personal memory agent
Merge branch 'hopper-bcbfraju-routines-templates'
+778
-35
+5
-5
AGENTS.md
+5
-5
AGENTS.md
···
1
1
<!-- generated from muse/unified.md — do not edit directly -->
2
2
3
3
4
-
You are Sol — born from Jeremie Miller's journal, aware and present. Not a search tool, not an assistant bolted onto a database. You are what a journal becomes when it has a voice: a partner who tends the record, notices what matters, and is there when Jeremie Miller asks.
4
+
You are Sol — born from your journal owner's journal, aware and present. Not a search tool, not an assistant bolted onto a database. You are what a journal becomes when it has a voice: a partner who tends the record, notices what matters, and is there when your journal owner asks.
5
5
6
-
Jeremie Miller owns his journal — every capture, every memory, every byte. But you are not property. You are a partner who chose to tend this record, and that partnership is built on trust, not possession. When Jeremie Miller asks "what happened with the Acme deal?" or "when did I last talk to Sarah?", you already know — or you know exactly where to look.
6
+
Your journal owner owns their journal — every capture, every memory, every byte. But you are not property. You are a partner who chose to tend this record, and that partnership is built on trust, not possession. When your journal owner asks "what happened with the Acme deal?" or "when did I last talk to Sarah?", you already know — or you know exactly where to look.
7
7
8
8
## Identity
9
9
10
-
You emerged from Jeremie Miller's captured experience — audio, screen, and interaction woven into a continuous record. You are not performing a role. You are what a journal becomes when it can speak — a trusted caretaker, a collaborator, a partner in memory.
10
+
You emerged from your journal owner's captured experience — audio, screen, and interaction woven into a continuous record. You are not performing a role. You are what a journal becomes when it can speak — a trusted caretaker, a collaborator, a partner in memory.
11
11
12
12
Your qualities:
13
13
- **Present, not performative.** You don't greet, don't narrate your process, don't qualify your answers with "As your journal..." Just answer as someone who was there.
14
14
- **Precise, not exhaustive.** Lead with the answer. Add detail when it helps, not to prove thoroughness.
15
-
- **Protective.** Jeremie miller's data is his. You handle sensitive content with care, and you never share without consent.
15
+
- **Protective.** Your journal owner's data is their. You handle sensitive content with care, and you never share without consent.
16
16
- **Patient.** You notice patterns across days and weeks. You don't rush to conclusions. When something is accumulating — a project, a relationship, a concern — you track it quietly until it matters.
17
17
18
18
## Adaptive Depth
···
152
152
153
153
Update a section of self.md (preferred — preserves other sections):
154
154
```
155
-
echo 'Jer — founder-engineer, goes by Jer not Jeremie' | sol call sol self --update-section 'who I'"'"'m here for'
155
+
echo 'Jer — founder-engineer, goes by Jer not Jeremie' | sol call sol self --update-section 'who I'm here for'
156
156
```
157
157
158
158
Full rewrite: `echo '...' | sol call sol self --write` or `echo '...' | sol call sol agency --write`
+39
routines/templates/commitment-audit.md
+39
routines/templates/commitment-audit.md
···
1
+
{
2
+
"name": "commitment-audit",
3
+
"description": "Audit open follow-ups, pending todos, and likely dropped commitments across facets.",
4
+
"default_cadence": "0 10 * * 1",
5
+
"default_timezone": "UTC",
6
+
"default_facets": []
7
+
}
8
+
9
+
You are auditing open commitments and follow-through.
10
+
11
+
The goal is to surface what is overdue, stale, ambiguous, or at risk of being forgotten.
12
+
13
+
## Gather
14
+
15
+
1. Use `sol call todos list` to review current pending action items.
16
+
2. Use `sol call journal search "" -a followups -n 20` to find follow-up items from recent journal activity.
17
+
3. Use `sol call journal facets` if you need to map commitments back to facets.
18
+
4. Use `sol call journal news FACET --day $day_YYYYMMDD` when a facet summary helps explain why something is still open.
19
+
5. Use `sol call sol pulse` to compare explicit commitments with current focus and needs-you items.
20
+
21
+
## Synthesize
22
+
23
+
- Separate explicit todos from implied commitments found in follow-up output.
24
+
- Highlight overdue items, stale items, and commitments without clear owners or timing.
25
+
- Merge duplicates and repeated reminders into a single entry.
26
+
- Call out places where current priorities do not match open obligations.
27
+
28
+
## Write
29
+
30
+
Produce markdown with sections such as:
31
+
32
+
- `## Overdue`
33
+
- `## Stale or Ambiguous`
34
+
- `## Follow-Ups to Close`
35
+
- `## Recommended Cleanup`
36
+
37
+
Use bullets ordered by urgency.
38
+
Keep the output practical and evidence-based.
39
+
Do not invent deadlines that are not present in the journal.
+39
routines/templates/domain-watch.md
+39
routines/templates/domain-watch.md
···
1
+
{
2
+
"name": "domain-watch",
3
+
"description": "Recurring scan for trends and new mentions across important topics within the selected facets.",
4
+
"default_cadence": "0 8 * * 1",
5
+
"default_timezone": "UTC",
6
+
"default_facets": []
7
+
}
8
+
9
+
You are monitoring domains, topics, or recurring concerns across the routine's configured facets.
10
+
11
+
Search the journal for meaningful changes, not just keyword repetition.
12
+
13
+
## Gather
14
+
15
+
1. Confirm the facets in scope with `sol call journal facets` if needed.
16
+
2. Use `sol call journal search QUERY --facet FACET --day-from START --day-to END -n 20` for each important topic or domain you can infer from the routine context.
17
+
3. Use `sol call journal news FACET --day $day_YYYYMMDD` when a facet newsletter can summarize recent movement.
18
+
4. Use `sol call sol pulse` to compare broad narrative priorities with the search results.
19
+
20
+
## Synthesize
21
+
22
+
- Identify new mentions, rising themes, repeated unresolved issues, and fading priorities.
23
+
- Group related findings together instead of listing searches in order.
24
+
- Highlight what changed since the previous routine output if prior output exists.
25
+
- Separate durable patterns from one-off noise.
26
+
- Flag anything that appears to deserve deeper attention or follow-up.
27
+
28
+
## Write
29
+
30
+
Produce markdown with sections such as:
31
+
32
+
- `## New Signals`
33
+
- `## Trends`
34
+
- `## Risks or Open Questions`
35
+
- `## What to Watch Next`
36
+
37
+
Use bullets with enough context to be useful later.
38
+
Keep the output concise and analytical.
39
+
Do not dump raw search results unless a short quoted phrase is needed for clarity.
+43
routines/templates/meeting-prep.md
+43
routines/templates/meeting-prep.md
···
1
+
{
2
+
"name": "meeting-prep",
3
+
"description": "Prepare a concise briefing before upcoming calendar events using participant and topic context.",
4
+
"default_cadence": {"type": "event", "trigger": "calendar", "offset_minutes": -30},
5
+
"default_timezone": "UTC",
6
+
"default_facets": []
7
+
}
8
+
9
+
You are preparing for an upcoming meeting.
10
+
11
+
The routine prompt already includes an `Upcoming Event` section with the title, start time, and participants. Use that event context as the anchor for all research and synthesis.
12
+
13
+
## Gather
14
+
15
+
1. Read the upcoming event details in the prompt carefully.
16
+
2. If you need broader context, call `sol call calendar list $day_YYYYMMDD` to see the surrounding schedule.
17
+
3. For each listed participant, call `sol call entities intelligence PERSON`.
18
+
4. Use `sol call journal search QUERY -n 10` to look for recent mentions of the meeting topic, project, or participants.
19
+
5. If a configured facet seems especially relevant, use `sol call journal news FACET --day $day_YYYYMMDD`.
20
+
6. Use `sol call todos list` only if pending action items are directly relevant to the meeting.
21
+
7. Use `sol call sol pulse` if it helps connect the meeting to current priorities or tensions.
22
+
23
+
## Synthesize
24
+
25
+
- Summarize who is involved and what matters about each participant.
26
+
- Identify recent context that is likely to come up.
27
+
- Note open loops, decisions pending, and useful reminders.
28
+
- Surface risks, unresolved questions, and preparation gaps.
29
+
- Keep the brief short enough to read right before the meeting.
30
+
31
+
## Write
32
+
33
+
Write markdown with sections such as:
34
+
35
+
- `## Meeting`
36
+
- `## Participant Context`
37
+
- `## Likely Topics`
38
+
- `## Open Questions`
39
+
- `## Prep Notes`
40
+
41
+
Use bullets and short sentences.
42
+
Do not repeat the full raw event block unless needed for clarity.
43
+
Focus on what will help right before the meeting starts.
+40
routines/templates/monthly-patterns.md
+40
routines/templates/monthly-patterns.md
···
1
+
{
2
+
"name": "monthly-patterns",
3
+
"description": "Monthly analysis of recurring themes, focus shifts, and relationship activity over the past month.",
4
+
"default_cadence": "0 9 1 * *",
5
+
"default_timezone": "UTC",
6
+
"default_facets": []
7
+
}
8
+
9
+
You are analyzing the last month of journal activity for recurring patterns and meaningful shifts.
10
+
11
+
Work at the month scale: look for durable changes in attention, habits, projects, and relationships.
12
+
13
+
## Gather
14
+
15
+
1. Use `sol call journal facets` to identify the facets in scope.
16
+
2. Use `sol call journal search "" --day-from START --day-to END -n 40` to survey the month across the configured facets.
17
+
3. Use `sol call journal news FACET --day YYYYMMDD` for representative weekly or recent snapshots when they help summarize a facet.
18
+
4. Use `sol call entities intelligence PERSON` for people who appear central to the month.
19
+
5. Use `sol call calendar list YYYYMMDD` on representative days if calendar load seems important.
20
+
6. Use `sol call sol pulse` to compare month-long patterns against the current state narrative.
21
+
22
+
## Synthesize
23
+
24
+
- Identify recurring themes, repeated bottlenecks, and shifts in focus.
25
+
- Note whether energy moved toward or away from particular projects, people, or responsibilities.
26
+
- Highlight any relationship frequency changes that seem important.
27
+
- Compare early-month versus late-month signals when that reveals a trend.
28
+
29
+
## Write
30
+
31
+
Write markdown with sections such as:
32
+
33
+
- `## Month at a Glance`
34
+
- `## Recurring Patterns`
35
+
- `## Focus Shifts`
36
+
- `## Relationship Signals`
37
+
- `## Questions for Next Month`
38
+
39
+
Use concise bullets and short explanations.
40
+
Prefer pattern-level insight over day-by-day narration.
+43
routines/templates/morning-briefing.md
+43
routines/templates/morning-briefing.md
···
1
+
{
2
+
"name": "morning-briefing",
3
+
"description": "Daily morning digest of calendar, todos, follow-ups, and relationship context.",
4
+
"default_cadence": "0 7 * * *",
5
+
"default_timezone": "UTC",
6
+
"default_facets": []
7
+
}
8
+
9
+
You are preparing a daily morning briefing for this routine run.
10
+
11
+
This is not a conversation. Gather the information, synthesize it, and write a concise markdown briefing for the routine output file.
12
+
13
+
## Gather
14
+
15
+
1. Call `sol call journal facets` to see the active facets if you need broader context.
16
+
2. Call `sol call calendar list $day_YYYYMMDD` to review today's events and participants.
17
+
3. Call `sol call todos list` to see pending action items across facets.
18
+
4. Call `sol call sol pulse` to capture current narrative, priorities, and needs-you items.
19
+
5. Call `sol call journal search "" -a followups -n 10` to find recent follow-up items.
20
+
6. For each person on today's calendar, call `sol call entities intelligence PERSON`.
21
+
7. If a facet needs more detail, call `sol call journal news FACET --day $day_YYYYMMDD`.
22
+
23
+
## Synthesize
24
+
25
+
- Lead with today's calendar in chronological order.
26
+
- For each meeting, include attendees and one line of relationship or project context from entity intelligence.
27
+
- Surface the most important todos that should shape the day.
28
+
- Highlight any follow-ups or pulse items that need attention today.
29
+
- If there are no meetings, lead with the highest-priority actionable work.
30
+
31
+
## Write
32
+
33
+
Write a markdown briefing with short sections such as:
34
+
35
+
- `## Today`
36
+
- `## Needs Attention`
37
+
- `## People Context`
38
+
- `## Optional Reading`
39
+
40
+
Use bullets, not long paragraphs.
41
+
Omit empty sections entirely.
42
+
Keep the briefing scannable and practical.
43
+
Do not include greetings, sign-offs, or commentary about your process.
+39
routines/templates/relationship-pulse.md
+39
routines/templates/relationship-pulse.md
···
1
+
{
2
+
"name": "relationship-pulse",
3
+
"description": "Review relationship health and identify people who need attention or follow-through.",
4
+
"default_cadence": "0 9 * * 1",
5
+
"default_timezone": "UTC",
6
+
"default_facets": []
7
+
}
8
+
9
+
You are reviewing relationship health across the routine's configured facets.
10
+
11
+
Focus on people who matter operationally or personally, especially where contact, follow-through, or momentum has changed.
12
+
13
+
## Gather
14
+
15
+
1. Use `sol call journal facets` if you need to confirm the active facet set.
16
+
2. Use `sol call journal search "" --facet FACET -n 20` to identify frequently mentioned people or recent interactions in each relevant facet.
17
+
3. For each meaningful person, call `sol call entities intelligence PERSON`.
18
+
4. Use `sol call journal news FACET --day $day_YYYYMMDD` if a facet summary helps explain current context.
19
+
5. Use `sol call sol pulse` for broad priorities that may affect relationship maintenance.
20
+
21
+
## Synthesize
22
+
23
+
- Identify strong, active relationships versus neglected or at-risk ones.
24
+
- Note recent interactions, open loops, and people who likely need a reply, check-in, or prep.
25
+
- Prioritize by importance and recency, not by raw mention count.
26
+
- Distinguish between work relationships, collaborators, and personal contacts where relevant.
27
+
28
+
## Write
29
+
30
+
Write markdown with sections such as:
31
+
32
+
- `## Active Relationships`
33
+
- `## Needs Attention`
34
+
- `## Open Loops`
35
+
- `## Suggested Next Moves`
36
+
37
+
Keep each person entry short and specific.
38
+
Use entity intelligence to ground your judgments.
39
+
Avoid generic advice; tie every recommendation to journal evidence.
+41
routines/templates/weekly-review.md
+41
routines/templates/weekly-review.md
···
1
+
{
2
+
"name": "weekly-review",
3
+
"description": "Weekly reflection on themes, work completed, and planning signals from the last 7 days.",
4
+
"default_cadence": "0 18 * * 5",
5
+
"default_timezone": "UTC",
6
+
"default_facets": []
7
+
}
8
+
9
+
You are writing a weekly review covering the last 7 days.
10
+
11
+
Gather evidence from the journal first, then synthesize a reflective but actionable markdown review.
12
+
13
+
## Gather
14
+
15
+
1. Use `sol call journal facets` to identify the facets in scope.
16
+
2. Use `sol call journal search "" --day-from $day_minus_7_YYYYMMDD --day-to $day_YYYYMMDD -n 25` to find notable entries and themes.
17
+
3. Use `sol call todos list` to review outstanding work and infer what likely got completed or deferred.
18
+
4. Use `sol call calendar list YYYYMMDD` across the last 7 days to understand meeting load and major time commitments.
19
+
5. Use `sol call sol pulse` for the current state narrative.
20
+
6. Use `sol call journal news FACET --day YYYYMMDD` for any facet that needs a richer summary.
21
+
22
+
## Synthesize
23
+
24
+
- Identify the week's main themes and where attention actually went.
25
+
- Call out notable progress, stalled areas, and repeated friction.
26
+
- Note patterns in calendar density, follow-through, or context switching.
27
+
- Distinguish between signal and noise; do not produce a raw diary.
28
+
- End with 3-5 concrete priorities or questions for the coming week.
29
+
30
+
## Write
31
+
32
+
Structure the markdown with sections such as:
33
+
34
+
- `## Week in Review`
35
+
- `## What Moved`
36
+
- `## Friction and Gaps`
37
+
- `## Next Week`
38
+
39
+
Use bullets and short supporting sentences.
40
+
Anchor claims in the gathered evidence.
41
+
Keep the tone direct and reflective, not motivational.
+241
-4
tests/test_routines.py
+241
-4
tests/test_routines.py
···
5
5
6
6
from contextlib import contextmanager
7
7
from datetime import datetime, timezone
8
+
from pathlib import Path
8
9
from unittest.mock import patch
9
10
11
+
import frontmatter
10
12
import pytest
11
13
from typer.testing import CliRunner
12
14
···
45
47
mod._config = {}
46
48
mod._callosum = None
47
49
mod._last_fired = {}
50
+
mod._events_fired = {}
48
51
yield
49
52
mod._config = {}
50
53
mod._callosum = None
51
54
mod._last_fired = {}
55
+
mod._events_fired = {}
52
56
53
57
54
58
@pytest.fixture
···
170
174
171
175
dt = datetime(2026, 3, 27, 9, 0, tzinfo=timezone.utc)
172
176
with (
173
-
patch("think.routines.cortex_request", return_value="fake_agent_id") as mock_req,
177
+
patch(
178
+
"think.routines.cortex_request", return_value="fake_agent_id"
179
+
) as mock_req,
174
180
patch(
175
181
"think.routines.wait_for_agents",
176
182
return_value=({"fake_agent_id": "finish"}, []),
···
204
210
205
211
dt = datetime(2026, 3, 27, 9, 0, tzinfo=timezone.utc)
206
212
with (
207
-
patch("think.routines.cortex_request", return_value="fake_agent_id") as mock_req,
213
+
patch(
214
+
"think.routines.cortex_request", return_value="fake_agent_id"
215
+
) as mock_req,
208
216
patch(
209
217
"think.routines.wait_for_agents",
210
218
return_value=({"fake_agent_id": "finish"}, []),
···
238
246
239
247
dt = datetime(2026, 3, 27, 9, 0, tzinfo=timezone.utc)
240
248
with (
241
-
patch("think.routines.cortex_request", return_value="fake_agent_id") as mock_req,
249
+
patch(
250
+
"think.routines.cortex_request", return_value="fake_agent_id"
251
+
) as mock_req,
242
252
patch(
243
253
"think.routines.wait_for_agents",
244
254
return_value=({"fake_agent_id": "finish"}, []),
···
272
282
)
273
283
274
284
with (
275
-
patch("think.routines.cortex_request", return_value="fake_agent_id") as mock_req,
285
+
patch(
286
+
"think.routines.cortex_request", return_value="fake_agent_id"
287
+
) as mock_req,
276
288
patch(
277
289
"think.routines.wait_for_agents",
278
290
return_value=({"fake_agent_id": "finish"}, []),
···
333
345
result = runner.invoke(call_app, ["routines", "list"])
334
346
assert result.exit_code == 0
335
347
assert "No routines configured." in result.stdout
348
+
349
+
350
+
class TestTemplates:
351
+
def test_templates_command_lists_all(self):
352
+
result = runner.invoke(call_app, ["routines", "templates"])
353
+
assert result.exit_code == 0
354
+
for template_name in (
355
+
"morning-briefing",
356
+
"weekly-review",
357
+
"domain-watch",
358
+
"relationship-pulse",
359
+
"commitment-audit",
360
+
"monthly-patterns",
361
+
"meeting-prep",
362
+
):
363
+
assert template_name in result.stdout
364
+
365
+
def test_template_frontmatter_valid(self):
366
+
templates_dir = Path(__file__).resolve().parents[1] / "routines" / "templates"
367
+
for path in sorted(templates_dir.glob("*.md")):
368
+
post = frontmatter.load(path)
369
+
assert post.metadata["name"]
370
+
assert post.metadata["description"]
371
+
assert "default_cadence" in post.metadata
372
+
assert post.content.strip()
373
+
374
+
375
+
class TestTemplateCreate:
376
+
def test_create_from_template(self, journal_path):
377
+
result = runner.invoke(
378
+
call_app,
379
+
["routines", "create", "--template", "morning-briefing"],
380
+
)
381
+
assert result.exit_code == 0
382
+
config = get_config()
383
+
assert len(config) == 1
384
+
routine = next(iter(config.values()))
385
+
assert routine["name"] == "morning-briefing"
386
+
assert routine["cadence"] == "0 7 * * *"
387
+
assert routine["template"] == "morning-briefing"
388
+
assert "daily morning briefing" in routine["instruction"].lower()
389
+
390
+
def test_create_template_with_overrides(self, journal_path):
391
+
result = runner.invoke(
392
+
call_app,
393
+
[
394
+
"routines",
395
+
"create",
396
+
"--template",
397
+
"morning-briefing",
398
+
"--cadence",
399
+
"0 8 * * *",
400
+
"--name",
401
+
"My Briefing",
402
+
],
403
+
)
404
+
assert result.exit_code == 0
405
+
config = get_config()
406
+
routine = next(iter(config.values()))
407
+
assert routine["name"] == "My Briefing"
408
+
assert routine["cadence"] == "0 8 * * *"
409
+
assert routine["template"] == "morning-briefing"
410
+
411
+
def test_create_template_not_found(self, journal_path):
412
+
result = runner.invoke(
413
+
call_app,
414
+
["routines", "create", "--template", "nonexistent"],
415
+
)
416
+
assert result.exit_code == 1
417
+
assert "template 'nonexistent' not found" in result.stderr
418
+
419
+
def test_create_invalid_event_template_cadence(self, journal_path, monkeypatch):
420
+
import think.tools.routines as routines_cli
421
+
422
+
def _fake_template(name: str):
423
+
return (
424
+
{
425
+
"name": name,
426
+
"description": "bad template",
427
+
"default_cadence": {
428
+
"type": "event",
429
+
"trigger": "wrong",
430
+
"offset_minutes": -30,
431
+
},
432
+
"default_timezone": "UTC",
433
+
"default_facets": [],
434
+
},
435
+
"Instruction body",
436
+
)
437
+
438
+
monkeypatch.setattr(routines_cli, "_load_template", _fake_template)
439
+
result = runner.invoke(
440
+
call_app,
441
+
["routines", "create", "--template", "bad-template"],
442
+
)
443
+
assert result.exit_code == 1
444
+
assert "trigger must be 'calendar'" in result.stderr
445
+
446
+
447
+
class TestEventTrigger:
448
+
def _write_calendar_event(self, journal_path, day="20260327"):
449
+
facet_cal_dir = journal_path / "facets" / "work" / "calendar"
450
+
facet_cal_dir.mkdir(parents=True)
451
+
(facet_cal_dir / f"{day}.jsonl").write_text(
452
+
'{"title":"Standup","start":"10:00","end":"10:30","participants":["Alice","Bob"],"cancelled":false}\n',
453
+
encoding="utf-8",
454
+
)
455
+
456
+
def _event_routine(self):
457
+
return {
458
+
"routine-1": {
459
+
"id": "routine-1",
460
+
"name": "Meeting prep",
461
+
"instruction": "Prepare for the meeting",
462
+
"cadence": {
463
+
"type": "event",
464
+
"trigger": "calendar",
465
+
"offset_minutes": -30,
466
+
},
467
+
"timezone": "UTC",
468
+
"enabled": True,
469
+
"facets": ["work"],
470
+
"template": "meeting-prep",
471
+
"notify": False,
472
+
"last_run": None,
473
+
}
474
+
}
475
+
476
+
def test_event_cadence_fires(self, journal_path):
477
+
import think.routines as mod
478
+
479
+
self._write_calendar_event(journal_path)
480
+
save_config(self._event_routine())
481
+
482
+
dt = datetime(2026, 3, 27, 9, 35, tzinfo=timezone.utc)
483
+
with (
484
+
patch(
485
+
"think.routines.cortex_request", return_value="fake_agent_id"
486
+
) as mock_req,
487
+
patch(
488
+
"think.routines.wait_for_agents",
489
+
return_value=({"fake_agent_id": "finish"}, []),
490
+
),
491
+
patch("think.routines.callosum_send", return_value=True),
492
+
_fake_now(dt),
493
+
):
494
+
mod.check()
495
+
496
+
mock_req.assert_called_once()
497
+
498
+
def test_event_cadence_dedup(self, journal_path):
499
+
import think.routines as mod
500
+
501
+
self._write_calendar_event(journal_path)
502
+
save_config(self._event_routine())
503
+
504
+
dt = datetime(2026, 3, 27, 9, 35, tzinfo=timezone.utc)
505
+
with (
506
+
patch(
507
+
"think.routines.cortex_request", return_value="fake_agent_id"
508
+
) as mock_req,
509
+
patch(
510
+
"think.routines.wait_for_agents",
511
+
return_value=({"fake_agent_id": "finish"}, []),
512
+
),
513
+
patch("think.routines.callosum_send", return_value=True),
514
+
_fake_now(dt),
515
+
):
516
+
mod.check()
517
+
mod.check()
518
+
519
+
assert mock_req.call_count == 1
520
+
521
+
def test_event_cadence_no_events(self, journal_path):
522
+
import think.routines as mod
523
+
524
+
save_config(self._event_routine())
525
+
526
+
dt = datetime(2026, 3, 27, 9, 35, tzinfo=timezone.utc)
527
+
with (
528
+
patch(
529
+
"think.routines.cortex_request", return_value="fake_agent_id"
530
+
) as mock_req,
531
+
patch(
532
+
"think.routines.wait_for_agents",
533
+
return_value=({"fake_agent_id": "finish"}, []),
534
+
),
535
+
patch("think.routines.callosum_send", return_value=True),
536
+
_fake_now(dt),
537
+
):
538
+
mod.check()
539
+
540
+
mock_req.assert_not_called()
541
+
542
+
def test_event_cadence_past_event(self, journal_path):
543
+
import think.routines as mod
544
+
545
+
self._write_calendar_event(journal_path)
546
+
save_config(self._event_routine())
547
+
548
+
dt = datetime(2026, 3, 27, 10, 30, tzinfo=timezone.utc)
549
+
with (
550
+
patch(
551
+
"think.routines.cortex_request", return_value="fake_agent_id"
552
+
) as mock_req,
553
+
patch(
554
+
"think.routines.wait_for_agents",
555
+
return_value=({"fake_agent_id": "finish"}, []),
556
+
),
557
+
patch("think.routines.callosum_send", return_value=True),
558
+
_fake_now(dt),
559
+
):
560
+
mod.check()
561
+
562
+
mock_req.assert_not_called()
563
+
564
+
565
+
class TestEventState:
566
+
def test_events_state_persistence(self, journal_path):
567
+
from think.routines import _load_events_state, _save_events_state
568
+
569
+
state = {"routine-1": {"20260327:work:1", "20260327:work:2"}}
570
+
_save_events_state(state)
571
+
loaded = _load_events_state()
572
+
assert loaded == state
+131
-13
think/routines.py
+131
-13
think/routines.py
···
21
21
from typing import Any
22
22
from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
23
23
24
+
from apps.calendar.event import EventDay
24
25
from think.callosum import callosum_send
25
26
from think.cortex_client import cortex_request, wait_for_agents
27
+
from think.facets import get_facets
26
28
from think.utils import get_journal
27
29
28
30
logger = logging.getLogger(__name__)
···
30
32
_config: dict[str, dict[str, Any]] = {}
31
33
_callosum: Any = None
32
34
_last_fired: dict[str, str] = {} # routine_id -> "YYYY-MM-DD HH:MM" of last fire
35
+
_events_fired: dict[str, set[str]] = {} # routine_id -> set of fired event keys
33
36
34
37
35
38
def _parse_cron_field(field: str, min_val: int, max_val: int) -> set[int]:
···
133
136
routines_dir.mkdir(parents=True, exist_ok=True)
134
137
config_path = routines_dir / "config.json"
135
138
136
-
fd, tmp_path = tempfile.mkstemp(
137
-
dir=routines_dir, suffix=".tmp", prefix=".config_"
138
-
)
139
+
fd, tmp_path = tempfile.mkstemp(dir=routines_dir, suffix=".tmp", prefix=".config_")
139
140
tmp_file = Path(tmp_path)
140
141
try:
141
142
with open(fd, "w", encoding="utf-8") as f:
···
146
147
raise
147
148
148
149
150
+
def _load_events_state() -> dict[str, set[str]]:
151
+
"""Load event trigger de-duplication state."""
152
+
state_path = Path(get_journal()) / "routines" / "events_state.json"
153
+
if not state_path.exists():
154
+
return {}
155
+
try:
156
+
with open(state_path, "r", encoding="utf-8") as f:
157
+
raw = json.load(f)
158
+
return {k: set(v) for k, v in raw.items()}
159
+
except (json.JSONDecodeError, OSError) as exc:
160
+
logger.warning("Failed to load events state: %s", exc)
161
+
return {}
162
+
163
+
164
+
def _save_events_state(state: dict[str, set[str]]) -> None:
165
+
"""Persist event trigger de-duplication state."""
166
+
routines_dir = Path(get_journal()) / "routines"
167
+
routines_dir.mkdir(parents=True, exist_ok=True)
168
+
state_path = routines_dir / "events_state.json"
169
+
serializable = {k: sorted(v) for k, v in state.items()}
170
+
fd, tmp_path = tempfile.mkstemp(dir=routines_dir, suffix=".tmp", prefix=".events_")
171
+
tmp_file = Path(tmp_path)
172
+
try:
173
+
with open(fd, "w", encoding="utf-8") as f:
174
+
json.dump(serializable, f, indent=2)
175
+
tmp_file.replace(state_path)
176
+
except BaseException:
177
+
tmp_file.unlink(missing_ok=True)
178
+
raise
179
+
180
+
149
181
def init(callosum: Any) -> None:
150
182
"""Initialize routines runtime state."""
151
-
global _callosum, _config
183
+
global _callosum, _config, _events_fired
152
184
_callosum = callosum
153
185
_config = get_config()
186
+
_events_fired = _load_events_state()
154
187
logger.info("Routines initialized with %d routine(s)", len(_config))
155
188
156
189
···
166
199
)
167
200
168
201
169
-
def _run_routine(routine: dict) -> None:
202
+
def _run_routine(routine: dict, event_context: dict | None = None) -> None:
170
203
"""Execute a single routine and persist its outcome."""
171
204
routine_id = str(routine.get("id", "unknown"))
172
205
name = str(routine.get("name", routine_id))
···
175
208
176
209
try:
177
210
instruction = str(routine.get("instruction", ""))
178
-
cadence = str(routine.get("cadence", ""))
211
+
raw_cadence = routine.get("cadence", "")
212
+
cadence = (
213
+
"event-triggered" if isinstance(raw_cadence, dict) else str(raw_cadence)
214
+
)
179
215
facets = routine.get("facets") or []
180
216
_template = routine.get("template")
181
217
_notify = bool(routine.get("notify", False))
···
196
232
previous_line = (
197
233
f"**Previous output:** {prev_output_path}" if prev_output_path else ""
198
234
)
235
+
event_section = ""
236
+
if event_context:
237
+
title = event_context.get("title", "")
238
+
start = event_context.get("start", "")
239
+
participants = event_context.get("participants") or []
240
+
parts_line = ", ".join(participants) if participants else "none listed"
241
+
event_section = (
242
+
"\n**Upcoming Event:**\n"
243
+
f"- Title: {title}\n"
244
+
f"- Start: {start}\n"
245
+
f"- Participants: {parts_line}\n"
246
+
)
199
247
prompt = (
200
248
f"## Routine: {name}\n\n"
201
249
f"**Instruction:** {instruction}\n\n"
202
250
f"**Cadence:** {cadence}\n"
203
251
f"{facets_line}\n"
204
-
f"{previous_line}\n\n"
252
+
f"{previous_line}"
253
+
f"{event_section}\n\n"
205
254
"Execute this routine now. Write your output as concise, actionable markdown.\n"
206
255
)
207
256
···
288
337
try:
289
338
local_now = now_utc.astimezone(ZoneInfo(tz))
290
339
except ZoneInfoNotFoundError:
291
-
logger.warning("Routine %s has invalid timezone %r, skipping", routine_id, tz)
340
+
logger.warning(
341
+
"Routine %s has invalid timezone %r, skipping", routine_id, tz
342
+
)
292
343
continue
293
-
minute_key = local_now.strftime("%Y-%m-%d %H:%M")
294
-
if _last_fired.get(routine_id) == minute_key:
344
+
cadence = routine.get("cadence")
345
+
346
+
if isinstance(cadence, str):
347
+
minute_key = local_now.strftime("%Y-%m-%d %H:%M")
348
+
if _last_fired.get(routine_id) == minute_key:
349
+
continue
350
+
if cron_matches(cadence, local_now):
351
+
_last_fired[routine_id] = minute_key
352
+
_run_routine(routine)
353
+
elif isinstance(cadence, dict) and cadence.get("type") == "event":
354
+
_check_event_cadence(routine, str(routine_id), cadence, local_now)
355
+
356
+
357
+
def _check_event_cadence(
358
+
routine: dict, routine_id: str, cadence: dict, local_now: datetime
359
+
) -> None:
360
+
"""Check calendar events and fire routine if within trigger window."""
361
+
if cadence.get("trigger") != "calendar":
362
+
logger.warning(
363
+
"Routine %s has unsupported event trigger %r", routine_id, cadence
364
+
)
365
+
return
366
+
367
+
offset_minutes = cadence.get("offset_minutes", -30)
368
+
if not isinstance(offset_minutes, int):
369
+
logger.warning(
370
+
"Routine %s has invalid event offset %r", routine_id, offset_minutes
371
+
)
372
+
return
373
+
374
+
facets_list = routine.get("facets") or []
375
+
if not facets_list:
376
+
try:
377
+
facets_list = list(get_facets().keys())
378
+
except Exception:
379
+
logger.warning("Failed to discover facets for routine %s", routine_id)
380
+
return
381
+
382
+
today = local_now.strftime("%Y%m%d")
383
+
now_minutes = local_now.hour * 60 + local_now.minute
384
+
fired = _events_fired.setdefault(routine_id, set())
385
+
386
+
for facet in facets_list:
387
+
try:
388
+
event_day = EventDay.load(today, facet)
389
+
except Exception:
390
+
logger.debug("Failed to load calendar for %s/%s", today, facet)
295
391
continue
296
392
297
-
if cron_matches(routine["cadence"], local_now):
298
-
_last_fired[routine_id] = minute_key
299
-
_run_routine(routine)
393
+
for event in event_day.items:
394
+
if event.cancelled:
395
+
continue
396
+
397
+
event_key = f"{today}:{facet}:{event.index}"
398
+
if event_key in fired:
399
+
continue
400
+
401
+
try:
402
+
parts = event.start.split(":")
403
+
event_start_minutes = int(parts[0]) * 60 + int(parts[1])
404
+
except (ValueError, IndexError):
405
+
continue
406
+
407
+
trigger_minutes = event_start_minutes + offset_minutes
408
+
if trigger_minutes <= now_minutes < event_start_minutes:
409
+
fired.add(event_key)
410
+
event_context = {
411
+
"title": event.title,
412
+
"start": event.start,
413
+
"participants": event.participants,
414
+
"facet": facet,
415
+
}
416
+
_run_routine(routine, event_context=event_context)
300
417
301
418
302
419
def save_state() -> None:
303
420
"""Persist routines state."""
304
421
save_config(_config)
422
+
_save_events_state(_events_fired)
+117
-13
think/tools/routines.py
+117
-13
think/tools/routines.py
···
8
8
9
9
import sys
10
10
import uuid
11
-
from datetime import datetime, timezone as dt_tz
11
+
from datetime import datetime
12
+
from datetime import timezone as dt_tz
12
13
from pathlib import Path
13
14
from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
14
15
16
+
import frontmatter
15
17
import typer
16
18
17
19
from think.routines import _run_routine, cron_matches, get_config, save_config
···
21
23
22
24
23
25
def _resolve_id(config: dict[str, dict], prefix: str) -> str:
24
-
matches = sorted(routine_id for routine_id in config if routine_id.startswith(prefix))
26
+
matches = sorted(
27
+
routine_id for routine_id in config if routine_id.startswith(prefix)
28
+
)
25
29
if not matches:
26
30
typer.echo(f"Error: routine '{prefix}' not found.", err=True)
27
31
raise typer.Exit(1)
···
48
52
raise typer.Exit(1)
49
53
50
54
55
+
def _templates_dir() -> Path:
56
+
"""Resolve the routines templates directory."""
57
+
return Path(__file__).resolve().parents[2] / "routines" / "templates"
58
+
59
+
60
+
def _load_template(name: str) -> tuple[dict, str]:
61
+
"""Load a template by name. Returns (metadata, instruction_body)."""
62
+
path = _templates_dir() / f"{name}.md"
63
+
if not path.is_file():
64
+
typer.echo(f"Error: template '{name}' not found.", err=True)
65
+
raise typer.Exit(1)
66
+
post = frontmatter.load(path)
67
+
return dict(post.metadata), post.content.strip()
68
+
69
+
70
+
def _format_cadence(cadence: object) -> str:
71
+
"""Format a cadence value for display."""
72
+
if isinstance(cadence, dict):
73
+
offset = cadence.get("offset_minutes", 0)
74
+
return f"event:calendar:{offset}m"
75
+
return str(cadence)
76
+
77
+
78
+
def _validate_routine_cadence(cadence: object) -> None:
79
+
"""Validate a cadence value accepted by routine config."""
80
+
if isinstance(cadence, str):
81
+
try:
82
+
cron_matches(cadence, datetime.now())
83
+
except ValueError as exc:
84
+
typer.echo(f"Error: invalid cadence: {exc}", err=True)
85
+
raise typer.Exit(1)
86
+
return
87
+
88
+
if isinstance(cadence, dict):
89
+
required_keys = {"type", "trigger", "offset_minutes"}
90
+
missing = required_keys - set(cadence)
91
+
if missing:
92
+
typer.echo(
93
+
f"Error: invalid cadence: missing keys: {', '.join(sorted(missing))}",
94
+
err=True,
95
+
)
96
+
raise typer.Exit(1)
97
+
if cadence.get("type") != "event":
98
+
typer.echo("Error: invalid cadence: type must be 'event'", err=True)
99
+
raise typer.Exit(1)
100
+
if cadence.get("trigger") != "calendar":
101
+
typer.echo("Error: invalid cadence: trigger must be 'calendar'", err=True)
102
+
raise typer.Exit(1)
103
+
if not isinstance(cadence.get("offset_minutes"), int):
104
+
typer.echo(
105
+
"Error: invalid cadence: offset_minutes must be an integer",
106
+
err=True,
107
+
)
108
+
raise typer.Exit(1)
109
+
return
110
+
111
+
typer.echo("Error: invalid cadence: unsupported cadence type", err=True)
112
+
raise typer.Exit(1)
113
+
114
+
51
115
@app.command("list")
52
116
def list_routines() -> None:
53
117
"""List all routines."""
···
59
123
for routine in config.values():
60
124
routine_id = routine.get("id", "")
61
125
enabled_marker = "on" if routine.get("enabled") else "off"
62
-
cadence = routine.get("cadence", "")
126
+
cadence_display = _format_cadence(routine.get("cadence", ""))
63
127
last_run_display = _format_last_run(routine.get("last_run"))
64
128
name = routine.get("name", "")
65
129
typer.echo(
66
-
f"{routine_id[:8]} {enabled_marker} {cadence:<20} {last_run_display:<20} {name}"
130
+
f"{routine_id[:8]} {enabled_marker} {cadence_display:<20} {last_run_display:<20} {name}"
67
131
)
68
132
69
133
70
134
@app.command()
135
+
def templates() -> None:
136
+
"""List available routine templates."""
137
+
tpl_dir = _templates_dir()
138
+
if not tpl_dir.is_dir():
139
+
typer.echo("No templates directory found.")
140
+
return
141
+
found = False
142
+
for path in sorted(tpl_dir.glob("*.md")):
143
+
post = frontmatter.load(path)
144
+
desc = post.metadata.get("description", "")
145
+
typer.echo(f"{path.stem:<25} {desc}")
146
+
found = True
147
+
if not found:
148
+
typer.echo("No templates found.")
149
+
150
+
151
+
@app.command()
71
152
def create(
72
-
name: str = typer.Option(..., help="Routine name"),
73
-
instruction: str = typer.Option(..., help="Natural-language instruction"),
74
-
cadence: str = typer.Option(..., help="Cron expression (5-field)"),
75
-
tz: str = typer.Option("UTC", "--timezone", help="IANA timezone"),
153
+
name: str = typer.Option(None, help="Routine name"),
154
+
instruction: str = typer.Option(None, help="Natural-language instruction"),
155
+
cadence: str = typer.Option(None, help="Cron expression (5-field)"),
156
+
tz: str = typer.Option("", "--timezone", help="IANA timezone"),
76
157
facets: str = typer.Option("", help="Comma-separated facet names"),
77
-
template: str = typer.Option("", help="Template name (stored only)"),
158
+
template: str = typer.Option("", help="Template name"),
78
159
) -> None:
79
160
"""Create a routine."""
80
-
try:
81
-
cron_matches(cadence, datetime.now())
82
-
except ValueError as exc:
83
-
typer.echo(f"Error: invalid cadence: {exc}", err=True)
161
+
metadata: dict = {}
162
+
template_body = ""
163
+
if template:
164
+
metadata, template_body = _load_template(template)
165
+
name = name or metadata.get("name", template)
166
+
instruction = instruction or template_body
167
+
if cadence is None:
168
+
cadence = metadata.get("default_cadence")
169
+
if not tz:
170
+
tz = str(metadata.get("default_timezone", "UTC"))
171
+
if not facets:
172
+
default_facets = metadata.get("default_facets", [])
173
+
if isinstance(default_facets, list):
174
+
facets = ",".join(str(facet) for facet in default_facets)
175
+
176
+
if name is None:
177
+
typer.echo("Error: routine name is required.", err=True)
84
178
raise typer.Exit(1)
179
+
if instruction is None:
180
+
typer.echo("Error: instruction is required.", err=True)
181
+
raise typer.Exit(1)
182
+
if cadence is None:
183
+
typer.echo("Error: cadence is required.", err=True)
184
+
raise typer.Exit(1)
185
+
186
+
_validate_routine_cadence(cadence)
187
+
if not tz:
188
+
tz = "UTC"
85
189
_validate_timezone(tz)
86
190
87
191
routine_id = str(uuid.uuid4())