Skip to content

Sonny: the lookout forwards everything and judges nothing #310

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lydakis merged 122 commits into from
Oct 29, 2026
Merged

Sonny: the lookout forwards everything and judges nothing #310

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
122 commits
Select commit Hold shift + click to select a range
9eb1557
Sonny: a lookout that wakes the lead when its tasks need it
claude Oct 26, 2026
4d26f37
Sonny: lookout-replay runs the lookout's checks over the store
claude Oct 26, 2026
04d61cd
Sonny: lookout fixes from the first real replay
claude Oct 26, 2026
86dbbea
Sonny: lookout reads the person's words as a task's brief
claude Oct 26, 2026
cf076f5
Sonny: lookout notes reach the lead as the person's words
claude Oct 26, 2026
fb7ae37
Sonny: a lookout note names every reason's action
claude Oct 26, 2026
41da630
Sonny: the lookout is on by default
claude Oct 26, 2026
19f3dd1
Sonny: a lookout note says pass it on once
claude Oct 26, 2026
834ec92
Approver: the memory script and task worktrees are not the approval c…
claude Oct 26, 2026
3121171
Sonny: describe the lookout in the README
claude Oct 26, 2026
01a0999
Sonny: lookout fixes from the first Codex round
claude Oct 26, 2026
9e8eb06
Sonny: lookout fixes from the second Codex round
claude Oct 26, 2026
5534df2
Sonny: an approval check reads the call the approval names
claude Oct 26, 2026
49131fe
Sonny: lookout-replay samples the newest checks across leads
claude Oct 26, 2026
ed4ee30
Merge main into the lookout branch; sandfs task copies are not the ap…
claude Oct 26, 2026
08349a9
Sonny: the lookout checks a turn the approver held back
claude Oct 26, 2026
5f2e07f
Sonny: lookout replay checks approvals still waiting
claude Oct 26, 2026
488b8e4
Merge main into the lookout branch
claude Oct 26, 2026
3cf73ab
Sonny: lookout checks the rest of an any-one wait; replay pages bots …
claude Oct 26, 2026
b8a3495
Sonny: a refused lookout note waits the gap before it is sent again
claude Oct 26, 2026
98c24ff
Sonny: a failed lookout check forgets its words so the retry reads them
claude Oct 26, 2026
245344c
Sonny: two approvals in one turn are two lookout notes
claude Oct 26, 2026
c4ee5aa
Sonny: a lookout item told unchecked keeps the words read for it
claude Oct 26, 2026
fc8e2c2
Sonny: the lookout runs on any model the daemon serves; AGENT_JUDGE p…
claude Oct 26, 2026
1a591ec
Merge the merged lookout branch, keeping main's version
claude Oct 26, 2026
95e537d
Sonny: the lookout's default reads the lead's provider; judges are ne…
claude Oct 26, 2026
f68ea80
Sonny: the lookout forks only a judge, refusing a bot that took its name
claude Oct 26, 2026
c1196f0
Merge origin/main into claude/project-thread-nzse3y
claude Oct 26, 2026
aa7b525
One agent lookout watches every lead, for the app and Sonny alike
claude Oct 26, 2026
89fd31a
Merge the merged #168 branch history (no changes)
claude Oct 26, 2026
63c2a46
Make the lookout `sonny lookout`, with task updates folded in
claude Oct 27, 2026
06fad95
Quieter lookout notes, after a day of real use
claude Oct 27, 2026
2a024ad
Lookout: pick judges before the replay, key each silence, bind results
claude Oct 27, 2026
2310b87
Lookout: a polled wait reads no silence, quiet checks see the call
claude Oct 27, 2026
4a6883c
Lookout: a person's first words in a task stay theirs
claude Oct 27, 2026
c7aa0b3
Merge main into the lookout branch
claude Oct 27, 2026
6f8b112
Merge main into the lookout branch
claude Oct 27, 2026
47a5da8
Lookout subscriptions: sonny lookout ls/add/rm, look-fors, /lookout
claude Oct 27, 2026
db594c7
Join the merged #180 branch history
claude Oct 27, 2026
d382c0f
Lookout subscriptions: plain names, one edit at a time, applied on se…
claude Oct 27, 2026
223f689
Lookout subscriptions: a wake acts only on high answers; late subscri…
claude Oct 27, 2026
46917b1
Lookout subscriptions: a watched root made again keeps its tasks; rep…
claude Oct 27, 2026
336580b
Lookout subscriptions: name the look-for that matched, grade replays …
claude Oct 27, 2026
1682ed0
Lookout subscriptions: a look-for replaced while it was judged or hel…
claude Oct 27, 2026
40b3a99
Lookout subscriptions: subscribers take any bot name; bots a glob nam…
claude Oct 27, 2026
85953f1
Merge main into lookout subscriptions
claude Oct 27, 2026
216d2ee
Lookout: judge forks expire on their own
claude Oct 27, 2026
b907ba7
Keep the merged branch history; this branch restarts from main
claude Oct 27, 2026
c370c3c
Lookout: one check for a task drifting from what the person means
claude Oct 27, 2026
9212b54
Keep the merged branch history; this branch restarts from main
claude Oct 27, 2026
65d97f3
Lookout eval: a fork of the lead as a judge, and scripted cases to co…
claude Oct 27, 2026
149d310
Keep the merged branch history; this branch restarts from main
claude Oct 27, 2026
7e3cf58
Lookout eval: no fork reads the lead's later words; a relative OUT works
claude Oct 27, 2026
7d394b2
Lookout eval: a closed fork point always in reach, the lead read at e…
claude Oct 27, 2026
4babcb1
Lookout serves subscribers only; Sonny subscribes each lead to its tasks
claude Oct 27, 2026
762c225
Lookout cases: write each lead's tasks subscription
claude Oct 27, 2026
6c51a21
Lookout: follow bots' turns whether or not task updates are on
claude Oct 27, 2026
3c7a769
Lookout: a task that subscribes drops out of the news already held
claude Oct 27, 2026
b4c6a8f
Lookout: a subscriber seated mid-wait starts from what its turn waits on
claude Oct 27, 2026
7e9e263
Lookout: judge only interruptions, and send the rest with task updates
claude Oct 27, 2026
44873fb
Keep the earlier eval branch history (no content)
claude Oct 27, 2026
4de8ff0
Lookout eval: score what reaches the lead with task updates
claude Oct 27, 2026
019c224
Merge remote-tracking branch 'origin/main' into merge-main
claude Oct 27, 2026
a0cda1a
Lookout: judge only interruptions, and send the rest with task updates
claude Oct 27, 2026
78432c4
Lookout eval: score what reaches the lead with task updates
claude Oct 27, 2026
ecdd18a
Lookout: a person changing or replacing their ask in a task is drift
claude Oct 27, 2026
72bd93b
Keep the earlier re-score history (no content)
claude Oct 27, 2026
f8c2005
Lookout: task results reach the lead within a minute
claude Oct 27, 2026
5bb6210
Keep the earlier eval branch history (no content)
claude Oct 27, 2026
30e8bc7
Lookout: send task news as soon as the lead rests
claude Oct 27, 2026
3bd21c4
Keep the earlier branch history (no content)
claude Oct 27, 2026
5900abd
Lookout: every task message reaches the lead; judge only interrupts (…
claude Oct 27, 2026
8d22cc7
Keep the branch fast-forward
claude Oct 27, 2026
586897c
Lookout: tell every task bot's news; news from while it was down stil…
claude Oct 27, 2026
28ae644
sonny: a turn the lookout's note started ends quietly unless it speaks
claude Oct 27, 2026
4f6083c
lookout docs for pick C; the folded note row names messages
claude Oct 27, 2026
1e71dfa
lookout note: tell the person the outcome of work the lead started
claude Oct 27, 2026
90bca6d
Merge remote-tracking branch 'origin/main' into claude/project-thread…
claude Oct 27, 2026
ad62c37
lead text: task updates carry every message; nothing for the person, …
claude Oct 27, 2026
2afb476
lookout: one line per message or turn end; a narrower ending line
claude Oct 28, 2026
fc4c195
docs: screenshot of a quiet note turn
claude Oct 28, 2026
bed957a
lookout: bound each note's bytes; a spilled reply still speaks; eval …
claude Oct 28, 2026
64aa289
lookout note: outcome once, no blanket silence on the person's own turns
claude Oct 28, 2026
4a2bda9
lookout: a task's waiting messages are bounded; eval reads items in b…
claude Oct 28, 2026
e4e3a87
Merge origin/main into claude/project-thread-nzse3y; at most two judg…
claude Oct 28, 2026
6971350
lookout: news held for a busy lead survives a restart; bounded check …
claude Oct 28, 2026
120633a
lookout: a reconnect starts with no checks under way; eval keys event…
claude Oct 28, 2026
4e2fb7f
lookout: a message is news once read, before its judge answers; eval …
claude Oct 28, 2026
dab79e5
Merge main
claude Oct 28, 2026
d18f197
lookout: a message told while it was judged does not steer again
claude Oct 28, 2026
b71f505
lookout: a reconnect starts from the oldest news not told; eval reads…
claude Oct 28, 2026
e935048
Merge main
claude Oct 28, 2026
6fd3e5e
bench: replay scoring counts every task message as task updates
claude Oct 28, 2026
24a69fb
lookout: the note asks the lead to settle what the person means; pers…
claude Oct 28, 2026
cdffd7a
Merge the merged branch's history, keeping main's tree
claude Oct 28, 2026
a1aabec
Merge main
claude Oct 28, 2026
dba7e21
lookout: count the person's messages in a task since the lead wrote t…
claude Oct 28, 2026
0d21080
bench: a drift arm picks what the judge read as drift; a failed perso…
claude Oct 28, 2026
49acf3d
Merge main
claude Oct 28, 2026
6a94875
Back to the shipped note; bench: drift arm is that note with the judg…
claude Oct 28, 2026
9a701cf
Lookout: the judge picks a task drifting from what the person means, …
claude Oct 28, 2026
726ca04
docs: the drift picks' measured effect
claude Oct 28, 2026
25adc1a
Lookout: a reconnect rebuilds each task's exchange; a drift check key…
claude Oct 28, 2026
b846dcb
Merge main
claude Oct 28, 2026
398d0e8
Merge main
claude Oct 28, 2026
e45a9ad
Lookout: only a brief the lead wrote makes repeated words someone else's
claude Oct 28, 2026
10df501
Sonny: the lookout forwards everything and judges nothing
claude Oct 29, 2026
10df526
Merge the merged #273 branch history (no changes)
claude Oct 29, 2026
2744902
Lookout: the person's words go first and keep cutting in; a reconnect…
claude Oct 29, 2026
b2dd6f7
Lookout: each ended turn comes with what the task replied, marked as …
claude Oct 29, 2026
f7f0f80
Lookout: a note admits a task's turns only when they fit as they will…
claude Oct 29, 2026
e47d30a
Lookout: a reply waits only with a trigger's answer to the lead
claude Oct 29, 2026
fb4f126
Lookout: a reply reads within 80 KB, so a task's turns always fit a note
claude Oct 29, 2026
034c7c0
Lookout: a subscriptions call counts from the next event, not the nex…
claude Oct 29, 2026
620cc57
Merge main
claude Oct 29, 2026
6cdada8
Merge main
claude Oct 29, 2026
d76a6e4
Merge main
claude Oct 29, 2026
37620f2
Merge main
claude Oct 29, 2026
8830892
Merge main: lookout sources in the judge-free lookout
claude Oct 29, 2026
e8f4ebb
Merge main
claude Oct 29, 2026
13e3e16
Merge main
claude Oct 29, 2026
dd5ba62
Merge main into the judge-free lookout
claude Oct 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Empty file removed bench/lookout_cases/__init__.py
View file
Open in desktop
Empty file.
306 changes: 0 additions & 306 deletions bench/lookout_cases/build.py
View file
Open in desktop

This file was deleted.

20 changes: 0 additions & 20 deletions bench/lookout_cases/run.sh
View file
Open in desktop

This file was deleted.

Loading
Loading