Skip to content

sync the code in 0624 - #107

Merged
SWangHash merged 120 commits into
mainfrom
wc/delete_workspace
Jun 25, 2026
Merged

SWangHash merged 120 commits into
mainfrom
wc/delete_workspace

Conversation

@SWangHash

Copy link
Copy Markdown
Owner

Summary

Fixes #

Type and Areas

Type:

Areas:

Motivation / Impact

Verification

Reviewer Notes

Checklist

  • This PR is focused and does not include secrets, temporary prompts, generated scratch files, or unrelated artifacts.
  • Relevant verification is recorded above, or skipped checks are explained.
  • User-facing strings, docs, and locales are updated where applicable.

wgqqqqq and others added 30 commits June 15, 2026 15:04
…y-compat

fix: keep mcp elicitation capability legacy-compatible
Replace staged multi-turn planning/render/audit flow with one minimal agent
prompt that delegates design rules to the pinned ppt-design skill, and trim
UI/backend code for faster end-to-end deck generation.
…-generation

feat(ppt-live): simplify deck generation to single cowork agent turn
Preserve internal thinking-panel tail follow across streaming Markdown/code-block
layout changes, while respecting explicit user scroll intent.

Add a guarded follow state for ModelThinkingDisplay so code block reflow no
longer strands the internal scroller above the latest thinking output. Pause
follow mode on upward wheel/touch/keyboard navigation and prevent queued
auto-scroll frames or near-bottom checks from immediately re-enabling follow.
Add a configurable subagent batch scheduling policy for tool execution. The new
ai.subagent_batch_execution_policy setting supports safe-only, force-parallel,
and serial behavior, with the UI exposing safe-only and force-parallel modes.

Wire the policy through round execution into the tool pipeline, preserve existing
safe-only behavior by default, and keep subagent capacity and DeepReview limits
intact. Also update Task tool guidance for parallel non-read-only subagents.
Route FlowChat session hot paths through exact service API modules instead of the broad API barrel, add a startup contract guard, and extend rapid-switch reports with post-click response metrics.
- Keep terminal tabs and inactive terminal scenes mounted so xterm buffers are hidden instead of destroyed and replayed.
- Add structured terminal replay events that preserve resize/write ordering across backend history recovery.
- Suspend terminal fit and PTY resize while right/bottom panels animate or drag-resize, then apply one final fit after settling.
- Gate PTY resize on successful local xterm resize to prevent frontend/backend geometry drift.
- Suppress incomplete Windows PTY resize repaint output that would overwrite preserved xterm scrollback.
- Add replay and resize repaint guard tests covering PowerShell, Cmd, Git Bash, and metadata-only replay.
- suppress Git Bash partial prompt redraw sequences that can move the cursor into old viewport rows after bottom terminal resize
- fix resize repaint guard regex state leakage that caused intermittent misses across repeated resize cycles
- add regression coverage for prompt-only redraw suppression across repeated resize attempts
Move prompt-cache persistence decisions and DeepReview queue event payload shaping into agent-runtime while keeping core as compatibility adapter and IO/event delivery layer.
…nd incremental flows

Restructure the PPT Live placeholder copy into three explicit usage patterns and refresh bundled assets and example dates.
…eholder-guidance

feat(ppt-live): clarify one-box prompt guidance for generate, edit, and incremental flows
…ght-report guidance

Add mandatory architecture/flow diagram triggers for engineering decks, editable CSS snippets for system diagrams, a dedicated scene template, and expanded insight-report rules for chart diversity, layout discipline, and readable density.
…itecture-visualization

feat(ppt-design): strengthen tech architecture visualization and insight-report guidance
- Add markdown-backed custom agent schema for modes and subagents
- Add desktop APIs for creating, editing, deleting, reloading, and detailing custom agents
- Integrate custom modes into agent registry, mode discovery, prompt/cache identity, and runtime policy
- Add custom agent management UI with tools, prompt, readonly/review, and runtime context controls
- Add ChatInput default mode override stored in app.flow_chat.default_mode_id
- Refresh mode/agent selectors when custom agents change
- Clean up custom agent model/profile/default-mode config on deletion
- Refine Agent scene cards, capability chips, and custom mode model display
- Add registry, serialization, custom mode, default mode, and ChatInput mode tests
…text

- Move non-visual model ComputerUse guidance out of system prompt templates
- Inject the guidance through runtime context only when ComputerUse is available
- Keep system prompt cache identity stable across primary model capability changes
- Add tests for text-only guidance inclusion, omission, and prompt-template isolation
bobleer and others added 23 commits June 23, 2026 17:04
Keep the resume card visible during background refresh and skip the
redundant post-init loadFirstPage call that caused an extra round-trip
and UI flash. Add session item enter animation for smoother list paint.
…oading-flicker

fix(mobile-web): eliminate session list loading flicker on init
…dupe invalid arg previews

- remove the OpenAI-compatible stop_reason-based arguments_is_snapshot special case
- keep stop_reason tool chunks on the normal incremental arguments path
- update OpenAI stream fixtures and tests to cover append behavior
- stop echoing raw arguments twice in invalid tool-call assistant errors
- keep a single Provided arguments preview in model-facing error text
Add stable data-testid coverage for app shell, welcome, navigation, session, agents, and skills surfaces, and document the locator planning and naming rules.
…ite-once

Replace post-generation page-by-page self-check and rework loops with upfront
budgeting and in-flow quality rules so deck generation completes in a single pass.
…-once-guidance

refactor(ppt-design): shift skill guidance from post-gen review to write-once
Add stable data-testid and data-* attributes for agent, skill, shell, browser, and terminal UI automation. Update UI test id documentation in English and Chinese.
- extract main-session activation from BTW pane helpers
- avoid auto-selecting child subagent sessions during workspace restore
- return to the welcome panel when deleting or archiving the current session
- align batch archive and delete flows with shared session manager behavior
…ng-main

test(web-ui): expose stable DOM locators for OH UI automation
- render runtime skill snapshots and listings with inline skill tags
- switch chat input skill insertion to token pills across slash and boost flows
- support inline / and $ skill triggers and wrap command picker navigation
@SWangHash
SWangHash force-pushed the wc/delete_workspace branch 3 times, most recently from bc8de42 to 8f0d274 Compare June 25, 2026 02:27
- BrowserPanel: remove stray ? in ternary condition (!isTauri || webviewUnavailable)
- ChatInput: add missing brace to close outer if block in useEffect
- AppLayout: drop duplicate TransitionDirection type and conflicting WorkspaceManager import
- SessionConfig: remove companion pet code brought back by merge (handlers, options, selected state)
- SkillsScene: drop unused systemAPI import
- SessionConfig: drop unused Plus/Trash2/Check icon imports
@SWangHash
SWangHash force-pushed the wc/delete_workspace branch from 8f0d274 to bf2f57e Compare June 25, 2026 02:40
@SWangHash
SWangHash merged commit 6fbcbd0 into main Jun 25, 2026
1 check failed
@hanlinyu1030
hanlinyu1030 deleted the wc/delete_workspace branch July 7, 2026 03:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

8 participants