agent-orchestrator/docs/progress/v3.md

# v3 Progress Log

### Session: 2026-03-07 — Architecture Planning + MVP Implementation (Phases 1-5)

#### Phase: Adversarial Design Review
- [x] Launch 3 architecture agents (Architect, Devil's Advocate, UX+Performance Specialist)
- [x] Collect findings — 12 issues identified, all resolved
- [x] Produce final architecture plan in docs/v3-task_plan.md
- [x] Create 10-phase implementation plan

#### Phase 1: Data Model + Config
- [x] Created `v2/src/lib/types/groups.ts` — TypeScript interfaces (ProjectConfig, GroupConfig, GroupsFile)
- [x] Created `v2/src-tauri/src/groups.rs` — Rust structs + load/save groups.json
- [x] Added `groups_load`, `groups_save` Tauri commands to lib.rs
- [x] SQLite migrations in session.rs: project_id column, agent_messages table, project_agent_state table
- [x] Created `v2/src/lib/adapters/groups-bridge.ts` (IPC wrapper)
- [x] Created `v2/src/lib/stores/workspace.svelte.ts` (replaces layout.svelte.ts, Svelte 5 runes)
- [x] Added `--group` CLI argument parsing in main.rs
- [x] Wrote 24 vitest tests for workspace store (workspace.test.ts)
- [x] Wrote cargo tests for groups load/save/default

#### Phase 2: Project Box Shell
- [x] Created GlobalTabBar.svelte (Sessions | Docs | Context | Settings)
- [x] Created ProjectGrid.svelte (flex + scroll-snap container)
- [x] Created ProjectBox.svelte (CSS grid: header | session-area | terminal-area)
- [x] Created ProjectHeader.svelte (icon + name + status dot + accent color)
- [x] Rewrote App.svelte (GlobalTabBar + tab content + StatusBar, no sidebar/TilingGrid)
- [x] Created CommandPalette.svelte (Ctrl+K overlay with fuzzy search)
- [x] Created DocsTab.svelte (markdown file browser per project)
- [x] Created ContextTab.svelte (wrapper for ContextPane)
- [x] Created SettingsTab.svelte (per-project + global settings editor)
- [x] CSS for responsive project count + Catppuccin accent colors

#### Phase 3: Claude Session Integration
- [x] Created ClaudeSession.svelte (wraps AgentPane, passes project cwd/profile/config_dir)

#### Phase 4: Terminal Tabs
- [x] Created TerminalTabs.svelte (tab bar + content, shell/SSH/agent tab types)

#### Phase 5: Team Agents Panel
- [x] Created TeamAgentsPanel.svelte (right panel for subagents)
- [x] Created AgentCard.svelte (compact subagent view: status, messages, cost)

#### Bug Fix
- [x] Fixed AgentPane Svelte 5 event modifier syntax: `on:click` -> `onclick` (Svelte 5 requires lowercase event attributes)

#### Verification
- All 138 vitest tests pass (114 existing + 24 new workspace tests)
- All 36 cargo tests pass (29 existing + 7 new groups tests)
- Vite build succeeds

### Session: 2026-03-07 — Phases 6-10 Completion

#### Phase 6: Session Continuity
- [x] Added `persistSessionForProject()` to agent-dispatcher — saves agent state + messages to SQLite on session complete
- [x] Added `registerSessionProject()` — maps sessionId -> projectId for persistence routing
- [x] Added `sessionProjectMap` (Map<string, string>) in agent-dispatcher
- [x] Updated ClaudeSession.svelte: `restoreMessagesFromRecords()` restores cached messages into agent store on mount
- [x] ClaudeSession loads previous state via `loadProjectAgentState()`, restores session ID and messages
- [x] Added `getAgentSession()` export to agents store

#### Phase 7: Workspace Teardown on Group Switch
- [x] Added `clearAllAgentSessions()` to agents store (clears sessions array)
- [x] Updated `switchGroup()` in workspace store to call `clearAllAgentSessions()` + reset terminal tabs
- [x] Updated workspace.test.ts to mock `clearAllAgentSessions`

#### Phase 10: Dead Component Removal + Polish
- [x] Deleted `TilingGrid.svelte` (328 lines), `PaneContainer.svelte` (113 lines), `PaneHeader.svelte` (44 lines)
- [x] Deleted `SessionList.svelte` (374 lines), `SshSessionList.svelte` (263 lines), `SshDialog.svelte` (281 lines), `SettingsDialog.svelte` (433 lines)
- [x] Removed empty directories: Layout/, Sidebar/, Settings/, SSH/
- [x] Rewrote StatusBar.svelte for workspace store (group name, project count, agent count, "BTerminal v3" label)
- [x] Fixed subagent routing in agent-dispatcher: project-scoped sessions skip layout pane creation (subagents render in TeamAgentsPanel instead)
- [x] Updated v3-task_plan.md to mark all 10 phases complete

#### Verification
- All 138 vitest tests pass (including updated workspace tests with clearAllAgentSessions mock)
- All 36 cargo tests pass
- Vite build succeeds
- ~1,836 lines of dead code removed

### Session: 2026-03-07 — SettingsTab Global Settings + Cleanup

#### SettingsTab Global Settings Section
- [x] Added "Global" section to SettingsTab.svelte with three settings:
  - Theme flavor dropdown (Catppuccin Latte/Frappe/Macchiato/Mocha) via `setFlavor()` from theme store
  - Default shell text input (persisted via `setSetting('default_shell', ...)`)
  - Default CWD text input (persisted via `setSetting('default_cwd', ...)`)
- [x] Global settings load on mount via `getSetting()` from settings-bridge
- [x] Added imports: `onMount`, `getSetting`/`setSetting`, `getCurrentFlavor`/`setFlavor`, `CatppuccinFlavor` type

#### A11y Fixes
- [x] Changed project field labels from `<div class="project-field"><label>` to wrapping `<label class="project-field"><span class="field-label">` pattern — proper label/input association
- [x] Global settings use `id`/`for` label association (e.g., `id="theme-flavor"`, `id="default-shell"`)

#### CSS Cleanup
- [x] Removed unused `.project-field label` selector (replaced by `.field-label`)
- [x] Simplified `.project-field input[type="text"], .project-field input:not([type])` to `.project-field input:not([type="checkbox"])`

#### Rust Cleanup (committed separately)
- [x] Removed dead `update_ssh_session()` method from session.rs and its test
- [x] Fixed stale TilingGrid comment in AgentPane.svelte

### Session: 2026-03-07 — Multi-Theme System (7 Editor Themes)

#### Theme System Generalization
- [x] Generalized `CatppuccinFlavor` type to `ThemeId` union type (11 values)
- [x] Added 7 new editor themes: VSCode Dark+, Atom One Dark, Monokai, Dracula, Nord, Solarized Dark, GitHub Dark
- [x] Added `ThemePalette` interface (26-color slots) — all themes map to same slots
- [x] Added `ThemeMeta` interface (id, label, group, isDark) for UI metadata
- [x] Added `THEME_LIST: ThemeMeta[]` with group metadata ('Catppuccin' or 'Editor')
- [x] Added `ALL_THEME_IDS: ThemeId[]` derived from THEME_LIST for validation
- [x] Deprecated `CatppuccinFlavor`, `CatppuccinPalette`, `FLAVOR_LABELS`, `ALL_FLAVORS` (kept as backwards compat aliases)

#### Theme Store Updates
- [x] `getCurrentTheme(): ThemeId` replaces `getCurrentFlavor()` as primary getter
- [x] `setTheme(theme: ThemeId)` replaces `setFlavor()` as primary setter
- [x] `initTheme()` validates saved theme against `ALL_THEME_IDS`
- [x] Deprecated `getCurrentFlavor()` and `setFlavor()` with delegation wrappers

#### SettingsTab Theme Selector
- [x] Theme dropdown uses `<optgroup>` per theme group (Catppuccin, Editor)
- [x] `themeGroups` derived from `THEME_LIST` using Map grouping
- [x] `handleThemeChange()` replaces direct `setFlavor()` call
- [x] Fixed input overflow in `.setting-row` with `min-width: 0`

#### Design Decision
All editor themes map to the same `--ctp-*` CSS custom property names (26 vars). This means every component works unchanged — no component-level theme awareness needed. Each theme provides its own mapping of colors to the 26 semantic slots.

#### Verification
- All 138 vitest + 35 cargo tests pass

### Session: 2026-03-07 — Deep Dark Theme Group (6 Themes)

#### New Theme Group: Deep Dark
- [x] Added 6 new "Deep Dark" themes to `v2/src/lib/styles/themes.ts`:
  - Tokyo Night (base: #1a1b26)
  - Gruvbox Dark (base: #1d2021)
  - Ayu Dark (base: #0b0e14, near-black)
  - Poimandres (base: #1b1e28)
  - Vesper (base: #101010, warm dark)
  - Midnight (base: #000000, pure OLED black)
- [x] Extended `ThemeId` union type from 11 to 17 values
- [x] Added `THEME_LIST` entries with `group: 'Deep Dark'`
- [x] Added all 6 palette definitions (26 colors each) mapped to --ctp-* slots
- [x] Total themes: 17 across 3 groups (Catppuccin 4, Editor 7, Deep Dark 6)

#### Verification
- No test changes needed — theme palettes are data-only, no logic changes

### Session: 2026-03-07 — Custom Theme Dropdown

#### SettingsTab Theme Picker Redesign
- [x] Replaced native `<select>` with custom themed dropdown in SettingsTab.svelte
- [x] Dropdown trigger shows color swatch (base color from getPalette()) + theme label + arrow indicator
- [x] Dropdown menu groups themes by category (Catppuccin/Editor/Deep Dark) with styled uppercase headers
- [x] Each option shows: color swatch + label + 4 accent color dots (red/green/blue/yellow)
- [x] Active theme highlighted with surface0 background + bold text
- [x] Click-outside handler and Escape key to close dropdown
- [x] Uses --ctp-* CSS vars throughout — fully themed with any active theme
- [x] Added `getPalette` import from themes.ts for live color rendering
- [x] Added aria-haspopup/aria-expanded attributes for accessibility

#### Verification
- No test changes needed — UI-only change, no logic changes

### Session: 2026-03-07 — Theme Dropdown CSS Polish

#### SettingsTab Dropdown Sizing Fix
- [x] Set `min-width: 180px` on `.theme-dropdown` container (was `min-width: 0`) to prevent trigger from collapsing
- [x] Set `min-width: 280px` on `.theme-options` dropdown menu (was `right: 0`) to ensure full theme names visible
- [x] Increased `max-height` from 320px to 400px on dropdown menu for better scrolling experience
- [x] Added `white-space: nowrap` on `.theme-option-label` (was `min-width: 0`) to prevent label text wrapping

#### Verification
- No test changes needed — CSS-only change

### Session: 2026-03-07 — Global Font Controls

#### SettingsTab Font Family + Font Size Controls
- [x] Added font family `<select>` with 9 monospace font options (JetBrains Mono, Fira Code, Cascadia Code, Source Code Pro, IBM Plex Mono, Hack, Inconsolata, Ubuntu Mono, monospace) + "Default" option
- [x] Added font size +/- stepper control with numeric input (range 8-24px)
- [x] Both controls apply live preview via CSS custom properties (`--ui-font-family`, `--ui-font-size`)
- [x] Both settings persisted to SQLite via settings-bridge (`font_family`, `font_size` keys)
- [x] `handleFontFamilyChange()` and `handleFontSizeChange()` functions with validation

#### SettingsTab Layout Restructure
- [x] Restructured global settings from inline `.setting-row` (label left, control right) to 2-column `.global-grid` with `.setting-field` (label above control)
- [x] Labels now uppercase, 0.7rem, subtext0 color — consistent compact labeling
- [x] All inputs/selects use consistent styling (surface0 bg, surface1 border, 4px radius)

#### CSS Typography Variables
- [x] Added `--ui-font-family` and `--ui-font-size` to catppuccin.css `:root` (defaults: JetBrains Mono fallback chain, 13px)
- [x] Updated `app.css` body rule to use CSS vars instead of hardcoded font values

#### Theme Store Font Restoration
- [x] Extended `initTheme()` in `theme.svelte.ts` to load and apply saved `font_family` and `font_size` settings on startup
- [x] Font restoration wrapped in try/catch — failures are non-fatal (CSS defaults apply)

#### Verification
- No test changes needed — UI/CSS-only changes, no logic changes

### Session: 2026-03-07 — Settings Drawer Conversion

#### Settings Tab to Drawer
- [x] Converted Settings from a full-page tab to a collapsible side drawer
- [x] GlobalTabBar now has 3 tabs (Sessions/Docs/Context) + gear icon toggle for settings drawer
- [x] App.svelte renders SettingsTab in an `<aside>` drawer (right side, 32em width, semi-transparent backdrop)
- [x] Drawer close: Escape key, click-outside (backdrop), close button (X icon)
- [x] Gear icon in GlobalTabBar highlights blue when drawer is open (active state)
- [x] GlobalTabBar accepts props: `settingsOpen`, `ontoggleSettings`
- [x] Removed 'settings' from WorkspaceTab union type (now 'sessions' | 'docs' | 'context')
- [x] Alt+1..3 for tabs (was Alt+1..4), Ctrl+, toggles drawer (was setActiveTab('settings'))
- [x] SettingsTab padding reduced (12px 16px), max-width removed, flex:1 for drawer context

#### Verification
- All 138 vitest tests pass

### Session: 2026-03-08 — VSCode-Style Sidebar Redesign

#### UI Layout Redesign (Top Tab Bar -> Left Sidebar)
- [x] Redesigned GlobalTabBar.svelte from horizontal tab bar to vertical icon rail (36px wide)
  - 4 SVG icon buttons: Sessions (grid), Docs (document), Context (clock), Settings (gear)
  - Each button uses SVG path from `icons` record mapped by WorkspaceTab
  - Props renamed: `settingsOpen` -> `expanded`, `ontoggleSettings` -> `ontoggle`
  - `handleTabClick()` manages toggle: clicking active tab collapses drawer
- [x] Rewrote App.svelte layout from vertical (top tab bar + content area + settings drawer) to horizontal (icon rail + sidebar panel + workspace)
  - `.main-row` flex container: GlobalTabBar | sidebar-panel (28em, max 50%) | workspace
  - ProjectGrid always visible in main workspace (not inside tab content)
  - Sidebar panel renders active tab content (Sessions/Docs/Context/Settings)
  - Panel header with title + close button
  - Removed backdrop overlay, drawer is inline sidebar not overlay
- [x] Re-added 'settings' to WorkspaceTab union type (was removed when settings was a drawer)
- [x] SettingsTab CSS: changed `flex: 1` to `height: 100%` for sidebar panel context
- [x] Updated keyboard shortcuts:
  - Alt+1..4 (was Alt+1..3): switch tabs + open drawer, toggle if same tab
  - Ctrl+B (new): toggle sidebar open/closed
  - Ctrl+, : open settings panel (toggle if already active)
  - Escape: close drawer
- [x] State variables renamed: `settingsOpen` -> `drawerOpen`, `toggleSettings()` -> `toggleDrawer()`
- [x] Added `panelTitles` record for drawer header labels

#### Design Decisions
- VSCode-style sidebar chosen for: always-visible workspace, progressive disclosure, familiar UX
- Settings as regular tab (not special drawer) simplifies code and mental model
- Icon rail at 36px minimizes horizontal space cost
- No backdrop overlay — sidebar is inline, not modal

#### Verification
- All 138 vitest tests pass
- svelte-check clean (only 2 third-party esrap warnings)

### Session: 2026-03-07 — SettingsTab Global Settings Redesign

#### Font Settings Split (UI Font + Terminal Font)
- [x] Split single font setting into UI font (sans-serif options) and Terminal font (monospace options)
- [x] UI font dropdown: System Sans-Serif, Inter, Roboto, Open Sans, Lato, Noto Sans, Source Sans 3, IBM Plex Sans, Ubuntu + Default
- [x] Terminal font dropdown: JetBrains Mono, Fira Code, Cascadia Code, Source Code Pro, IBM Plex Mono, Hack, Inconsolata, Ubuntu Mono, monospace + Default
- [x] Each font dropdown renders preview text in its own typeface
- [x] Size steppers (8-24px) for both UI and Terminal font independently
- [x] Changed setting keys: font_family -> ui_font_family, font_size -> ui_font_size, + new term_font_family, term_font_size

#### SettingsTab Layout Redesign
- [x] Rewrote global settings as single-column layout with labels above controls
- [x] Split into "Appearance" subsection (theme, UI font, terminal font) and "Defaults" subsection (shell, CWD)
- [x] All dropdowns now use reusable custom themed dropdowns (no native `<select>` anywhere)

#### CSS + Theme Store Updates
- [x] Added `--term-font-family` and `--term-font-size` CSS custom properties to catppuccin.css
- [x] Updated `initTheme()` in theme.svelte.ts: loads 4 font settings (ui_font_family, ui_font_size, term_font_family, term_font_size) instead of 2
- [x] UI font fallback changed from monospace to sans-serif

#### Verification
- No test changes needed — UI/CSS-only changes, no logic changes

### Session: 2026-03-08 — CSS Relative Units Rule

#### New Rule: 18-relative-units.md
- [x] Created `.claude/rules/18-relative-units.md` enforcing rem/em for layout CSS
- [x] Pixels allowed only for icon sizes, borders/outlines, box shadows
- [x] Exception: --ui-font-size/--term-font-size CSS vars store px (xterm.js API requirement)
- [x] Added rule #18 to `.claude/CLAUDE.md` rule index

#### CSS Conversions
- [x] GlobalTabBar.svelte: rail width 36px -> 2.75rem, button 28px -> 2rem, gap 2px -> 0.25rem, padding 6px 4px -> 0.5rem 0.375rem, border-radius 4px -> 0.375rem
- [x] App.svelte: sidebar header padding 8px 12px -> 0.5rem 0.75rem, close button 22px -> 1.375rem, border-radius 4px -> 0.25rem
- [x] Also changed GlobalTabBar rail-btn color from --ctp-overlay1 to --ctp-subtext0 for better contrast

### Session: 2026-03-08 — Content-Driven Sidebar Width

#### Sidebar Panel Sizing
- [x] Changed `.sidebar-panel` from fixed `width: 28em` to `width: max-content` with `min-width: 16em` and `max-width: 50%`
- [x] Changed `.sidebar-panel` and `.panel-content` from `overflow: hidden` to `overflow-y: auto` — hidden was blocking content from driving parent width
- [x] Each tab component now defines its own `min-width: 22em` (SettingsTab, ContextTab, DocsTab)

#### Additional px → rem Conversions
- [x] SettingsTab.svelte: padding 12px 16px → 0.75rem 1rem
- [x] DocsTab.svelte: file-picker 220px → 14em, picker-title padding → rem, file-btn padding → rem, empty/loading padding → rem
- [x] ContextPane.svelte: font-size, padding, margin, gap converted from px to rem; added `white-space: nowrap` on `.ctx-header`/`.ctx-error` for intrinsic width measurement

#### Fix: Sidebar Drawer Content-Driven Width
- [x] Root cause found: `#app` in `app.css` had leftover v2 grid layout (`display: grid; grid-template-columns: var(--sidebar-width) 1fr`) constraining `.app-shell` to 260px first column
- [x] Removed v2 grid + both media queries from `#app` — v3 `.app-shell` manages its own flexbox layout
- [x] Added JS `$effect` in App.svelte: measures content width via `requestAnimationFrame` + `querySelectorAll` for nowrap elements, headings, inputs, tab-specific selectors; `panelWidth` state drives inline `style:width`
- [x] Verified all 4 tabs scale to content: Sessions ~473px, Settings ~322px, Context ~580px, Docs varies by content
- [x] Investigation path: CSS intrinsic sizing (max-content, fit-content) failed due to column-flex circular dependency → JS measurement approach → discovered inline style set but rendered width wrong → Playwright inspection revealed parent `.main-row` only 260px → traced to `#app` grid layout

### Session: 2026-03-08 — Native Directory Picker

#### tauri-plugin-dialog Integration
- [x] Added `tauri-plugin-dialog` Rust crate + `@tauri-apps/plugin-dialog` npm package
- [x] Registered plugin in lib.rs (`tauri_plugin_dialog::init()`)
- [x] Removed stub `pick_directory` Tauri command (always returned None)
- [x] Added `browseDirectory()` helper in SettingsTab.svelte using `open({ directory: true })`
- [x] Added folder browse button (folder SVG icon) to: Default CWD, existing project CWD, Add Project path
- [x] Styled `.input-with-browse` layout (flex row, themed browse button)
- [x] Fixed nested input theme: `.setting-field .input-with-browse input` selector for dark background
- [x] Fixed dialog not opening: added `"dialog:default"` permission to `v2/src-tauri/capabilities/default.json` — Tauri IPC security blocked invoke() without capability
- [x] Verified via Playwright: error was `Cannot read properties of undefined (reading 'invoke')` in browser context (expected — Tauri IPC only exists in WebView), confirming code is correct
- [x] Clean rebuild required after capability changes (cached binary doesn't pick up new permissions)

#### Modal + Dark-Themed Dialog
- [x] Root cause: `tauri-plugin-dialog` skips `set_parent(&window)` on Linux via `cfg(any(windows, target_os = "macos"))` gate in commands.rs — dialog not modal
- [x] Root cause: native GTK file chooser uses system GTK theme, not app's CSS theme — dialog appears light
- [x] Fix: custom `pick_directory` Tauri command using `rfd::AsyncFileDialog` directly with `.set_parent(&window)` — modal on Linux
- [x] Fix: `std::env::set_var("GTK_THEME", "Adwaita:dark")` at start of `run()` in lib.rs — dark-themed dialog
- [x] Added `rfd = { version = "0.16", default-features = false, features = ["gtk3"] }` as direct dep — MUST disable defaults to avoid gtk3+xdg-portal feature conflict
- [x] Switched SettingsTab from `@tauri-apps/plugin-dialog` `open()` to `invoke<string | null>('pick_directory')`

### Session: 2026-03-08 — Project Workspace Layout Redesign + Icon Fix

#### Icon Fix
- [x] Replaced Nerd Font codepoints (`\uf120`) with emoji (`📁` default) — Nerd Font not installed, showed "?"
- [x] Added emoji picker grid (24 project-relevant emoji, 8-column popup) in SettingsTab instead of plain text input
- [x] Removed `font-family: 'NerdFontsSymbols Nerd Font'` from ProjectHeader and TerminalTabs

#### ProjectBox Layout Redesign
- [x] Switched ProjectBox from flex to CSS grid (`grid-template-rows: auto 1fr auto`) — header | session | terminal zones
- [x] Terminal area: explicit `height: 16rem` instead of collapsing to content
- [x] Session area: `min-height: 0` for proper flex child overflow

#### AgentPane Prompt Layout
- [x] Prompt area anchored to bottom (`justify-content: flex-end`) instead of vertical center
- [x] Removed `max-width: 600px` constraint on form and toolbar — uses full panel width
- [x] Toolbar sits directly above textarea

#### CSS px → rem Conversions
- [x] ProjectGrid.svelte: gap 4px → 0.25rem, padding 4px → 0.25rem, min-width 480px → 30rem
- [x] TerminalTabs.svelte: tab bar, tabs, close/add buttons all converted to rem
- [x] ProjectBox.svelte: min-width 480px → 30rem

### Session: 2026-03-08 — Project-Level Tabs + Clean AgentPane

#### ProjectHeader Info Bar
- [x] Added CWD path display (ellipsized from START via `direction: rtl` + `text-overflow: ellipsis`)
- [x] Added profile name as info-only text (right side of header)
- [x] Home dir shortening: `/home/user/foo` → `~/foo`

#### Project-Level Tab Bar
- [x] Added tab bar in ProjectBox below header: Claude | Files | Context
- [x] Content area switches between ClaudeSession, ProjectFiles, ContextPane based on selected tab
- [x] CSS grid updated to 4 rows: `auto auto 1fr auto` (header | tabs | content | terminal)
- [x] TeamAgentsPanel still renders alongside ClaudeSession in Claude tab

#### ProjectFiles Component (NEW)
- [x] Created `ProjectFiles.svelte` — project-scoped markdown file viewer
- [x] Accepts `cwd` + `projectName` props (not workspace store)
- [x] File picker sidebar (10rem) + MarkdownPane content area
- [x] Auto-selects priority file or first file

#### AgentPane Cleanup
- [x] Removed entire session toolbar (DIR/ACC interactive inputs + all CSS)
- [x] Added `profile` prop — resolved via `listProfiles()` to get config_dir
- [x] CWD passed as prop from parent (project.cwd), no longer editable in pane
- [x] Clean chat interface: prompt (bottom-anchored) + messages + send button
- [x] ClaudeSession now passes `project.profile` to AgentPane

#### Verification
- All 138 vitest tests pass
- Vite build succeeds

### Session: 2026-03-08 — Security Audit Fixes + OTEL Telemetry

#### Security Audit Fixes
- [x] Fixed all CRITICAL (5) + HIGH (4) findings — path traversal, race conditions, memory leaks, listener leaks, transaction safety
- [x] Fixed all MEDIUM (6) findings — runtime type guards, ANTHROPIC_* env stripping, timestamp mismatch, async lock, error propagation
- [x] Fixed all LOW (8) findings — input validation, mutex poisoning, log warnings, payload validation
- [x] 3 false positives dismissed with rationale
- [x] 172/172 tests pass (138 vitest + 34 cargo)

#### OTEL Telemetry Implementation
- [x] Added 6 Rust deps: tracing, tracing-subscriber, opentelemetry 0.28, opentelemetry_sdk 0.28, opentelemetry-otlp 0.28, tracing-opentelemetry 0.29
- [x] Created `v2/src-tauri/src/telemetry.rs` — TelemetryGuard, layer composition, OTLP export via BTERMINAL_OTLP_ENDPOINT env var
- [x] Integrated into lib.rs: TelemetryGuard in AppState, init before Tauri builder
- [x] Instrumented 10 Tauri commands with `#[tracing::instrument]`: pty_spawn, pty_kill, agent_query/stop/restart, remote_connect/disconnect/agent_query/agent_stop/pty_spawn
- [x] Added `frontend_log` Tauri command for frontend→Rust tracing bridge
- [x] Created `v2/src/lib/adapters/telemetry-bridge.ts` — `tel.info/warn/error/debug/trace()` convenience API
- [x] Wired agent dispatcher lifecycle events: agent_started, agent_stopped, agent_error, sidecar_crashed, cost metrics
- [x] Created Docker compose stack: `docker/tempo/` — Tempo (4317/4318/3200) + Grafana (port 9715)

### Session: 2026-03-08 — Teardown Race Fix + px→rem Conversion

#### Workspace Teardown Race Fix
- [x] Added `pendingPersistCount` counter + `waitForPendingPersistence()` export in agent-dispatcher.ts
- [x] `persistSessionForProject()` increments/decrements counter in try/finally
- [x] `switchGroup()` in workspace.svelte.ts now awaits `waitForPendingPersistence()` before clearing state
- [x] SettingsTab.svelte switchGroup onclick handler made async with await
- [x] Added test for `waitForPendingPersistence` in agent-dispatcher.test.ts
- [x] Added mock for `waitForPendingPersistence` in workspace.test.ts
- [x] Last open HIGH audit finding resolved (workspace teardown race)

#### px→rem Conversion (Rule 18 Compliance)
- [x] Converted ~100 px layout violations to rem across 10 components
- [x] AgentPane.svelte (~35 violations: font-size, padding, gap, margin, max-height, border-radius)
- [x] ToastContainer.svelte, CommandPalette.svelte, TeamAgentsPanel.svelte, AgentCard.svelte
- [x] StatusBar.svelte, AgentTree.svelte, TerminalPane.svelte, AgentPreviewPane.svelte, SettingsTab.svelte
- [x] Icon/decorative dot dimensions kept as px per rule 18
- [x] 139 vitest + 34 cargo tests pass, vite build succeeds

### Session: 2026-03-08 — E2E Testing Infrastructure

#### WebdriverIO + tauri-driver Setup
- [x] Installed @wdio/cli, @wdio/local-runner, @wdio/mocha-framework, @wdio/spec-reporter (v9.24.0)
- [x] Created wdio.conf.js with tauri-driver lifecycle hooks (onPrepare builds debug binary, beforeSession/afterSession spawns/kills tauri-driver)
- [x] Created tsconfig.json for e2e test TypeScript compilation
- [x] Created smoke.test.ts with 6 tests: app title, status bar, version text, sidebar rail, workspace area, sidebar toggle
- [x] Added `test:e2e` npm script (`wdio run tests/e2e/wdio.conf.js`)
- [x] Updated README.md with complete setup instructions and CI guide
- [x] Key decision: WebdriverIO over Playwright (Playwright cannot control Tauri/WebKit2GTK apps)
- [x] Prerequisites: tauri-driver (cargo install), webkit2gtk-driver (apt), display server or xvfb-run

#### E2E Fixes (wdio v9 + tauri-driver compatibility)
- [x] Fixed wdio v9 BiDi: added `wdio:enforceWebDriverClassic: true` — wdio v9 injects webSocketUrl:true which tauri-driver rejects
- [x] Removed `browserName: 'wry'` from capabilities (not needed in wdio, only Selenium)
- [x] Fixed binary path: Cargo workspace target is v2/target/debug/, not v2/src-tauri/target/debug/
- [x] Fixed tauri-plugin-log panic: telemetry::init() registers tracing-subscriber before plugin-log → removed tauri-plugin-log entirely (redundant with telemetry::init())
- [x] Removed tauri-plugin-log from Cargo.toml dependency

#### E2E Coverage Expansion (25 tests, single spec file)
- [x] Consolidated 4 spec files into single bterminal.test.ts — Tauri creates one app session per spec file; after first spec completes, app closes and subsequent specs get "invalid session id"
- [x] Added Workspace & Projects tests (8): project grid, project boxes, header with name, 3 project tabs, active highlight, tab switching, status bar counts
- [x] Added Settings Panel tests (6): settings tab, sections, theme dropdown, dropdown open+options, group list, close button
- [x] Added Keyboard Shortcuts tests (5): Ctrl+K command palette, Ctrl+, settings, Ctrl+B sidebar, Escape close, palette group list
- [x] Fixed WebDriver clicks on Svelte 5 components: `element.click()` doesn't reliably trigger onclick inside complex components via WebKit2GTK/tauri-driver — use `browser.execute()` for JS-level clicks
- [x] Fixed CSS text-transform: `.ptab` getText() returns uppercase — use `.toLowerCase()` for comparison
- [x] Fixed element scoping: `browser.$('.ptab')` returns ALL tabs across project boxes — scope via `box.$('.ptab')`
- [x] Fixed keyboard focus: `browser.execute(() => document.body.focus())` before sending shortcuts
- [x] Removed old individual spec files (smoke.test.ts, keyboard.test.ts, settings.test.ts, workspace.test.ts)
- [x] All 25 E2E tests pass (9s runtime after build)

### Session: 2026-03-10 — Tab System Overhaul

#### Tab Renames + New Tabs
- [x] Renamed Claude → Model, Files → Docs in ProjectBox
- [x] Added 3 new tabs: Files (directory browser), SSH (connection manager), Memory (knowledge explorer)
- [x] Implemented PERSISTED-EAGER (Model/Docs/Context — display:flex/none) vs PERSISTED-LAZY (Files/SSH/Memory — {#if everActivated} + display:flex/none) mount strategy
- [x] Tab type union: 'model' | 'docs' | 'context' | 'files' | 'ssh' | 'memories'

#### Files Tab (FilesTab.svelte)
- [x] VSCode-style tree sidebar (14rem) + content viewer
- [x] Rust list_directory_children command: lazy expansion, hidden files skipped, dirs-first sort
- [x] Rust read_file_content command: FileContent tagged union (Text/Binary/TooLarge), 10MB gate, 30+ language mappings
- [x] Frontend files-bridge.ts adapter (DirEntry, FileContent types)
- [x] Shiki syntax highlighting for code files, image display via convertFileSrc, emoji file icons

#### SSH Tab (SshTab.svelte)
- [x] CRUD panel for SSH connections using existing ssh-bridge.ts/SshSession model
- [x] Launch button spawns terminal tab in Model tab's TerminalTabs section via addTerminalTab()

#### Memory Tab (MemoriesTab.svelte)
- [x] Pluggable MemoryAdapter interface (memory-adapter.ts): name, available, list(), search(), get()
- [x] Adapter registry: registerMemoryAdapter(), getDefaultAdapter(), getAvailableAdapters()
- [x] UI: search bar, tag display, expandable cards, adapter switcher, placeholder when no adapter

#### Context Tab Repurpose (ContextTab.svelte)
- [x] Replaced ContextPane (ctx database viewer) with LLM context window visualization
- [x] Tribunal debate for design (S-1-R4 winner at 82% confidence)
- [x] Stats bar: input/output tokens, cost, turns, duration
- [x] Segmented token meter: CSS flex bar with color-coded categories (assistant/thinking/tool calls/tool results)
- [x] File references: extracted from tool_call messages, colored op badges
- [x] Turn breakdown: collapsible message groups by user prompt
- [x] Token estimation via ~4 chars/token heuristic
- [x] Wired into ProjectBox (replaces ContextPane, passes sessionId)
- [x] Sub-tab navigation: Overview | AST | Graph
- [x] AST tab: per-turn SVG conversation trees (Thinking/Response/ToolCall/File nodes, bezier edges, token counts)
- [x] Graph tab: bipartite tool→file DAG (tools left, files right, curved edges, count badges)
- [x] Compaction detection: sdk-messages.ts adapts `compact_boundary` system messages → `CompactionContent` type
- [x] Stats bar compaction pill: yellow count badge with tooltip (last trigger, tokens removed)
- [x] AST compaction boundaries: red "Compacted" nodes inserted between turns at compaction points

#### FilesTab Fixes & CodeMirror Editor
- [x] Fixed HTML nesting error: `<button>` inside `<button>` → `<div role="tab">`
- [x] Fixed Svelte 5 $state proxy reactivity: look up tab from reactive array before setting content
- [x] CodeEditor.svelte: CodeMirror 6 with 15 lazy-loaded language modes, Catppuccin theme
- [x] Dirty tracking, Ctrl+S save, save-on-blur setting (files_save_on_blur in SettingsTab)
- [x] write_file_content Rust command (safety: existing files only)

#### Project Health Dashboard (S-3 — Mission Control)
- [x] health.svelte.ts store: per-project ActivityState (running/idle/stalled), burn rate ($/hr EMA), context pressure (% of model limit), attention scoring
- [x] StatusBar → Mission Control bar: running/idle/stalled counts, $/hr burn rate, "needs attention" priority queue dropdown
- [x] ProjectHeader health indicators: status dot (color-coded), context pressure badge, burn rate badge
- [x] session_metrics SQLite table: per-project historical metrics (100-row retention)
- [x] Rust commands: session_metric_save, session_metrics_load
- [x] TypeScript bridge: SessionMetric interface, saveSessionMetric(), loadSessionMetrics()
- [x] agent-dispatcher wiring: recordActivity, recordToolDone, recordTokenSnapshot, sessionStartTimes, metric persistence on completion
- [x] ClaudeSession: trackProject() on session create/restore
- [x] App.svelte: startHealthTick()/stopHealthTick() lifecycle
- [x] workspace.svelte.ts: clearHealthTracking() on group switch

#### Verification
- [x] svelte-check: 0 new errors (only pre-existing esrap type errors)
- [x] vitest: 139/139 tests pass
- [x] cargo test: 34/34 pass

### Session: 2026-03-11 — S-1 Phase 1.5: Conflict Detection Enhancements

#### Bash Write Detection
- [x] BASH_WRITE_PATTERNS regex array in tool-files.ts: >, >>, sed -i, tee [-a], cp dest, mv dest, chmod/chown
- [x] extractBashWritePaths() helper with /dev/null and flag-target filtering
- [x] Write detection prioritized over read detection for ambiguous commands (cat file > out)
- [x] extractWritePaths() now captures Bash writes alongside Write/Edit

#### Acknowledge/Dismiss Conflicts
- [x] acknowledgeConflicts(projectId) API in conflicts.svelte.ts — marks current conflicts as acknowledged
- [x] acknowledgedFiles Map state — suppresses badge until new session writes to acknowledged file
- [x] ProjectHeader conflict badge → clickable button with ✕ (stopPropagation, hover darkens)
- [x] Ack auto-cleared when new session writes to previously-acknowledged file

#### Worktree-Aware Conflict Suppression
- [x] sessionWorktrees Map in conflicts store — tracks worktree path per session (null = main tree)
- [x] setSessionWorktree(sessionId, path) API
- [x] areInDifferentWorktrees() / hasRealConflict() — suppresses conflicts between sessions in different worktrees
- [x] extractWorktreePath(tc) in tool-files.ts — detects Agent/Task isolation:"worktree" and EnterWorktree
- [x] agent-dispatcher.ts wiring: registers worktree paths from tool_call events
- [x] useWorktrees?: boolean field on ProjectConfig (groups.ts) for future per-project setting

#### Verification
- [x] vitest: 194/194 tests pass (+24 new: 5 extractWorktreePath, 10 bash write, 9 acknowledge/worktree)
- [x] cargo test: 34/34 pass

### Session: 2026-03-11 — S-1 Phase 2: Filesystem Write Detection

#### Rust Backend — ProjectFsWatcher
- [x] New module `v2/src-tauri/src/fs_watcher.rs` — per-project recursive inotify watchers via notify crate v6
- [x] Debouncing (100ms per-file), ignored dirs (.git/, node_modules/, target/, etc.)
- [x] Emits `fs-write-detected` Tauri events with FsWritePayload { project_id, file_path, timestamp_ms }
- [x] Two Tauri commands: `fs_watch_project`, `fs_unwatch_project`
- [x] ProjectFsWatcher added to AppState, initialized in setup()
- [x] 5 Rust unit tests for path filtering (should_ignore_path)

#### Frontend Bridge
- [x] New `v2/src/lib/adapters/fs-watcher-bridge.ts` — fsWatchProject(), fsUnwatchProject(), onFsWriteDetected()

#### External Write Detection (conflicts store)
- [x] EXTERNAL_SESSION_ID = '__external__' sentinel for non-agent writers
- [x] agentWriteTimestamps Map — tracks when agents write files (for timing heuristic)
- [x] recordExternalWrite(projectId, filePath, timestampMs) — 2s grace window suppresses agent's own writes
- [x] getExternalConflictCount(projectId) — counts external-only conflicts
- [x] FileConflict.isExternal flag, ProjectConflicts.externalConflictCount field
- [x] clearAllConflicts/clearProjectConflicts clear timestamp state

#### Health Store Integration
- [x] externalConflictCount added to ProjectHealth interface
- [x] Attention reason includes "(N external)" note when external conflicts present

#### UI Updates
- [x] ProjectBox $effect: starts/stops fs watcher per project CWD, listens for events, calls recordExternalWrite
- [x] ProjectHeader: split conflict badge into orange "ext write" badge + red "agent conflict" badge
- [x] Toast notification on new external write conflict

#### Verification
- [x] vitest: 202/202 tests pass (+8 new external write tests)
- [x] cargo test: 39/39 pass (+5 new fs_watcher tests)

### Session: 2026-03-11 — Files Tab: PDF Viewer + CSV Table View

#### PDF Viewer
- [x] Added pdfjs-dist@5.5.207 dependency (WebKit2GTK has no built-in PDF viewer)
- [x] Created PdfViewer.svelte — canvas-based multi-page renderer
- [x] Zoom controls (0.5x–3x, 25% steps), HiDPI-aware (devicePixelRatio scaling)
- [x] Reads PDF via convertFileSrc() → pdfjs (no new Rust commands needed)
- [x] Page shadow, themed toolbar, error handling

#### CSV Table View
- [x] Created CsvTable.svelte — RFC 4180 CSV parser (no external dependency)
- [x] Auto-detects delimiter (comma, semicolon, tab)
- [x] Sortable columns (numeric-aware), sticky header, row numbers
- [x] Row hover, text truncation at 20rem, themed via --ctp-* vars

#### FilesTab Routing
- [x] Binary+pdf → PdfViewer (via isPdfExt check)
- [x] Text+csv → CsvTable (via isCsvLang check)
- [x] Updated file icons: 📕 PDF, 📊 CSV
- [x] Both viewers are read-only

#### Verification
- [x] vitest: 202/202 tests pass (no regressions)
- [x] Vite build: clean
- [x] cargo check: clean

### Session: 2026-03-11 — S-2 Session Anchors

#### Implementation
- [x] Created types/anchors.ts — AnchorType, SessionAnchor, AnchorSettings, budget constants
- [x] Created adapters/anchors-bridge.ts — 5 Tauri IPC functions (save, load, delete, clear, updateType)
- [x] Created stores/anchors.svelte.ts — Svelte 5 rune store (per-project anchor management)
- [x] Created utils/anchor-serializer.ts — observation masking, turn grouping, token estimation
- [x] Created utils/anchor-serializer.test.ts — 17 tests (4 describe blocks)
- [x] Added session_anchors SQLite table + SessionAnchorRecord struct + 5 CRUD methods (session.rs)
- [x] Added 5 Tauri commands for anchor persistence (lib.rs)
- [x] Auto-anchor logic in agent-dispatcher.ts on first compaction event per project
- [x] Re-injection in AgentPane.startQuery() via system_prompt field
- [x] Pin button on AgentPane text messages
- [x] Anchor section in ContextTab: budget meter, promote/demote, remove

#### Verification
- [x] vitest: 219/219 tests pass (+17 new anchor tests)
- [x] cargo test: 42/42 pass (+3 new session_anchors tests)

### Session: 2026-03-11 — Configurable Anchor Budget + Truncation Fix

#### Research-backed truncation fix
- [x] Removed 500-char assistant text truncation in anchor-serializer.ts
- [x] Research consensus (JetBrains NeurIPS 2025, SWE-agent, OpenDev ACC): reasoning must never be truncated, only tool outputs get masked

#### Configurable anchor budget scale
- [x] Added AnchorBudgetScale type ('small'|'medium'|'large'|'full') with preset map (2K/6K/12K/20K)
- [x] Added anchorBudgetScale? field to ProjectConfig (persisted in groups.json)
- [x] Updated getAnchorSettings() to resolve budget from scale
- [x] Added 4-stop range slider to SettingsTab per-project settings
- [x] Updated ContextTab to derive budget from anchorBudgetScale prop
- [x] Updated agent-dispatcher to look up project's budget scale

#### Cleanup
- [x] Removed Ollama-specific warning toast from AgentPane (budget slider handles generically)
- [x] Removed unused notify import from AgentPane

#### Verification
- [x] vitest: 219/219 tests pass (no regressions)
- [x] cargo test: 42/42 pass (no regressions)

### Session: 2026-03-11 — S-1 Phase 3: Worktree Isolation Per Project

#### UI toggle
- [x] Added 'Worktree Isolation' checkbox to SettingsTab per-project card (card-field-row CSS layout)
- [x] ProjectConfig.useWorktrees? already existed — wired to toggle

#### Spawn with worktree flag
- [x] Added worktree_name: Option<String> to AgentQueryOptions (Rust sidecar.rs)
- [x] Added worktree_name?: string to TS AgentQueryOptions (agent-bridge.ts)
- [x] Sidecar JSON passes worktreeName field to claude-runner.ts
- [x] claude-runner.ts passes extraArgs: { worktree: name } to SDK query() (maps to --worktree CLI flag)
- [x] AgentPane: added useWorktrees prop, passes worktree_name=sessionId when enabled
- [x] AgentSession: passes useWorktrees={project.useWorktrees} to AgentPane
- [x] Rebuilt sidecar bundle (claude-runner.mjs)

#### CWD-based worktree detection
- [x] Added detectWorktreeFromCwd() to agent-dispatcher.ts (matches .claude/.codex/.cursor worktree patterns)
- [x] Init event handler now calls setSessionWorktree() when CWD contains worktree path
- [x] Dual detection: CWD-based (primary) + tool_call-based extractWorktreePath (subagent fallback)

#### Tests
- [x] Added 7 new tests to agent-dispatcher.test.ts (detectWorktreeFromCwd unit tests + init CWD integration)
- [x] vitest: 226/226 tests pass
- [x] cargo test: 42/42 pass

### Session: 2026-03-11 — Provider Runners (Codex + Ollama)

#### Codex Provider
- [x] providers/codex.ts — ProviderMeta (gpt-5.4, hasSandbox, supportsResume)
- [x] adapters/codex-messages.ts — adaptCodexMessage (ThreadEvents → AgentMessage[])
- [x] sidecar/codex-runner.ts — @openai/codex-sdk wrapper (dynamic import, graceful failure)
- [x] adapters/codex-messages.test.ts — 19 tests

#### Ollama Provider
- [x] providers/ollama.ts — ProviderMeta (qwen3:8b, modelSelection only)
- [x] adapters/ollama-messages.ts — adaptOllamaMessage (streaming chunks → AgentMessage[])
- [x] sidecar/ollama-runner.ts — Direct HTTP to localhost:11434 (zero deps)
- [x] adapters/ollama-messages.test.ts — 11 tests

#### Registration + Build
- [x] App.svelte: register CODEX_PROVIDER + OLLAMA_PROVIDER
- [x] message-adapters.ts: register codex + ollama adapters
- [x] package.json: build:sidecar builds all 3 runners
- [x] vitest: 256/256 tests pass
- [x] cargo test: 42/42 pass

### 2026-03-11 — Register Memora Adapter

**Duration:** ~15 min

**What happened:**
Registered a concrete MemoraAdapter that bridges the MemoryAdapter interface to the Memora SQLite database. Direct read-only SQLite access (no MCP/CLI dependency at runtime).

#### Rust Backend
- [x] memora.rs — MemoraDb struct (read-only SQLite, Option<Connection>, graceful absence)
- [x] list() with tag filtering via json_each() + IN clause
- [x] search() via FTS5 MATCH on memories_fts, optional tag join
- [x] get() by ID
- [x] 4 Tauri commands: memora_available, memora_list, memora_search, memora_get
- [x] 7 cargo tests (missing-db error paths)

#### TypeScript Bridge + Adapter
- [x] memora-bridge.ts — IPC wrappers + MemoraAdapter class implementing MemoryAdapter
- [x] App.svelte — registers MemoraAdapter on mount with async availability check
- [x] memora-bridge.test.ts — 16 tests (IPC + adapter)

#### Results
- [x] vitest: 272/272 tests pass
- [x] cargo test: 49/49 pass

### 2026-03-11 — Configurable Stall Threshold

**Duration:** ~10 min

**What happened:**
Made the hardcoded 15-minute stall threshold configurable per-project via a range slider in SettingsTab (5–60 min, step 5).

#### Changes
- [x] groups.ts — Added `stallThresholdMin?: number` to ProjectConfig
- [x] health.svelte.ts — Replaced hardcoded constant with per-project `stallThresholds` Map + `setStallThreshold()` API, fallback to DEFAULT_STALL_THRESHOLD_MS (15 min)
- [x] SettingsTab.svelte — Range slider per project card (5–60 min, step 5, default 15)
- [x] ProjectBox.svelte — `$effect` syncs `project.stallThresholdMin` → `setStallThreshold()` on mount/change

#### Results
- [x] No test changes — UI/config wiring only
- [x] vitest: 272/272 tests pass
- [x] cargo test: 49/49 pass

### 2026-03-11 — Nemesis Security Audit + Reconnect Loop Fix

**Duration:** ~15 min

**What happened:**
Ran nemezis-audit on Rust backend. 0 verified exploitable findings, 10 recon targets identified (all previously known from 2026-03-08 security audit). Fixed Priority 8 reconnect loop race condition.

#### Nemesis Audit
- [x] Ran nemezis orchestrator on v2/src-tauri (Rust backend, 496s, $0.57)
- [x] 0 verified findings, 10 attack surface targets in recon hit list
- [x] All targets match previous 2026-03-08 security audit — no new vulnerabilities

#### Reconnect Loop Fix
- [x] remote.rs — Added `cancelled: Arc<AtomicBool>` to RemoteMachine struct
- [x] remove_machine() and disconnect() set cancelled=true before aborting tasks
- [x] connect() resets cancelled=false for new connections
- [x] Reconnect loop checks flag at top of each iteration, exits immediately when set

#### Results
- [x] cargo check: clean
- [x] cargo test: 49/49 pass

### Session 2026-03-11 (SOLID Phase 1 Refactoring)

#### SOLID Analysis
- [x] Ran /solid on full v2 codebase (TypeScript + Rust)
- [x] Identified 3 critical, 5 high, 5 medium issues; 8 good-practice modules

#### Phase 1 Refactoring — High-impact, Low-risk
- [x] **AttentionScorer extraction**: scoreAttention() pure function from health.svelte.ts → utils/attention-scorer.ts (14 tests)
- [x] **Shared type guards**: str()/num() from 3 adapter copies → utils/type-guards.ts
- [x] **lib.rs command module split**: 976 → 170 lines, 48 commands → 11 domain modules under commands/
- [x] Skipped withRemoteSupport() HOF — parameter shapes differ, 3-line duplication doesn't justify abstraction

#### Results
- [x] vitest: 286/286 pass (14 new attention-scorer tests)
- [x] cargo check: clean
- [x] cargo test: 49/49 pass

### Session 2026-03-11 (SOLID Phase 2 Refactoring)

#### agent-dispatcher.ts Split (496→260 lines)
- [x] Extracted utils/worktree-detection.ts — detectWorktreeFromCwd() pure function (17 lines, 5 tests)
- [x] Extracted utils/session-persistence.ts — session maps + persistSessionForProject (107 lines)
- [x] Extracted utils/auto-anchoring.ts — triggerAutoAnchor (48 lines)
- [x] Extracted utils/subagent-router.ts — spawnSubagentPane + SUBAGENT_TOOL_NAMES (73 lines)
- [x] Dispatcher is now thin coordinator with re-exports for backward compat

#### session.rs Split (1,008 lines → 7 sub-modules)
- [x] session/mod.rs — SessionDb struct + open() + migrate() + re-exports (153 lines)
- [x] session/sessions.rs — Session CRUD (9 tests)
- [x] session/layout.rs — LayoutState save/load (3 tests)
- [x] session/settings.rs — Settings CRUD (5 tests)
- [x] session/ssh.rs — SshSession CRUD (4 tests)
- [x] session/agents.rs — AgentMessageRecord + ProjectAgentState
- [x] session/metrics.rs — SessionMetric save/load
- [x] session/anchors.rs — SessionAnchorRecord CRUD
- [x] conn field: pub(in crate::session) for sub-module access

#### Results
- [x] vitest: 286/286 pass (5 worktree tests moved to new file)
- [x] cargo check: clean
- [x] cargo test: 49/49 pass

### Session 2026-03-11 (SOLID Phase 3 — Branded Types)

#### Implementation
- [x] Introduced SessionId/ProjectId branded types (types/ids.ts)
- [x] Applied branded types to conflicts.svelte.ts and health.svelte.ts Map keys
- [x] Branded sessionId at sidecar boundary in agent-dispatcher
- [x] Applied branded types at Svelte component call sites

#### Results
- [x] cargo check: clean
- [x] vitest: 286/286 pass

### Session 2026-03-11 — Multi-Agent Orchestration System

#### btmsg Group Agent Messenger CLI
- [x] Created btmsg CLI tool — inter-agent messaging (inbox, send, reply, contacts, history, channels)
- [x] btmsg graph command — visual agent hierarchy with status
- [x] Admin role (tier 0), channel messaging (create/list/send/history), mark-read, global feed

#### btmsg Rust Backend + Tauri Bridge
- [x] Created btmsg.rs module — SQLite-backed messaging (shared DB: ~/.local/share/bterminal/btmsg.db)
- [x] 8+ Tauri commands: btmsg_inbox, btmsg_send, btmsg_read, btmsg_contacts, btmsg_feed, btmsg_channels, etc.
- [x] CommsTab: sidebar chat interface with activity feed, DMs, channels (Ctrl+M)

#### Agent Unification (Tier 1 → ProjectBoxes)
- [x] agentToProject() converter in groups.ts — Tier 1 agents rendered as full ProjectBoxes
- [x] getAllWorkItems() in workspace store combines agents + projects for ProjectGrid
- [x] GroupAgentsPanel: click-to-navigate agent cards to their ProjectBox

#### Agent System Prompts
- [x] Created utils/agent-prompts.ts — generateAgentPrompt() builds comprehensive introductory context
  - Sections: Identity, Environment, Team hierarchy, btmsg docs, bttask docs, Custom context, Workflow
  - Role-specific workflows (Manager: check inbox → review board → coordinate; Architect: code review focus; Tester: write/run tests)
- [x] AgentSession builds system prompt: Tier 1 gets full generated prompt, Tier 2 gets custom context

#### BTMSG_AGENT_ID Environment Passthrough (5-layer chain)
- [x] agent-bridge.ts: added extra_env?: Record<string,string> to AgentQueryOptions
- [x] bterminal-core/sidecar.rs: added extra_env: HashMap<String,String> with #[serde(default)]
- [x] claude-runner.ts: extraEnv merged into cleanEnv after provider var stripping
- [x] codex-runner.ts: same extraEnv pattern
- [x] AgentSession injects { BTMSG_AGENT_ID: project.id } for agent projects

#### Tier 1 Agent Config in SettingsTab
- [x] Agent cards: icon + name + role badge + enable toggle + CWD + model + wake interval (manager)
- [x] Custom Context textarea per agent (appended to auto-generated prompt)
- [x] Collapsible preview of full generated introductory prompt
- [x] updateAgent() function in workspace store for Tier 1 config persistence

#### Custom Context for Tier 2 Projects
- [x] Custom Context textarea in SettingsTab project cards
- [x] Stored as project.systemPrompt, passed through AgentSession → AgentPane

#### Periodic System Prompt Re-injection
- [x] AgentSession: 1-hour timer (REINJECTION_INTERVAL_MS = 3,600,000ms)
- [x] Checks every 60s if elapsed > 1 hour, sets contextRefreshPrompt
- [x] AgentPane: autoPrompt prop consumed only when agent is idle (done/error state)
- [x] Different refresh messages: Tier 1 (check inbox + task board), Tier 2 (review instructions)

#### bttask Kanban Backend + UI
- [x] Created bttask.rs — Task/TaskComment structs, 6 operations (list, comments, update_status, create, delete, add_comment)
- [x] Created commands/bttask.rs — 6 Tauri commands registered in lib.rs
- [x] Created bttask-bridge.ts — TypeScript IPC adapter
- [x] Created TaskBoardTab.svelte — Kanban board with 5 columns (todo/progress/review/done/blocked)
  - Task creation form (title, description, priority)
  - Expandable task detail with status actions, comments, delete
  - 5-second polling
  - Pending count badge

#### ArchitectureTab (PlantUML Diagrams)
- [x] Created ArchitectureTab.svelte — PlantUML diagram viewer/editor
  - Sidebar with diagram list + new diagram form (4 templates: Class, Sequence, State, Component)
  - PlantUML source editor + SVG preview via plantuml.com server (~h hex encoding)
  - Stores .puml files in .architecture/ directory
  - Read/write via files-bridge.ts

#### TestingTab (Selenium + Automated Tests)
- [x] Created TestingTab.svelte — dual-mode component
  - Selenium mode: screenshot gallery (.selenium/screenshots/), session log viewer, 3s polling
  - Tests mode: discovers test files in standard dirs (tests/, test/, spec/, __tests__/, e2e/), file content viewer

#### Role-Specific Tabs in ProjectBox
- [x] Extended ProjectTab type: added 'tasks' | 'architecture' | 'selenium' | 'tests'
- [x] Conditional tab buttons: Manager→Tasks, Architect→Arch, Tester→Selenium+Tests
- [x] PERSISTED-LAZY rendering via {#if everActivated[tab]} pattern
- [x] .ptab-role CSS class (mauve accent color for agent-specific tabs)

#### Bug Fix
- [x] Fixed FileContent type case: 'text' → 'Text' in ArchitectureTab and TestingTab (files-bridge uses capital T)

#### Verification
- [x] cargo check: clean (bttask module + commands)
- [x] svelte-check: 0 project errors
- [x] Sidecar rebuilt with extraEnv support

### Night Integration Session — Bug Fixes for dexter_changes (2026-03-11)

Reviewed and integrated Dexter's multi-agent orchestration branch (dexter_changes, ~6,500 lines, 36 files). Fast-forward merged. Fixed 5 critical bugs + 2 hardening improvements.

#### Critical Bug Fixes
- [x] **Bug 1 — btmsg.rs column index mismatch**: `get_agents()` used `SELECT a.*` with positional index 7 for `status`, but column 7 is `system_prompt` (column 8 is `status`). Converted ALL query functions in btmsg.rs and bttask.rs from positional to named column access (`row.get("column_name")`). Added SQL aliases for JOIN columns.
- [x] **Bug 2 — btmsg-bridge.ts serde camelCase mismatch**: `BtmsgAgent` and `BtmsgMessage` TypeScript interfaces used snake_case (`group_id`, `unread_count`, `from_agent`) but Rust `#[serde(rename_all = "camelCase")]` sends camelCase. Fixed interfaces to match wire format.
- [x] **Bug 3 — GroupAgentsPanel event propagation**: toggleAgent button click propagated to parent card click handler (setActiveProject). Added `e.stopPropagation()`.
- [x] **Bug 4 — ArchitectureTab PlantUML encoding**: `rawDeflate()` was a no-op, `encode64()` did hex encoding. Collapsed into single `plantumlEncode()` using PlantUML's `~h` hex encoding.
- [x] **Bug 5 — TestingTab Tauri 2.x asset URL**: Used `asset://localhost/` (Tauri 1.x pattern). Fixed to `convertFileSrc()` from `@tauri-apps/api/core`.

#### Hardening
- [x] Added WAL mode + 5s busy_timeout to both btmsg.rs and bttask.rs `open_db()` for safe concurrent access from Python CLIs + Rust backend
- [x] Fixed misleading "Read-only access" comment in btmsg.rs (it opens READ_WRITE)

#### Verification
- [x] cargo check: clean
- [x] svelte-check: 0 project errors (2 pre-existing in node_modules/esrap)

### Regression Tests + Sidecar Security Session (2026-03-11 night, continued)

#### Regression Tests (49 new tests total)
- [x] **btmsg.rs cargo tests (8)**: In-memory SQLite with full schema, verifies named column access returns status (not system_prompt), unread counts, JOIN alias disambiguation (all_feed, unread_messages, channel_messages), serde camelCase serialization
- [x] **bttask.rs cargo tests (7)**: Named column access (list_tasks, task_comments), serde camelCase serialization, status validation
- [x] **sidecar strip_provider_env_var tests (8)**: All prefix combinations (CLAUDE, CODEX, OLLAMA, ANTHROPIC stripped; CLAUDE_CODE_EXPERIMENTAL whitelisted; OPENAI kept), integration test with full env simulation
- [x] **btmsg-bridge.test.ts (17)**: camelCase field verification for all 5 interfaces, IPC command names, error propagation
- [x] **bttask-bridge.test.ts (10)**: camelCase field verification (Task, TaskComment), IPC command names
- [x] **plantuml-encode.test.ts (7)**: ~h hex prefix, ASCII/Unicode encoding, URL safety, plantuml.com URL generation
- [x] Fixed pre-existing groups.rs test (missing `agents` field from Dexter's schema change)

#### Security: Sidecar Env Allowlist
- [x] Added ANTHROPIC_* to Rust-level strip_provider_env_var() — defense-in-depth (Claude CLI uses credentials file, not env var for auth)
- [x] OPENAI_* intentionally NOT stripped at Rust level (Codex runner needs OPENAI_API_KEY from env)
- [x] Documented dual-layer stripping design in code comments

#### Test Counts
- Vitest: 327 passed (was 286, +41)
- Cargo src-tauri: 64 passed (was 49, +15)
- Cargo bterminal-core: 8 passed (was 0, +8)

### E2E Testing Engine — Phase A (2026-03-12)

#### Test Mode Infrastructure
- [x] Rust: watcher.rs — test mode bypass (skip file watching, return content directly)
- [x] Rust: fs_watcher.rs — test mode bypass (skip inotify watchers)
- [x] Rust: commands/misc.rs — added `is_test_mode` Tauri command
- [x] Rust: lib.rs — registered is_test_mode in invoke handler
- [x] Frontend: wake-scheduler.svelte.ts — `disableWakeScheduler()` export + early return
- [x] Frontend: App.svelte — test mode detection via IPC, disables wake scheduler

#### E2E Test Anchors (data-testid attributes)
- [x] AgentPane: data-testid="agent-pane", data-agent-status, agent-messages, agent-stop, agent-prompt, agent-submit
- [x] ProjectBox: data-testid="project-box", data-project-id, project-tabs, terminal-toggle
- [x] StatusBar: data-testid="status-bar"
- [x] AgentSession: data-testid="agent-session"
- [x] GlobalTabBar: data-testid="sidebar-rail", settings-btn
- [x] CommandPalette: data-testid="command-palette", palette-input
- [x] TerminalTabs: data-testid="terminal-tabs", tab-add

#### WebDriverIO Config Improvements
- [x] TCP readiness probe replaces blind 2s sleep for tauri-driver startup
- [x] BTERMINAL_TEST=1 env var passed to Tauri app via capabilities
- [x] Optional BTERMINAL_TEST_DATA_DIR / BTERMINAL_TEST_CONFIG_DIR passthrough

#### Test Infrastructure Files
- [x] `v2/tests/e2e/fixtures.ts` — isolated test fixture generator (temp dirs, git repos, groups.json)
- [x] `v2/tests/e2e/results-db.ts` — JSON-based test results store (no native deps)
- [x] `v2/tests/e2e/specs/agent-scenarios.test.ts` — 7 Phase A scenarios (22 test cases)

#### Phase A Scenarios (7 scenarios, 22 tests)
1. **App Structural Integrity** — verifies all data-testid anchors render correctly
2. **Settings Panel (data-testid)** — open/close settings via stable selector
3. **Agent Pane Initial State** — idle status, prompt textarea, empty messages
4. **Terminal Tab Management** — add/close tabs via data-testid, empty state
5. **Command Palette (data-testid)** — open, focus, filter, close
6. **Project Focus & Tab Switching** — focus, tab persistence, agent status preservation
7. **Agent Prompt Submission** — textarea input, submit button state, graceful Claude CLI skip

#### Verification
- [x] cargo test: 68 passed, 0 failed
- [x] vitest: 345 passed across 18 files, 0 failed
- [x] svelte-check: 0 project errors (2 pre-existing esrap node_modules)

#### Test Counts
- Vitest: 345 passed (was 327, +18 — new wake-scorer + metrics tests from prior session)
- Cargo src-tauri: 68 passed (was 64, +4)
- E2E scenarios: 22 new test cases across 7 scenarios

### Session: 2026-03-12 — E2E Testing Engine Phase B+

#### LLM Judge Helper
- [x] Created `v2/tests/e2e/llm-judge.ts` — Claude API-based test assertion judge
  - Raw fetch to Anthropic API (zero new deps), uses claude-haiku-4-5 for speed/cost
  - `judge()` evaluates actual output against criteria, returns structured verdict (pass/fail, reasoning, confidence)
  - `assertWithJudge()` convenience with minimum confidence threshold (default 0.7)
  - `isJudgeAvailable()` check — tests skip gracefully when ANTHROPIC_API_KEY absent

#### Phase B Scenarios (6 scenarios, ~15 tests)
- [x] Created `v2/tests/e2e/specs/phase-b.test.ts`
  - **B1: Multi-Project Grid** — renders multiple project boxes, unique IDs, independent agent panes, CWD paths, focus/active styling
  - **B2: Independent Tab Switching** — different tabs active in different project boxes simultaneously
  - **B3: Status Bar Fleet State** — agent count display, burn rate $0.00 when all idle
  - **B4: LLM-Judged Agent Response** — sends file listing prompt, evaluates response quality + tool usage via LLM judge (requires ANTHROPIC_API_KEY)
  - **B5: LLM-Judged Code Generation** — sends code explanation prompt, evaluates correctness via LLM judge
  - **B6: Context Tab After Activity** — verifies context tab shows token usage data after agent activity
- [x] Per-project helper functions: focusProject(), getAgentStatus(), sendPromptInProject(), waitForProjectAgentStatus(), getAgentMessages(), switchProjectTab()
- [x] All LLM-judged tests skip gracefully when ANTHROPIC_API_KEY not set
- [x] Added phase-b.test.ts to wdio.conf.js specs array

#### CI Workflow
- [x] Created `.github/workflows/e2e.yml`
  - Triggers: push to v2-mission-control, PRs to master/v2-mission-control, manual dispatch
  - Path filters: v2/src/**, v2/src-tauri/**, v2/tests/e2e/**
  - 3 jobs: unit-tests (vitest), cargo-tests, e2e-tests (needs both)
  - E2E job: installs xvfb + tauri-driver, builds debug binary, runs Phase A + Phase B specs
  - LLM-judged tests gated on ANTHROPIC_API_KEY secret availability
  - Uploads test-results/ artifact on all outcomes

#### Verification
- [x] Vitest: 388 passed, 0 failed (was 345, +43 from prior sessions)
- [x] Cargo: 68 passed, 0 failed
- [x] No regressions

---

### Session: v3 Production Readiness — Tribunal Implementation (2026-03-12)

Implemented ALL 13 features from tribunal assessment in 3 parallel waves (11 sub-agents total). 60+ files changed, 3861 insertions.

#### Wave 1: Core Infrastructure (3 agents)
- [x] **Sidecar supervisor** — `bterminal-core/src/supervisor.rs`: SidecarSupervisor with exponential backoff (1s-30s, 5 retries), SidecarHealth enum, 5min stability window, 17 tests
- [x] **FTS5 search** — rusqlite `bundled-full`, SearchDb with 3 FTS5 virtual tables, SearchOverlay.svelte (Ctrl+Shift+F), search-bridge.ts
- [x] **Secrets management** — `keyring` crate (linux-native/libsecret), SecretsManager, secrets-bridge.ts, SettingsTab section

#### Wave 2: Features (5 agents)
- [x] **Notification system** — notify-rust + NotificationCenter.svelte (bell icon, history, 6 types), notifications-bridge.ts
- [x] **Keyboard-first UX** — Alt+1-5 jump, Ctrl+H/L vi-nav, Ctrl+Shift+1-9 tabs, Ctrl+J terminal, CommandPalette rewrite (18+ cmds, 6 categories)
- [x] **Agent health monitoring** — heartbeats + dead_letter_queue tables, 15s poll, ProjectHeader heart indicator, StatusBar badge
- [x] **Plugin system** — plugins.rs discovery, plugin-host.ts sandboxed runtime, plugins.svelte.ts store, example plugin
- [x] **Landlock sandbox** — bterminal-core/src/sandbox.rs, SandboxConfig, pre_exec() integration, per-project toggle

#### Wave 3: Integration (3 agents)
- [x] **Error classification** — error-classifier.ts (6 types, retry logic), 20 tests
- [x] **Audit log** — audit_log table, AuditLogTab.svelte, audit-bridge.ts
- [x] **Team agent orchestration** — install_cli_tools(), register_agents_from_groups(), bidirectional contacts, review channels
- [x] **Optimistic locking** — version column in bttask, conflict detection in Rust + Python CLI
- [x] **Usage meter** — UsageMeter.svelte, AgentPane integration

#### Key Fixes
- PRAGMA journal_mode=WAL: changed to query_row for bundled-full compatibility (session/mod.rs, btmsg.rs, bttask.rs)
- SidecarConfig: changed to `Mutex<SidecarConfig>` for interior mutability (sandbox updates)
- New Cargo deps: notify-rust 4, keyring 3 (linux-native), landlock 0.4

#### Verification
- [x] Vitest: 409 passed, 0 failed (+21 from prior)
- [x] Cargo: 109 passed, 0 failed (+41 from prior)
- [x] No regressions

---

### Session: v3 Hardening Sprint (2026-03-12)

Executed tribunal-recommended hybrid S-2/S-1 hardening sprint. Fixed 3 security/resilience issues, added TLS, fixed gitignore bug.

#### Subagent Delegation Fix
- [x] Root cause: Manager system prompt had no mention of Agent tool / delegation capability
- [x] Added "Multi-Agent Delegation" section to Manager workflow in `agent-prompts.ts`
- [x] Inject `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` env var for Manager agents in `AgentSession.svelte`

#### TLS for bterminal-relay
- [x] Added `--tls-cert` and `--tls-key` optional CLI args to relay binary
- [x] `build_tls_acceptor()` using `native_tls::Identity::from_pkcs8`
- [x] Refactored to generic `accept_ws_with_auth<S>` and `run_ws_session<S>` (avoids code duplication)
- [x] Client side already supports `wss://` via `connect_async` with native-tls feature — no changes needed
- [x] Certificate pinning deferred to v3.1 per tribunal risk matrix

#### Security Hardening
- [x] **WAL checkpoint** — `checkpoint_wal()` + `spawn_wal_checkpoint_task()` in lib.rs. Runs `PRAGMA wal_checkpoint(TRUNCATE)` every 5 minutes on sessions.db + btmsg.db. 2 tests
- [x] **Landlock logging** — Improved fallback message: "Kernel 6.2+ required for enforcement" + 3-state enforcement comments
- [x] **Plugin sandbox** — Already hardened (13 shadowed globals, `this` binding to undefined). Documented known `new Function()` escape vectors in JSDoc

#### Gitignore Fix
- [x] Root `.gitignore` had `plugins/` which matched `v2/src/lib/plugins/` (source code). Narrowed to `/plugins/` and `/v2/plugins/` (runtime dirs only)
- [x] Tracked previously-ignored `plugin-host.ts` source file
- [x] Added `plugin-host.test.ts` with 35 tests (sandbox globals, permissions, lifecycle)

#### Verification
- [x] Vitest: 444 passed, 0 failed (+35 plugin sandbox tests)
- [x] Cargo: 111 passed, 0 failed (+2 WAL checkpoint tests)
- [x] Full workspace compiles clean