facets = sense_json.get("facets") or []
+1
-1
think/thinking.py
+1
-1
think/thinking.py
···
645
645
return (total_success, total_failed + 1, failed_names)
646
646
647
647
write_sense_outputs(sense_json, seg_dir, stream=stream)
648
-
density = sense_json.get("density") or "active"
648
+
density = sense_json["density"]
649
649
_jsonl_log(
650
650
"sense.complete",
651
651
mode=target_schedule,