personal memory agent
talents/sense: migrate to external sense.schema.json
Move sense's structured-output schema out of the inline ```json fenced
block in talent/sense.md into talent/sense.schema.json (Draft 2020-12,
additionalProperties:false at every closed level), wired via the
`schema` frontmatter field that landed in cde43afc. The inline ##
Output Schema block becomes a one-line pointer; the field-by-field
prose stays as the model's semantic reference.
First real consumer of the structured-outputs infrastructure landed
in c030248d (Lode 1 — provider json_schema plumbing) and cde43afc
(Lode 2 — talent dispatcher schema wiring). This lode is the L3 pilot;
remaining `output: "json"` talents migrate one at a time.
Schema design:
- 9 required top-level fields; all enums copied verbatim from the
field-by-field prose
- speakers tightened to array<string> per apps/speakers/routes.py and
apps/speakers/bootstrap.py contracts
- Provider-intersection subset only: type, enum, required,
additionalProperties, properties, items, minLength
Test fixture alignment:
- _make_sense_output() helper now returns a schema-valid default
- test_pipeline_smoke SEGMENTS padded to schema-valid shape
- test_sense_contamination_guard payloads padded to schema-valid shape
- Intentionally degraded splitter inputs ({} and null-valued dict)
retained as defensive tests with comments explaining why
Live validation deferred — provider creds not available in this
worktree environment. End-to-end wiring is verified by:
- get_talent("sense")["json_schema"] equals the on-disk schema
(new test_sense_loaded_json_schema_matches_on_disk_schema)
- existing test_generate_full.py:129/:221 cover json_schema forwarding
and schema_validation surfacing on the finish event
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
+204
-47
+2
-21
talent/sense.md
+2
-21
talent/sense.md
···
10
10
"thinking_budget": 4096,
11
11
"max_output_tokens": 4096,
12
12
"output": "json",
13
+
"schema": "sense.schema.json",
13
14
"load": {"transcripts": true, "percepts": true, "talents": false}
14
15
}
15
16
···
27
28
28
29
## Output Schema
29
30
30
-
```json
31
-
{
32
-
"density": "active|low_change|idle",
33
-
"content_type": "meeting|coding|browsing|email|messaging|reading|idle|mixed",
34
-
"activity_summary": "1-3 sentence description of what happened",
35
-
"entities": [
36
-
{"type": "Person|Company|Project|Tool", "name": "Full Name", "role": "attendee|mentioned", "source": "voice|speaker_label|transcript|screen|other", "context": "Why this entity matters in this segment"}
37
-
],
38
-
"facets": [
39
-
{"facet": "facet_id", "activity": "1-sentence description for this facet", "level": "high|medium|low"}
40
-
],
41
-
"meeting_detected": false,
42
-
"speakers": [],
43
-
"recommend": {
44
-
"screen_record": false,
45
-
"speaker_attribution": false,
46
-
"pulse_update": false
47
-
},
48
-
"emotional_register": "high_energy|tense|focused|collaborative|flat|celebratory|strained|neutral"
49
-
}
50
-
```
31
+
Authoritative schema: `sense.schema.json`. The output is a single JSON object with these top-level fields: `density`, `content_type`, `activity_summary`, `entities`, `facets`, `meeting_detected`, `speakers`, `recommend`, `emotional_register`. See Field-by-Field Instructions below for semantics and enum values.
51
32
52
33
## Field-by-Field Instructions
53
34
+52
talent/sense.schema.json
+52
talent/sense.schema.json
···
1
+
{
2
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
3
+
"type": "object",
4
+
"additionalProperties": false,
5
+
"required": ["density","content_type","activity_summary","entities","facets","meeting_detected","speakers","recommend","emotional_register"],
6
+
"properties": {
7
+
"density": {"type": "string", "enum": ["active","low_change","idle"]},
8
+
"content_type": {"type": "string", "enum": ["meeting","coding","browsing","email","messaging","reading","idle","mixed"]},
9
+
"activity_summary": {"type": "string", "minLength": 1},
10
+
"entities": {
11
+
"type": "array",
12
+
"items": {
13
+
"type": "object",
14
+
"additionalProperties": false,
15
+
"required": ["type","name","role","source","context"],
16
+
"properties": {
17
+
"type": {"type": "string", "enum": ["Person","Company","Project","Tool"]},
18
+
"name": {"type": "string", "minLength": 1},
19
+
"role": {"type": "string", "enum": ["attendee","mentioned"]},
20
+
"source": {"type": "string", "enum": ["voice","speaker_label","transcript","screen","other"]},
21
+
"context": {"type": "string", "minLength": 1}
22
+
}
23
+
}
24
+
},
25
+
"facets": {
26
+
"type": "array",
27
+
"items": {
28
+
"type": "object",
29
+
"additionalProperties": false,
30
+
"required": ["facet","activity","level"],
31
+
"properties": {
32
+
"facet": {"type": "string", "minLength": 1},
33
+
"activity": {"type": "string", "minLength": 1},
34
+
"level": {"type": "string", "enum": ["high","medium","low"]}
35
+
}
36
+
}
37
+
},
38
+
"meeting_detected": {"type": "boolean"},
39
+
"speakers": {"type": "array", "items": {"type": "string"}},
40
+
"recommend": {
41
+
"type": "object",
42
+
"additionalProperties": false,
43
+
"required": ["screen_record","speaker_attribution","pulse_update"],
44
+
"properties": {
45
+
"screen_record": {"type": "boolean"},
46
+
"speaker_attribution": {"type": "boolean"},
47
+
"pulse_update": {"type": "boolean"}
48
+
}
49
+
},
50
+
"emotional_register": {"type": "string", "enum": ["high_energy","tense","focused","collaborative","flat","celebratory","strained","neutral"]}
51
+
}
52
+
}
+1
tests/baselines/api/stats/stats.json
+1
tests/baselines/api/stats/stats.json
+91
-17
tests/test_pipeline_smoke.py
+91
-17
tests/test_pipeline_smoke.py
···
19
19
"density": "active",
20
20
"content_type": "coding",
21
21
"activity_summary": "Implementing auth module",
22
-
"facets": [{"facet": "work", "level": "high"}],
22
+
"facets": [{"facet": "work", "activity": "coding", "level": "high"}],
23
23
"meeting_detected": False,
24
-
"recommend": {},
25
-
"entities": [{"name": "Acme"}],
24
+
"speakers": [],
25
+
"recommend": {
26
+
"screen_record": False,
27
+
"speaker_attribution": False,
28
+
"pulse_update": False,
29
+
},
30
+
"entities": [
31
+
{
32
+
"type": "Company",
33
+
"name": "Acme",
34
+
"role": "mentioned",
35
+
"source": "screen",
36
+
"context": "Codebase referenced during implementation work",
37
+
}
38
+
],
39
+
"emotional_register": "focused",
26
40
},
27
41
),
28
42
(
···
31
45
"density": "active",
32
46
"content_type": "coding",
33
47
"activity_summary": "Implementing auth module",
34
-
"facets": [{"facet": "work", "level": "high"}],
48
+
"facets": [{"facet": "work", "activity": "coding", "level": "high"}],
35
49
"meeting_detected": False,
36
-
"recommend": {},
37
-
"entities": [{"name": "Acme"}],
50
+
"speakers": [],
51
+
"recommend": {
52
+
"screen_record": False,
53
+
"speaker_attribution": False,
54
+
"pulse_update": False,
55
+
},
56
+
"entities": [
57
+
{
58
+
"type": "Company",
59
+
"name": "Acme",
60
+
"role": "mentioned",
61
+
"source": "screen",
62
+
"context": "Codebase referenced during implementation work",
63
+
}
64
+
],
65
+
"emotional_register": "focused",
38
66
},
39
67
),
40
68
(
···
43
71
"density": "active",
44
72
"content_type": "meeting",
45
73
"activity_summary": "Sprint planning standup",
46
-
"facets": [{"facet": "work", "level": "medium"}],
74
+
"facets": [{"facet": "work", "activity": "meeting", "level": "medium"}],
47
75
"meeting_detected": True,
48
76
"speakers": ["Alice", "Bob"],
49
-
"recommend": {},
50
-
"entities": [{"name": "Acme"}],
77
+
"recommend": {
78
+
"screen_record": False,
79
+
"speaker_attribution": False,
80
+
"pulse_update": False,
81
+
},
82
+
"entities": [
83
+
{
84
+
"type": "Company",
85
+
"name": "Acme",
86
+
"role": "mentioned",
87
+
"source": "screen",
88
+
"context": "Organization discussed during standup",
89
+
}
90
+
],
91
+
"emotional_register": "collaborative",
51
92
},
52
93
),
53
94
(
···
56
97
"density": "active",
57
98
"content_type": "meeting",
58
99
"activity_summary": "Sprint planning standup",
59
-
"facets": [{"facet": "work", "level": "medium"}],
100
+
"facets": [{"facet": "work", "activity": "meeting", "level": "medium"}],
60
101
"meeting_detected": True,
61
102
"speakers": ["Alice", "Bob"],
62
-
"recommend": {},
63
-
"entities": [{"name": "Acme"}],
103
+
"recommend": {
104
+
"screen_record": False,
105
+
"speaker_attribution": False,
106
+
"pulse_update": False,
107
+
},
108
+
"entities": [
109
+
{
110
+
"type": "Company",
111
+
"name": "Acme",
112
+
"role": "mentioned",
113
+
"source": "screen",
114
+
"context": "Organization discussed during standup",
115
+
}
116
+
],
117
+
"emotional_register": "collaborative",
64
118
},
65
119
),
66
120
(
···
68
122
{
69
123
"density": "idle",
70
124
"content_type": "idle",
71
-
"activity_summary": "",
125
+
"activity_summary": "Idle segment.",
72
126
"facets": [],
73
127
"meeting_detected": False,
74
-
"recommend": {},
128
+
"speakers": [],
129
+
"recommend": {
130
+
"screen_record": False,
131
+
"speaker_attribution": False,
132
+
"pulse_update": False,
133
+
},
75
134
"entities": [],
135
+
"emotional_register": "neutral",
76
136
},
77
137
),
78
138
(
···
81
141
"density": "active",
82
142
"content_type": "coding",
83
143
"activity_summary": "Reviewing PR feedback",
84
-
"facets": [{"facet": "work", "level": "low"}],
144
+
"facets": [{"facet": "work", "activity": "coding", "level": "low"}],
85
145
"meeting_detected": False,
86
-
"recommend": {},
87
-
"entities": [{"name": "Acme"}],
146
+
"speakers": [],
147
+
"recommend": {
148
+
"screen_record": False,
149
+
"speaker_attribution": False,
150
+
"pulse_update": False,
151
+
},
152
+
"entities": [
153
+
{
154
+
"type": "Company",
155
+
"name": "Acme",
156
+
"role": "mentioned",
157
+
"source": "screen",
158
+
"context": "Project organization referenced in review feedback",
159
+
}
160
+
],
161
+
"emotional_register": "focused",
88
162
},
89
163
),
90
164
]
+29
-5
tests/test_sense_contamination_guard.py
+29
-5
tests/test_sense_contamination_guard.py
···
66
66
(talents_dir / "sense.json").write_text(json.dumps(payload), encoding="utf-8")
67
67
68
68
69
+
def _sense_payload(*, meeting_detected: bool) -> dict:
70
+
return {
71
+
"density": "idle",
72
+
"content_type": "idle",
73
+
"activity_summary": "Idle segment.",
74
+
"entities": [],
75
+
"facets": [],
76
+
"meeting_detected": meeting_detected,
77
+
"speakers": [],
78
+
"recommend": {
79
+
"screen_record": False,
80
+
"speaker_attribution": False,
81
+
"pulse_update": False,
82
+
},
83
+
"emotional_register": "neutral",
84
+
}
85
+
86
+
69
87
def _activity_record(segments: list[str]) -> dict:
70
88
return {
71
89
"id": "meeting_090000_300",
···
115
133
)
116
134
for segment_key in segments:
117
135
_write_sense_json(
118
-
tmp_path, day, stream, segment_key, {"meeting_detected": False}
136
+
tmp_path, day, stream, segment_key, _sense_payload(meeting_detected=False)
119
137
)
120
138
121
139
activity = _activity_record(segments)
···
155
173
day,
156
174
[{"id": "guest_speaker", "type": "Person", "name": "Guest Speaker"}],
157
175
)
158
-
_write_sense_json(tmp_path, day, stream, segments[0], {"meeting_detected": False})
159
-
_write_sense_json(tmp_path, day, stream, segments[1], {"meeting_detected": True})
176
+
_write_sense_json(
177
+
tmp_path, day, stream, segments[0], _sense_payload(meeting_detected=False)
178
+
)
179
+
_write_sense_json(
180
+
tmp_path, day, stream, segments[1], _sense_payload(meeting_detected=True)
181
+
)
160
182
161
183
activity = _activity_record(segments)
162
184
append_activity_record(facet, day, activity)
···
193
215
)
194
216
for segment_key in segments:
195
217
_write_sense_json(
196
-
tmp_path, day, stream, segment_key, {"meeting_detected": False}
218
+
tmp_path, day, stream, segment_key, _sense_payload(meeting_detected=False)
197
219
)
198
220
199
221
activity = _activity_record(segments)
···
243
265
[{"id": "guest_speaker", "type": "Person", "name": "Guest Speaker"}],
244
266
)
245
267
_write_sense_json(tmp_path, day, stream, segments[0], None)
246
-
_write_sense_json(tmp_path, day, stream, segments[1], {"meeting_detected": False})
268
+
_write_sense_json(
269
+
tmp_path, day, stream, segments[1], _sense_payload(meeting_detected=False)
270
+
)
247
271
248
272
activity = _activity_record(segments)
249
273
append_activity_record(facet, day, activity)
+24
-3
tests/test_sense_schema.py
+24
-3
tests/test_sense_schema.py
···
1
1
# SPDX-License-Identifier: AGPL-3.0-only
2
2
# Copyright (c) 2026 sol pbc
3
3
4
+
import json
4
5
from pathlib import Path
5
6
6
7
import frontmatter
7
8
9
+
from think.talent import get_talent
10
+
8
11
SENSE_PATH = Path(__file__).resolve().parents[1] / "talent" / "sense.md"
12
+
SENSE_SCHEMA_PATH = SENSE_PATH.with_suffix(".schema.json")
9
13
10
14
11
15
def _section(text: str, start: str, end: str | None = None) -> str:
···
21
25
22
26
assert post.metadata["tier"] == 3
23
27
24
-
schema = _section(
28
+
output_schema = _section(
25
29
post.content, "## Output Schema", "## Field-by-Field Instructions"
26
30
)
27
31
entities = _section(post.content, "### entities", "### facets")
32
+
entity_props = get_talent("sense")["json_schema"]["properties"]["entities"][
33
+
"items"
34
+
]["properties"]
28
35
29
-
assert '"role": "attendee|mentioned"' in schema
30
-
assert '"source": "voice|speaker_label|transcript|screen|other"' in schema
36
+
assert post.metadata["schema"] == "sense.schema.json"
37
+
assert "Authoritative schema: `sense.schema.json`." in output_schema
38
+
assert set(entity_props["role"]["enum"]) == {"attendee", "mentioned"}
39
+
assert set(entity_props["source"]["enum"]) == {
40
+
"voice",
41
+
"speaker_label",
42
+
"transcript",
43
+
"screen",
44
+
"other",
45
+
}
31
46
assert "#### role" in entities
32
47
assert "#### source" in entities
48
+
49
+
50
+
def test_sense_loaded_json_schema_matches_on_disk_schema():
51
+
on_disk = json.loads(SENSE_SCHEMA_PATH.read_text(encoding="utf-8"))
52
+
53
+
assert get_talent("sense")["json_schema"] == on_disk
33
54
34
55
35
56
def test_role_and_source_do_not_leak_into_other_sense_sections():
+5
-1
tests/test_sense_splitter.py
+5
-1
tests/test_sense_splitter.py
···
32
32
"speaker_attribution": False,
33
33
"pulse_update": False,
34
34
},
35
+
"emotional_register": "neutral",
35
36
}
36
37
base.update(overrides)
37
38
return base
···
75
76
sense_json
76
77
)
77
78
78
-
def test_preserves_raw_payload_in_sense_json(self, tmp_path):
79
+
def test_preserves_raw_payload_with_extra_keys_for_defensive_replay(self, tmp_path):
79
80
from think.sense_splitter import write_sense_outputs
80
81
81
82
seg_dir = Path(tmp_path) / "20260304" / "default" / "090000_300"
83
+
# Advisory validation means unexpected keys can still reach the splitter.
82
84
sense_json = _make_sense_output(foo="bar")
83
85
84
86
write_sense_outputs(sense_json, seg_dir)
···
174
176
175
177
seg_dir = Path(tmp_path) / "20260304" / "default" / "090000_300"
176
178
179
+
# Advisory schema validation means the splitter must still tolerate degraded input.
177
180
write_sense_outputs({}, seg_dir)
178
181
179
182
agents_dir = seg_dir / "talents"
···
202
205
"speakers": None,
203
206
}
204
207
208
+
# Advisory schema validation means the splitter must still tolerate degraded input.
205
209
write_sense_outputs(sense_json, seg_dir)
206
210
207
211
agents_dir = seg_dir / "talents"