Commit Graph
16 Commits
Author SHA1 Message Date
weishu d29713aeda fix(native): render plan proposals inline
Render ExitPlanMode and exit_plan_mode Markdown from input.plan in iOS and Android conversations. Preserve approvals, raw source and diagnostics while hiding empty output placeholders and prewarming plan documents.

Add generated protocol fixtures and native regression coverage for long plans, live updates, recycling, themes and typography.
2026-09-12 16:06:16 +08:00
weishu 418f11f80e feat(ios): improve inline question answering 2026-09-12 15:20:01 +08:00
weishu 0c7f557ba3 feat(ios): browse tool groups in a native inspector 2026-09-12 13:03:27 +08:00
weishu f0fe2f2775 fix(ios): clean up localization catalog extraction
Translate the unknown delivery state, keep decorative content verbatim, and pin integer interpolation formats to existing catalog keys.
2026-09-12 12:07:54 +08:00
weishu 0c4abcb3d1 feat(codex): share sessions across terminal and web
Use one native app-server for terminal, Web and phone clients while retaining the existing CLI and Runner lifecycle.

Synchronize native queues, permissions, question history and steering state; preserve explicit permission precedence and per-turn usage models. Resume inactive clear commands through Runner and reject independent child cold resumes.

Add shared-runtime regression tests, generated protocol fixtures and lifecycle documentation.
2026-09-12 10:59:17 +08:00
weishu f5e5bcfa10 fix(native): render question tool answers in details 2026-09-11 21:04:48 +08:00
weishu 7995334399 fix(ios): prevent long user messages from blocking chat
Keep normal multi-screen prompts inline and bound oversized messages to a preview. Add a screen-owned paged reader with exact full-text copy and preserved reading state, plus threshold, Unicode, layout, and presentation tests.
2026-09-11 17:14:21 +08:00
weishu 0e98c97a34 fix(native): improve tool input and output previews 2026-09-11 13:08:04 +08:00
weishu 6c0ef32350 feat(ios): add native tool inspector and compact tool groups
Move tool output into a live sheet with group navigation and full-content copying. Keep lightweight grouped summaries inline, add agent process pages, and preserve transcript reading position.\n\nAdd real transcript and sheet regression coverage with localized summaries.
2026-09-11 11:30:13 +08:00
weishu 10042f79c6 feat(ios): refine chat typography and reading layout
Use unified 16pt Dynamic Type body text, clearer Markdown spacing, and a shared reading column. Preserve transcript anchors across typography changes and bound tool command previews to two lines. Add layout and typography regression tests.
2026-09-10 16:22:36 +08:00
weishu c2e3d16b7e feat(native): improve anchored chat scrolling and history loading
Add viewport-driven paging with layout acknowledgements, bounded retries, cancellation gates, and epoch-safe history retention.

Preserve transcript anchors and expansion state, fix tool-group identity collisions, and serialize Android history coordination on Main.

Reduce per-scroll composition and layout work; add native regression tests, CI coverage, and profiling guidance.
2026-09-10 16:22:36 +08:00
weishu 41c0db1aad fix(ios): missing argument label in terminalTitle call 2026-08-18 16:35:38 +08:00
weishu 6066f20c20 feat(ios): zh-CN localization + catalog dedupe (A-M5)
Part 1 — Simplified Chinese localization of the iOS app layer:
- ios/Hapi/Resources/Localizable.xcstrings: hand-authored String Catalog
  (395 keys, zh-Hans; en implicit as source). Terminology mined from
  web/src/lib/locales/zh-CN.ts (会话/新建会话/权限模式/允许/拒绝/工作树/
  机器/语音输入/用量/草稿夹 …).
- Mechanical edits only: merged multi-part string concatenations into
  single LocalizedStringKey literals, converted ternary/plain-String user
  copy to String(localized:), switched helper params (detailRow,
  DashboardCard, optionPicker, notice) to LocalizedStringKey.
- LocalizedNoticeMapper (app layer): display-point translation of the 27
  known HapiKit-emitted strings (ChatInteractor notices, dictation errors,
  window-sync warnings, files fallbacks, worktree-name validation) with
  verbatim passthrough for server-originated text; package stays
  language-free.
- Deliberately untranslated (web parity / non-copy): event rows
  (EventPresentation — web renders presentation.ts verbatim), catalog
  option labels (Default/Auto/Sonnet/permission modes — web shows them
  verbatim), tool names, code-like titles (grep(pattern:), MCP:, Skill,
  Task), unit suffixes (B/KB/MB, m/h/d/w), decorative separators.
- Language row wired: AppLanguage gains .system (follow system, new
  default); explicit picks write the AppleLanguages override, Follow
  system removes it; footer notes a relaunch applies it (no supported
  in-place SwiftUI locale swap).
- project.pbxproj: zh-Hans added to knownRegions.

Part 2 — #39 catalog dedupe (HapiProtocol):
- NewSessionCatalogs.claudeModels/claudeEfforts now derive from
  ClaudeModels/ClaudeEfforts (single source shared with ModelCatalog);
  codexReasoningEfforts stays own data (web CODEX_REASONING_EFFORT_OPTIONS
  minus max); effortLabel delegates to ModelCatalog.capitalizedFirst.
- New CatalogTests lock test for the derived option lists.

Gate: bash ios/scripts/linux-test.sh — 459 tests green.
2026-08-18 16:04:03 +08:00
weishu 05d063cc33 feat(ios): composer attachments + dictation (A-M3f)
Attachments (Android B-M3f semantics ported verbatim):
- HapiClient/Attachments/AttachmentPolicy — pure plan matrix (>4 MB
  recompressible image -> 2048 px JPEG q85 with .jpg rename, 50 MB hard
  reject, 192 MB image read cap, 512 px q80 previewUrl data-URL thumbs,
  data-URL parse/round-trip).
- HapiClient/Attachments/ComposerAttachments — upload-on-pick tray over an
  AttachmentUploading seam (APIClient conforms): uploading/ready/failed
  chips, retained payload for retry, remove -> best-effort delete,
  mid-upload removal deletes the orphan on completion, consume() ->
  AttachmentMetadata with JPEG data-URL previewUrl, discardAllDetached +
  deinit orphan cleanup (Android onCleared analogue).
- ChatInteractor: tray ownership, unsettled chips refuse the send with a
  notice, attachments-only sends post empty text, optimistic rows carry
  the metadata, appendDictatedText/postNotice/discardAttachments.
- App: AttachmentPreparer (capped security-scoped reads, ImageIO
  downscale/encode with EXIF transform, HEIC-undecodable fallback),
  PhotosPicker multi (videos via FileRepresentation temp files),
  UIImagePickerController camera capture, fileImporter; composer chip row
  (thumb/spinner/tap-to-retry/remove) with attachment-aware send gating;
  user bubbles upgrade chips to off-main-decoded previewUrl thumbnails
  (web-sent attachments included).

Dictation (Android B-M3ce port):
- HapiProtocol/Models/VoiceApi — TranscriptionResponse,
  TranscriptionProvidersResponse, TranscriptionProviderInfo.
- HapiClient/Endpoints/VoiceEndpoints — GET
  /api/voice/transcription/providers + multipart POST
  /api/voice/transcription (file/provider/mode/language, Android part
  order) over MultipartFormData; DictationTranscribing conformance.
- HapiClient/Voice/DictationController — idle/starting/recording/
  transcribing, transcribed/noProvider/error events, provider memoized
  (first standard-capable entry), appendTranscript port.
- App: AVAudioRecorderDictation (m4a/AAC mono 44.1 kHz 96 kbps, session
  activate/deactivate), mic button + recording chip (elapsed + cancel),
  record-permission request via AVAudioApplication.

Info.plist gains NSMicrophoneUsageDescription; the camera string now
covers attachment capture (modern PhotosPicker needs no photo-library
permission). Tests: policy matrix, tray over the real client with exact
base64 upload bodies + gated in-flight scenarios, dictation controller
suite with fake recorder/transport, voice endpoint request shapes, and
the interactor attachment-send matrix transcribed from the Android VM
tests (wire bodies byte-for-byte).
2026-08-18 11:27:28 +08:00
weishu eed3dddb94 feat(ios): composer, permission actions, session config (A-M3ab)
Interaction layer turning the read-only chat into a working remote control,
mirroring the merged Android B-M3ab feature-for-feature (web authority where
the ports disagreed):

- ChatInteractor (HapiClient, fully swift-test covered): optimistic composer
  sends (appendOptimistic -> POST -> status settle), queue-by-default with a
  long-press Send&Steer intent while a turn is active, tap-to-retry on failed
  rows (steer retries degrade to queue), per-session drafts
  (UserDefaultsChatDrafts, hub-scoped keys, debounced + flushed on close).
- session_inactive (409) recovery: one POST /resume (current permissionMode)
  then retry; a superseding session id seeds the new window
  (MessageWindowControllers.seed), migrates the draft, retargets the
  optimistic row and emits sessionSuperseded -- ChatView/HomeView replace the
  navigation entry in place.
- Queued bar: uninvoked sends in web sort order with Cancel (optimistic
  DELETE; invoked-race ingests the authoritative row as sent), Edit
  (cancel + composer prefill, newer-draft guard) and Steer (invoked answers
  reconcile a missed consume); single-flight per-row op guard.
  reconcileQueuedState now runs on chat open and on session-pipe gap.
- Permission actions: flavor-exact bodies mirroring PermissionFooter.tsx --
  claude {} / allowTools (Bash(cmd)) / mode:acceptEdits, codex-family
  decision approved / approved_for_session / abort via deny -- plus
  AskUserQuestion flat answers (option cards, Other free text, no-questions
  fallback, cursor stable ids) and request_user_input nested answers
  (user_note suffix, required validation); optimistic Resolving /
  AlreadyHandled (404/409) overrides settled by the agentState patch.
  ChatPipeline now re-attaches the window row's client status so failed
  user rows actually render the retry affordance (web normalize.ts parity;
  the Android reference misses this overlay).
- Session config sheet (toolbar gear): catalog-driven permission-mode picker
  with tones, claude static model/effort catalogs (ModelCatalog port), codex
  models via new GET /sessions/:id/codex-models endpoint + wire types with
  per-model reasoning efforts; optimistic detail updates
  (SessionListStore.updateDetailLocal, new) rolled forward to server truth
  on error.
- Lifecycle: VisibilityReporter posts POST /api/visibility per tracked
  handshake subscriptionId on scene-phase flips (404 prunes); ChatSession
  exposes its subscriptionId and feeds the reporter; the global SSE pipe was
  already HubSession-lifetime on iOS.
- Tests: ChatInteractorTests transcribes the Android interaction suite
  against the real APIClient/AuthManager/SessionListStore/window registry
  with only HTTP scripted -- canonical JSON bodies asserted byte-for-byte
  (send/approve/deny/config), optimistic send happy/fail/retry, 409 resume
  both id paths, cancel invoked-race, steer reconcile, edit prefill,
  override lifecycle, config optimistic + rollback, drafts, abort.
2026-08-18 10:33:04 +08:00
weishu a4355d1837 feat(ios): read-only chat assembly (A-M2f)
Wires the merged M2 pieces into a usable pair -> list -> live chat flow:

- pbxproj: link HapiUI into the app target (UUIDs C7/C8, mirroring the
  existing HapiProtocol/HapiClient product references; only project change).
- HubSession: MessageWindowControllers registry + per-chat ChatSession
  factory with in-memory per-session SSE resume cursors and scene-phase
  forwarding to the active chat.
- ChatSession (app): session-scope SSEClient while the chat is open;
  window opened/activated before subscribing; a single consume task awaits
  every event into the window actor, giving Android-channel-equivalent
  arrival-order ingestion; session-updated and friends route through the
  shared SyncEventRouter (detail patching), session-removed clears the
  window, a gap handshake triggers full resync + detail refetch + catch-up
  tail sync; stop hands the cursor back for seamless reopen.
- ChatPipeline (HapiClient): actor running the reduction path off-main -
  queued-row filter, normalization memoized by row instance identity,
  reduce + buildVisibleChatBlocks with previousGroups-stable group ids.
- ChatModel: window state + detail agentState + machine labels -> serial
  ~100 ms-coalesced pipeline loop (first run immediate, publishes strictly
  ordered), header title cascade, loading/empty/error states, loadOlder /
  retry, last-seen stamping on every update.
- ChatView: bottom-anchored ScrollView/LazyVStack (defaultScrollAnchor +
  scrollPosition(id:)), auto-stick at bottom, new-messages pill, top
  sentinel paging with documented scroll re-anchoring, degraded banners.
- Block views: user bubble w/ attachment chips, agent markdown, collapsed
  reasoning, tool cards (knownTools.tsx-parity presentation, status chip,
  read-only permission row w/ pending banner, per-tool bodies: terminal,
  before/after edits, write content, unified-diff detection -> DiffTextView,
  checklists, read-only questions, pretty-JSON fallback, children rail),
  tool groups, centered event rows (presentation.ts port), cli output,
  generated images (authed bytes + in-memory cache + full-screen viewer),
  codex review verdict card.
- Navigation/links: session list now pushes ChatView; app-level
  \.hapiOpenURL handler (https/http -> SFSafariViewController, custom
  schemes confirm first, hapi-file:// -> M4 placeholder) + theme injection.
- Tests: ChatPipelineTests drive the runner with fixture-derived windows
  (stable ids, memo-stable recomputes, queued filter, group-id stability
  across an older-page arrival).
- README: M2f app-layer description.
2026-08-17 20:55:36 +08:00