Skip to content

Releases

v0.1.485

Choose a tag to compare

@github-actions github-actions released this 14 Nov 12:04
740b4e3
Fix the failed workflow "Release" on main (#1920)

* test: scroll the chat list to its end in each screenshot step

The step chatEnd only read the gap to the end of the chat list. The list
can get shorter after its first scroll, for example when the tray shows,
and the gap then stayed for the full time of the wait. Scroll to the end
after the list is ready, and read the gap after two frames. In the step
chat-agents, wait for the end of the chat after the other elements.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* test: wait for the Inbox tray before the end of the chat list in each screenshot step

The tray of the real Inbox item shows after the first scroll of the list and makes the list shorter.
Each step of owner/shop/12 that uses chatEnd now waits for the tray first.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* docs: keep only the hazard in the comment of chatTray

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

---------

Co-authored-by: mobius-toolkit[bot] <335179271+mobius-toolkit[bot]@users.noreply.github.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

v0.1.484

Choose a tag to compare

@github-actions github-actions released this 14 Nov 09:10
0972b58
Fix the failed workflow "Screenshots" on main (#1915)

* fix: publish the screenshots of a large change

The publish job gave the section to python3 in an environment
variable. Linux refuses one value of more than 128 KB, so python3
failed with "Argument list too long" when many screens changed.
The job now writes the section to a file and reads it from there.

GitHub also refuses a pull request body of more than 65,535
characters. The job now lists at most 60 screens and gives the
number of the other screens with a link to the run.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* fix: publish the screenshots of a large change

The publish job gave the section to python3 in an environment
variable. Linux refuses one value of more than 128 KiB, so python3
failed with "Argument list too long" when many screens changed.
Both copies of set_section now write the section to a file, and
python3 reads it from there.

GitHub also refuses a pull request body of more than 65,535
characters. Python now removes table rows from the end until the
body fits. It then adds one line with the number of the hidden
screens and a link to the folder of the pull request on the branch
screenshots.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

---------

Co-authored-by: mobius-toolkit[bot] <335179271+mobius-toolkit[bot]@users.noreply.github.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

v0.1.483

Choose a tag to compare

@github-actions github-actions released this 14 Nov 08:46
a1b1614
Name the users of a Harness pause on its banner, and show a Workstrea…

v0.1.482

Choose a tag to compare

@github-actions github-actions released this 14 Nov 08:23
6827c3a
Add the shared composer to the task detail view (#1910)

* feat(web): add the shared composer to the task detail view

The detail view of a task in the Tasks tab shows the composer of
web/src/composer at its bottom. It uses the draft of the Lead chat, so
both composers show the same draft. The first change of the draft from
the view attaches the task once for each opening of the view, and it does
not close the phone Tasks sheet. A send queues the message in the outbox,
and the Lead chat sends it.

The lifted Subscriptions of the composer have their own keys, because
the Lead chat has the same Subscriptions in the same application.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* test(web): wait for the send request in the phone test of the task view

The test now waits for the request to /api/chat/messages before it ends, so the end of the test does not stop the request.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* test(web): let the send of the task view reach the server

The two send tests of the composer of the task view no longer fake the answer of the chat. They wait for the real request, and then read the stored message through GET /api/chat and check its Inbox item. The text of each test has a time stamp, so no other test reads it.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* fix(web): stop the composer of the Tasks tab while its field is not in the page

On a phone, the panel keeps its detail view when the Tasks sheet closes. The voice input of the view stayed registered without its field, so the text of a conversion went into no field. The composer Subscriptions of the Tasks tab now run only while the panel is in the page.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* test(web): cover a tab change during a conversion in the task view

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

---------

Co-authored-by: mobius-toolkit[bot] <335179271+mobius-toolkit[bot]@users.noreply.github.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

v0.1.481

Choose a tag to compare

@github-actions github-actions released this 14 Nov 08:06
c55e059
feat: refuse a save with an effort that the Harness does not give (#1…

v0.1.480

Choose a tag to compare

@github-actions github-actions released this 14 Nov 07:44
2cac7e2
Read the effort levels of each Harness from the Harness (#1907)

* feat: read the effort levels of each Harness from the Harness

Remove HasEffortLevels. ReadChoices reads the efforts of each model of each
Harness. CheckRoles allows an effort and no effort for each Harness. mobius
init asks for the effort when the Harness gives effort levels for the model.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* test: rename the init effort test

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* test: drop the two effort refusals from the task Implementer endpoint test

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

---------

Co-authored-by: mobius-toolkit[bot] <335179271+mobius-toolkit[bot]@users.noreply.github.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

v0.1.479

Choose a tag to compare

@github-actions github-actions released this 14 Nov 07:24
5e7e98c
test: fix the scroll of the chat screenshots (#1906)

The chat list makes its first scroll when it is ready, and that scroll
can go to the first unread message. Wait for the ready list before the
scroll of the test. Scroll the usage limit chat after the pause banner
and the tray show. Wait for the end of the cancelled chat.

Co-authored-by: mobius-toolkit[bot] <335179271+mobius-toolkit[bot]@users.noreply.github.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

v0.1.478

Choose a tag to compare

@github-actions github-actions released this 14 Nov 07:07
831bded
Open a detail view of a task from its row in the Tasks tab (#1770)

* feat(panel): open a detail view of a task from its row in the Tasks tab

Add the GetTask endpoint to get the issue body and pull request of a
task. In the Tasks tab, open a detail view when the Owner selects the row
text of a task. Render the back button, the reference and title, the
state badge and blockers, the pull request link, the Open on GitHub link,
the Chat button, and the issue body. Restore the previous scroll position
when the Owner returns to the task list.

* fix(panel): ignore task detail responses for other tasks and match Send button exactly

Include the repository and number of the task in FetchTaskDetail,
TaskDetailLoaded, and TaskDetailFailed so update ignores responses
that do not match the opened task. Use exact match for the composer
Send button in e2e tests so the locator does not match task row buttons
whose title contains Send.

* fix(panel): show attach task error in detail view

Render the row error with a destructive badge in the detail view below
the action buttons when Chat fails to attach the task. Add a Story test
verifying that an attach error appears in the rendered detail view.

* test(panel): check the attach error in the detail view with a Scene test

The Story test passed the inert HTML builder to the view, which tsc
refuses. Use the Scene test of Foldkit, which gives the view its own
builder.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

---------

Co-authored-by: mobius-toolkit[bot] <335179271+mobius-toolkit[bot]@users.noreply.github.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

v0.1.477

Choose a tag to compare

@github-actions github-actions released this 13 Nov 20:13
c759a42
Send an approved task outside the merge queue back for approval when …

v0.1.476

Choose a tag to compare

@github-actions github-actions released this 13 Nov 15:01
8da9ae2
Put the Lead events of the CI fix tasks first (#1895)

* feat: put the Lead events of the CI fix tasks first

A CI fix event gets the next turn of its Lead after the messages of the Owner. The Lead session of a new chat for such an event waits for its Lead slot at the CI rank.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* fix: order the CI fix events in Go

The hot query test refuses a scan, and the view ci_fix_issues needs one in the order of ListReadyLeadEvents.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* fix: write the queue insert of give as a switch

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* fix: queue the freed held events behind the CI fix events

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* fix: follow the chat for the CI flag of a Lead session

A waiting Lead session takes its CI flag from the next turn of its chat, and each new session of a chat reads the flag again from the first item.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* test: queue the new Lead session while the slot is held

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

---------

Co-authored-by: mobius-toolkit[bot] <335179271+mobius-toolkit[bot]@users.noreply.github.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>