5 Commits
Author SHA1 Message Date
Aleksander DygońandClaude Opus 5.5 03c2ff34c2 Remove code comments and add the no-comments rule (#12604)
* Remove code comments and add the no-comments rule

Remove the comments from first-party code and configuration files:
TS/JS, JSX, CSS, SQL migrations, Swift, Kotlin, Gradle, shell,
Docker, systemd, Nix, YAML, TOML, JSONC and GritQL.

Keep only what a tool reads: license headers, shebangs, compiler
and lint directives, JSDoc types in plain .js files, generated file
headers, GritQL scope lines, the install-server.sh help text and the
openbot-boat-pre-stop marker. Vendored code, fixtures and released
schemas keep their comments.

Add the rule to AGENTS.md, and change the docs that asked for
comments. Add two i18n keys that only comments named to
RELEASED_SOURCE_KEYS.

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

* Keep comments that are data in strings

The benchmark task files in opencode-free-models-bench.ts tell the
model what each function must do. The sample file and HTML page in
the two stories are display content. Put their comments back.

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

* Keep released migrations and mobile native inputs unchanged

check:migrations compares each released migration with v0.73.0 byte
for byte, so restore src/backend/migrations. A change to a native
build input of apps/mobile makes a new runtime version, so restore
those files and the extra fingerprint sources.

List these files and the comment markers that a tool reads in
AGENTS.md. Remove the last requests for comments in
src/renderer/AGENTS.md.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-19 12:19:28 +01:00
NBandClaude Opus 5.5 7ff90ccd1c Show the cards above the message box as one card (#3045)
* Show the cards above the message box as one card

The waiting block, open waits, queue, reply preview and notices are now in one
stack. The stack draws one border and one radius and goes under the input. The
parts in it have no gaps and no corners of their own, and a 1px line divides
them. Before, the queue slot clip cut the lower edge of each card flat, and a
card edge showed through the notice below it.

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

* Check the one composer stack and keep its narrow and exit behavior

The task plan e2e check now finds the cards in the stack. It checks that
each card sits flush on the card above with one shared line, and that the
cards fill the stack's width. It also checks the story with notices.

On a narrow pane the stack keeps the 2px side inset that the queue had.
When the last card is a compaction notice that leaves, the frame drops
behind the input with it, so no empty frame stays.

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

* Remove the attachment exceptions that the stack made dead

The queue and the last waiting block did not go under the message box when it
held an attachment. In the stack these cards have no bottom margin, so the two
rules did nothing. The stack always goes under the message box, as the notices
already did. The attachment row is in the opaque message box, which paints
above the stack, so the stack cannot hide it.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-24 19:02:33 +02:00
NBandClaude Opus 5.5 b7ab4b96a9 Remove the old channel view (#2738)
* Bring channels back on the host

The host no longer archives channels at start or refuses channel work. It
advertises channel-chats-v1 and channel-delete-v1 again. Channels that the
earlier release archived stay archived; the user restores each one by hand.

- Agents get the channel tools again, and the instructions, tool guidance
  and history text name them again. The section_teammates block stays.
- Agent import offers group chats as channels again.
- Remove error.backend.channelsRetired and the deprecation notes.
- The desktop renderer still hides channels; a later change shows them.

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

* Show desktop channels in the conversation view

The desktop opens a channel in the same view as an agent chat. Plan
cards, visual replies, ask_ui blocks, approvals and browser takeovers
of channel work show as they do in an agent chat. The old channel view
stays for the web client until it moves too.

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

* Show web client channels in the conversation view

The web client opens a channel in the same view as an agent chat, with
the desktop channel source. Files upload through the agent runtime of
the host, as in the old channel view.

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

* Make channel work robust with the newer agent features

- Channel work holds only the member agent that does it. Its tasks run
  beside the channel turn. Only declared browser and workspace resources
  are serialized.
- A hidden lead does not route, and hiding an agent gives its queued
  channel work back to the channel.
- A provider change hands over the channel context threads too.
- A channel task moves to the backup model at a usage limit, and the
  channel shows the model switch.
- Channel turns can use ask_ui, the plan card, html_render and screen
  capture. Each capture in a channel asks and names the channel.

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

* Add release notes for the desktop channel view

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

* Add release notes for the web client channel view

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

* Show channels again on servers that offer them

The desktop channels domain is supported when the active server advertises channel-chats-v1.
Open, edit and create do nothing on a server without channels, so a saved selection or a
notification cannot open one there. The "+" menu label names channels when it offers them.

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

* Notify channel replies and name the channel in channel prompts

Add the optional channel-attention-v1 capability. For a peer that sends it, the channel list
gives the threads of each member and the author and status of the newest message, and the
channel page gives the ui-blocks-v1 block of each message. Released peers get the same wire.

The desktop app and the web client notify an agent reply in a channel from the channel list.
A prompt, an approval or a takeover on a channel thread names the channel, and a click opens
the channel.

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

* Describe the per-agent channel rules in the docs

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

* Let the conversation view show a channel

The view takes an optional channel source. Each place that assumes one
agent takes a branch that is fixed for the life of the view. No caller
passes a channel yet.

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

* Remove the old channel view

Both surfaces show a channel in the conversation view now, so the old
view, its timeline helper and the interface text that only it used go.
The channel stories show the real view.

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

* Keep a sent channel message when the refresh fails

When the refresh after a send failed, `send` gave the earlier page, which did not hold the
message. The conversation then showed the stored message as not sent. Now `send` reads the page
from the host that stored the message when the refresh did not bring a newer page, or when the
reader changed scope during the send.

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

* Name the current channel view in the web client docs

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

* Restore a deleted channel from the sidebar

No client sent the `restore` command, so a deleted channel stayed preview-only. The channels
that the earlier release archived stayed locked. Each row in Deleted channels now has a menu
with "Restore channel", on desktop and web. It shows to the people who can delete a channel. The
delete dialog now says that the channel can be restored.

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

* Describe only the channel changes that this part ships

The restore action ships with the desktop channel gate.

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

* Show a toast when a channel restore fails

The sidebar menu shows no channel error. A failed restore now shows a
toast and clears the failed command, so the error and its Retry do not
show in the next channel that the user opens.

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

* Read channel lists only for servers whose replies can notify

A streaming reply changes the channels on each frame. The main process
read the channel list for each change, also when notifications were off,
muted, or not at the "all" level. It now drops the baseline of such a
server and does not read. A prompt still reads its channel when the
list has none.

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

* Check the channel thread of a prompt only where channels list threads

The threads of a channel summary come with channel-attention-v1, in the
next part of the stack. The channel tools E2E read them here, and the
node typecheck failed. The next part adds the check again.

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

* Check that a channel prompt is on the agent's channel thread

This part adds the threads of a channel summary, so the channel tools
E2E checks the thread of the ask_ui prompt and of the plan card again.

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

* Notify a channel reply once, when the channel work ends

A turn that writes many messages notified once for each message. Now a
reply notifies only when its newest message is an agent's, it stopped
streaming, and no task runs in the channel.

The desktop read the agent list of a remote server on each channel change.
Now it reads the list only when a reply is new.

A known local channel thread no longer waits on a promise, so the
desktop shows its notification before forwardAgentEvent returns.

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

* Count waiting and queued channel tasks before a reply notifies

When an agent passed work to other agents, its task waited and the new
tasks were queued. The released activeTasks counts only running tasks,
so it was 0, and each step of the work notified.

channel-attention-v1 now adds openTasks to each channel summary: the
count of queued, waiting and running tasks. A reply notifies only when
openTasks is 0. A host without the field falls back to activeTasks. The
meaning of activeTasks does not change.

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

* Do not export the channel source types before a caller uses them

Knip rejects an exported type that no file imports. The desktop channel
view exports them again when it imports them.

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

* Export the channel source types for the desktop channel view

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

* Let selectAgent close the channel when an agent opens from a channel

The wrapper closed the channel before the guard of selectAgent. When the
guard returned, the channel closed and no agent chat opened. selectAgent
already closes the channel after its guard.

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

* Delete the old channel view when the web client stops using it

The web client was the last user of ChannelConversation. Knip rejects
the unused files and the two exports that only they used. This moves
these deletions, and the text keys that only the old view read, from
the next change in the stack.

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

* Name the shared channel view in the web client document

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

* Open the renamed channel story and share the story mount

The task-plan E2E opened the old id of the channel story, so its stack
check could not find the story. The channel story copied the mock bridge,
the controller and the providers of the agent story. Both now use
StoryConversation, which takes any ConversationView props.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-21 23:38:14 +02:00
NB 62d13d3067 Hold teammate answers until every teammate of the request is done (#961)
* Hold teammate answers until every teammate of the request is done

When an agent asked several teammates, it ran one turn for each answer. It answered the person
again after each teammate.

- An answer to a request that went to two or more teammates waits while another teammate of that
  request still has its copy queued, starting or running.
- When the last copy ends, all waiting answers start in one turn. The prompt says how many
  messages the turn has.
- If a copy ended with no answer (failed, interrupted or cancelled), the prompt names that teammate
  and tells the agent not to wait for it.
- A message from the person does not wait. It starts at once and takes the answers that are
  already in, so the agent does not answer twice.
- Each end of a copy, a failed start and a cancel from the queue schedules a drain for the agent
  that asked, so the held answers start.
- The waiting block shows "reads it next" only when the answers are no longer held.

* Release held answers after a process loss, and keep a batch start consistent

- When the own provider process of a teammate exits, its requesters get a drain, so the answers
  that they hold start.
- An answer whose attachment changed fails alone. The person's message and the other answers of
  the batch still start.
- A pending runtime refresh applies to a batch start. Before, the first held answer counted as a
  start that was not confirmed, and the refresh waited for the next drain.
- After a restart, each delivery of a batch that was not confirmed gets the state of the turn that
  its first message started. Before, the other answers showed as interrupted.
- interrupt_agent cancels the caller's requests only, not its answers, because the target can hold
  an answer until its other teammates answer.
- docs/ARCHITECTURE.md describes the hold rule.

* Return the answers of a batch to the queue when its start fails

Only the message that starts the turn fails. The provider did not read the other answers, so they wait for the next turn.

* Let interrupt_agent cancel a linked follow-up request

Only a linked answer that expects no reply stays in the queue. A linked message that asks for a reply is work that the caller gave.

* Hold and group only teammate answers, not linked requests

An answer is a linked agent message that expects no reply. A linked question asks for a reply, so it runs at once, and the queue shows it, not the waiting block.
2026-09-27 15:22:57 +02:00
NB af99355ec9 Show agent plans as task lists, and show teammate replies in a waiting block (#960)
* Show agent plans as task lists, and show teammate replies in a waiting block

An agent's plan was lost, and an agent that asked teammates showed nothing until the answers came.
The answers then showed in the queue like the person's own messages, with Steer and Edit.

- Each turn with a plan shows one task list in the chat. It updates in place and stays in history.
  Codex update_plan, Claude TodoWrite and TaskCreate/TaskUpdate, and ACP plans all feed it. The
  plan is saved in the message JSON, so there is no migration.
- Codex gets tools.update_plan. The Claude CLI gets CLAUDE_CODE_ENABLE_TODO_TOOLS, because without
  a terminal it gives the newer models no plan tool.
- A "Waiting for replies" block above the composer lists each teammate that the agent asked, and
  the child tasks of a waiting channel task. Each row shows asked, working, replied or failed. The
  answers leave the queue panel.
- A plan is never the preview, the unread count or the result that goes back to a teammate.
- Peers and mobile get the plan as checklist text. The Team API wire does not change.

* Leave plans out of the unread counts, and fix four review findings

- The agent and channel unread counts in the backend leave out plan rows, as the renderer does.
  The channel divider walk skips plans too.
- The activity line skips the plan row, which streams for the whole turn, so it still shows the
  latest commentary.
- A paused channel sub-task does not show as failed in the waiting block. It shows in the stopped
  tasks with Resume.
- A new Claude turn drops the finished tasks of earlier turns. A TaskUpdate with a subject for an
  unknown task, after a runtime release, adds it again.
- Remove two unused exports (knip) and the unused chat.taskList.state.failed key.

* Remove a turn's plan when the agent deletes its last task

An empty plan update was dropped, so the old list stayed in the transcript. Now it removes the plan message of the running turn.

* Accept a missing value in the optional task list and avatar props

The type ratchet found new exactOptionalPropertyTypes errors in the task list, the waiting block, and their callers.

* Lower the type baseline for the avatar callers that no longer fail
2026-09-27 15:02:58 +02:00