feat(web): subagents in the header with a child view, model-written titles, and a quieter sidebar - #922
Merged
Conversation
…itles, and a quieter sidebar The browser client showed nothing of a session's delegations: the gateway dropped every agent_progress frame, and the Agent tool's result envelope reached neither the wire nor the saved session. Titles were the first prompt line, and every abandoned "New session" press left an untitled row behind. Web - "N subagents ▾" beside the title, with a live dot while any run; the list names each delegation with type, model, state, tokens and duration, and opens it in the conversation column: prompt, its own tool rows, a read-only seat, the parent title as the way back and a sibling switcher. - Agent rows read `Agent · description` with live activity or `N tools · duration`, and a body of prompt, report and a button into the run. - Bash rows summarise with the model's description of the command; the command itself stays in the terminal card. - Blank sessions are listed only while on screen, as "New session". Backend - agent_progress → subagent.progress events, carrying the tool_use_id of the Agent call so a running row and its progress find each other. - The Agent result envelope (agent_id, status, model, duration, tokens, tool count) rides tool.complete as result.agent and is persisted beside the stored result, so a resumed session lists its subagents like the live one. - Foreground subagents keep a sidechain transcript like background ones; subagent.transcript reads it in the session.resume message shape. - A generate_title control asks the session's own provider for a short title after the heuristic name lands; an explicit rename wins. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Brings the browser client up to the DeepSeek Harness reference on the three gaps that showed in a side-by-side: subagents were invisible, session titles were the raw first prompt, and the sidebar filled with untitled rows.
N subagents ▾beside the session title, a live dot while any run, and a dropdown naming each delegation with its type, model, state, tokens and duration. A row opens the run in the conversation column: its prompt, everything it did as tool rows, and a read-only seat where the composer would be, with the parent title as the way back and a sibling switcher on the child's name.agent_progressframes intosubagent.progressevents (they were dropped before) carrying thetool_use_idof the Agent call; the Agent tool's result envelope ridestool.completeasresult.agentand is persisted beside the stored result, so a resumed session lists its subagents exactly as the live one did. Foreground subagents keep the same sidechain transcript background ones do, and a newsubagent.transcriptmethod serves it.generate_titlecontrol asks the session's own provider for a short title and swaps it in; a rename typed in between wins. Not the Anthropic-onlygenerate_llm_titlepath.Agent · descriptionwith live activity orN tools · duration; Bash rows summarise with the model's description of the command; blank sessions are listed only while on screen, as "New session".Test plan
ui-web:npm run typecheckclean,vitest588 passing, including new suites for the subagent catalog fold, progress folding, envelope hydration, Agent/Bash row descriptions and blank-session filtering.tests/server/test_desktop_translate.py,test_desktop_sessions.py,test_desktop_gateway.py,tests/test_session_title.py, the Agent tool admission and terminal-progress suites, and the message-construction suites all pass, with new tests forsubagent.progress,result.agent, transcript loading, the RPC and title generation.2 subagents, 1 running, the to-dos panel moving to3 completed, both child views with their transcripts, and a model-written title replacing the heuristic within thirty seconds.🤖 Generated with Claude Code