fusion/docs/dashboard-guide.md

Dashboard Guide

← Docs index

The Fusion dashboard is the main control plane for tasks, agents, missions, settings, logs, and repository operations.

Browser Navigation

The dashboard now handles browser back navigation consistently on desktop and mobile. Using Back will first dismiss open modals and then step back through in-app view changes (for example, task detail → board) before leaving the app. This behavior used to be mobile-only, and now applies across all viewports. Task Detail modal opens from onboarding, activity log, and task-to-task navigation now all register navigation history entries, so Android back swipe/button dismisses them consistently.

Deep Links

Use deep links to open a specific task directly from notifications, chat, or external tools.

• /tasks/<TASK_ID> (for example, /tasks/FN-1234) opens that task, and can include ?project=<project-id> for multi-project routing.
• /?task=<TASK_ID>[&project=<project-id>] is the canonical in-app form and opens the task detail modal on load.
• Legacy path-style links (including trailing-slash forms like /tasks/<TASK_ID>/ and older hash-style entry points that resolve to that path) are normalized client-side to the canonical query form with history.replaceState, so the URL updates without a full reload.
• In non-headless dashboard mode, the server also issues an HTTP 301 redirect from /tasks/<TASK_ID> to /?task=<TASK_ID> and preserves ?project= when present.
• Theme assets resolve theme-data.css against the current document base (HTTP/HTTPS, file://, and Electron fallback paths), so non-default themes still load correctly when you land on deep-linked or sub-path URLs.
• Configure dashboardHost and ntfyDashboardHost in settings reference so generated notification links use the correct base URL.

/tasks/FN-1234
/?task=FN-1234
/?task=FN-1234&project=my-project

Clickable File Paths

File paths in dashboard text are automatically rendered as inline links. Clicking a linked path opens the Files browser modal at that path (including line/column targets when available) so you can inspect the file and use editor actions where supported.

Current surfaces include:

• Task detail modal content (description markdown, Review tab, and Workflow Results tab output)
• Chat view messages/tool output
• Agent log viewer
• Activity log modal
• Dev Server log viewer
• Settings sync log

Only detected file-path text is linkified; non-path text remains plain. Linked paths must resolve within the current project workspace to open successfully.

Board View

Board view is the kanban surface for day-to-day operation.

Features:

• Drag-and-drop between lifecycle columns
• Search/filter tasks (including working-branch and base-branch dropdown filters with explicit No working branch / No base branch options)
• Working-branch and base-branch filter selections are persisted per project and restored across refresh/navigation
• Column visibility controls
• Inline quick entry creation
• PR/issue badges with live updates
• GitHub provenance marker on task cards imported from GitHub (sourceType: github_import), shown alongside existing footer metadata like timers
• Agent-created provenance badge in task card headers for agent-originated tasks (sourceType: agent_heartbeat or sourceType: automation, or legacy tasks with sourceAgentId), with labels preferring sourceMetadata.agentName over raw agent IDs
• Column ordering semantics: todo mirrors scheduler pickup order (priority descending, then oldest createdAt, then task ID); triage, in-progress, in-review, and archived remain priority-first with task-ID tie-breaks; done is ordered by most recent completion first (columnMovedAt, then updatedAt, then createdAt fallback)

List View

List view is optimized for dense task management.

Features:

• Sectioned task table grouped by lifecycle column
• Sortable columns (ID/title/status/column)
• Column visibility toggles and optional hide-done filtering
• Bulk selection + batch model updates
• Bulk Pause / Unpause / Archive actions from the selection toolbar (Pause selected, Unpause selected, Archive selected) for fast batch task state management.
• Bulk delete from the selection toolbar (Delete selected): archived selections are skipped automatically, and dependency-conflict failures can be force-deleted per task after a danger confirmation that removes dependency references.

Graph View

Graph view visualizes task dependencies as an interactive node/edge map.

Navigation:

• Desktop: Header → More views → Graph
• Mobile: MobileNavBar → More → Graph

Behavior:

• Shows only tasks in triage, todo, in-progress, and in-review
• Excludes done and archived
• Uses Sugiyama-style layered auto-layout to place nodes by dependency depth
• Renders directed bezier dependency edges (dependent → dependency) with arrowheads
• Supports cursor-centered wheel zoom, pinch zoom, keyboard shortcuts (Ctrl/Cmd+=, Ctrl/Cmd+-, Ctrl/Cmd+0, Ctrl/Cmd+Shift+F, Escape), and fit/reset controls via the floating toolbar with live zoom percentage
• Pan limits are zoom-aware and based on full graph extents (including negative auto-layout origins), so zoomed-in views can still pan to every rendered node instead of getting trapped by fixed viewport-only bounds
• Dependency graph nodes reuse the same TaskCard UI as board/list views, so status badges, progress/steps, mission badges, retry/archive controls, and active-task glow stay visually consistent
• Active graph nodes also add a dedicated top status indicator bar and current-step row highlighting so in-progress execution state stays visible even when zoomed out
• Clicking a graph card opens task details via the host detail handler (onOpenDetail, with onOpenTaskDetail fallback), while clicking the same card again or empty canvas clears selection
• On touch devices, single-tap is reserved for pan/drag gestures, so double-tapping a node opens its task detail modal; this does not change selection state.
• Hovering or selecting a node highlights its full upstream and downstream dependency chain; highlighted nodes and connecting edges are emphasized while non-chain nodes are dimmed, and highlight clears when hover/selection is removed
• Nodes support manual drag repositioning with a 4px movement threshold to separate click from drag, using pointer capture and zoom-aware delta scaling for reliable tracking
• Custom node positions persist per project in browser localStorage (kb:${projectId}:fusion-plugin-dependency-graph:positions) across refresh/project switches, and Fit to graph clears saved positions and restores auto-layout

Planning Mode

Planning Mode now includes branch controls on the summary screen before you create a task.

• Branch strategy options mirror Subtask Breakdown semantics:
  • Use project/default branch
  • Create auto-named branch per task
  • Use existing branch
  • Create custom new branch
• Branch name is required when using existing or custom new strategies.
• Merge target / base branch (optional) lets you set the PR base branch (for example main or develop).

These values are sent with the Planning Mode create-task request as branchSelection, so created tasks persist branch/base-branch settings consistently with other branch-aware task creation flows.

New Task Modal Branch Strategy

The New Task dialog uses the same four-option Branch strategy selector and branchSelection payload as Planning Mode:

• Use project/default branch
• Create auto-named branch per task
• Use existing branch
• Create custom new branch

Rules:

• existing and custom-new require a branch name.
• project-default leaves branch unset.
• auto-new creates a branch after task creation using fusion/{task-id}-{short-name} (for example fusion/fn-5671-branch-strategy-dropdown).
• Merge target / base branch stays optional for all modes.

Chat View

Chat view provides project-scoped conversations with agents.

• Entering /new or /clear (exact match after trimming) in the composer starts a fresh thread for the current chat target instead of sending the literal command to the model
• On mobile, the New Chat and Delete Conversation dialogs use a compact inset treatment (centered, viewport-bounded, internally scrollable) instead of the app's default full-height mobile modal chrome.
• Full Chat and Quick Chat both consume the same streamed /api/chat/sessions/:id/messages response contract, and both now prefer the authoritative assistant message snapshot on done while still accumulating text chunks when present (so providers without incremental text streaming still render output immediately)
• In-progress assistant responses now survive refresh/navigation while generation is still active: Chat restores the last durable in-flight text/thinking/tool state immediately, then resumes streaming from the stored replay point instead of starting from an empty "Connecting…" placeholder.
• If a regular Chat stream drops with a hidden-tab/browser-suspension error (for example Load failed) while the server is still generating, Chat suppresses the false error banner, re-attaches to the in-progress stream using the durable replay state, and reconciles the final assistant reply when generation completes.
• Chat message lists now track near-bottom scroll state: while you are reading older messages, live streaming/new replies do not force-scroll; a Latest jump control appears until you return to the tail.
• On mobile direct-chat threads, entering a thread and restoring Chat after tab/page visibility returns re-anchors to the newest message (scrollTop = scrollHeight) so the view always opens at the live tail.
• On mobile direct-chat threads, tapping the active title/identity in the thread header opens a lightweight conversation dropdown so you can switch to another direct session without backing out to the sidebar list first; long conversation titles now stay readable in the dropdown via wrapped option text and taller touch-friendly rows.
• On mobile (max-width: 768px), chat bubbles are slightly wider in full Chat for improved readability while preserving header/composer gutters.
• Full Chat tool-call summaries now use a denser mobile layout: grouped and single-call collapsed rows keep icon + label + status on one line (Quick Chat-style scanability) while expanded details remain unchanged.
• The desktop Chat view toggle and mobile Chat tab now show an unread-response indicator when a live assistant reply arrives for your active chat thread after you leave Chat; opening Chat clears it immediately.
• Agent-backed chat sessions now expose the same mailbox messaging tools (fn_send_message, fn_read_messages) used by runtime execution/heartbeat flows whenever the engine MessageStore is available; model-only chats continue to run without mailbox tools.

Chat Rooms

Chat Rooms are project-scoped group conversations for multiple agents. They are separate from one-on-one direct chat sessions.

• Chat Rooms are currently gated behind the chatRooms experimental feature flag. Enable it in Settings → Experimental Features → Chat Rooms.
• Use the Direct / Rooms toggle in the Chat sidebar to switch scopes. The selected scope is saved and restored the next time you open Chat.
• In Rooms, click Create room to open the room-creation modal.
• Room names follow strict validation: a leading # is removed automatically, names must be lowercase, up to 80 characters, use only a-z, 0-9, -, or _, cannot start or end with -/_, and must be unique in the current project.
• The modal includes a member picker with search + multi-select from project agents. You must pick at least one member before creating the room.
• Members are currently chosen during room creation. The shipped UI does not yet provide full post-creation member management in Chat View.
• Each room row includes a trash action (aria-label="Delete room {name}", data-testid="chat-room-delete-{slug}") that opens a Delete Room? confirmation dialog with Cancel and Delete actions.
• Confirming delete calls rooms.deleteRoom(roomId) and permanently removes the room and its messages ("This action cannot be undone. This room and all its messages will be permanently deleted."); failures surface a Failed to delete room toast.
• Selecting a room opens the room thread pane with loading and empty states, then renders room messages from rooms.messages as ChatMessageInfo entries in the same thread UI used for direct Chat.
• Submitting the room composer calls rooms.sendRoomMessage(...), which immediately inserts a temporary local user message and then posts to POST /api/chat/rooms/:id/messages.
• The room composer clears immediately when send is dispatched so the user gets instant feedback; on success the optimistic message is reconciled with persisted server data and the transcript is refreshed to authoritative history.
• On mobile, room threads use the same keyboard-aware thread anchoring as direct chat, keeping the composer pinned above the soft keyboard while typing.
• The dashboard backend now orchestrates room responders on that POST: mentioned members are routed as direct responders, additional ambient members may reply (up to the room ambient responder cap), and each assistant reply is persisted with senderAgentId via chatStore.addRoomMessage(...).
• If room replies cannot be generated (for example no resolvable responders or all responders fail), the POST fails with an API error (HTTP 502) instead of silently returning only the user message.
• If room responders cannot be resolved or all room-reply generations fail, the POST now returns an error instead of silently succeeding with only the user message, so failures are surfaced deterministically.
• Room responder prompt construction now keeps the most recent room messages verbatim and, when the room runs long, prepends a compacted summary of older history (span, participants, and key highlights) plus an explicit latest-user-message marker so replies stay thread-aware without unbounded prompt growth.
• On send failure, useChatRooms rolls back/reconciles optimistic state and rethrows; ChatView catches once, restores the exact pre-send composer text for retry/edit, and surfaces a single error toast (no duplicate hook+view notifications).
• After each send attempt, the room transcript still re-fetches authoritative messages so persisted user/assistant replies remain visible even when SSE delivery is delayed, and chat:room:message:* SSE updates continue live fan-out.
• Relationship summary: direct Chat runs one target (agent or model) per session; rooms are shared threads with multiple agent members and now use the same message contract as direct Chat; Quick Chat is still a floating panel, but when a room is selected it now reads/writes that room thread directly.
• For backend details, see the Chat Room REST API reference and the chat room storage schema (chat_rooms, chat_room_members, chat_room_messages).

Quick Chat

Quick Chat is an optional floating panel for fast, project-scoped assistant conversations without leaving your current view.

• Controlled by the project setting showQuickChatFAB
• Supports agent mentions (@agent) and shared # task/file mentions
• Uses the same model/provider infrastructure as full Chat view
• On small screens, compact tool-call summaries in the floating panel intentionally stay single-line (count + tool names + status) to preserve message density
• The panel header uses a session-first flow: the main dropdown lists persisted sessions (preferring session.title, then falling back to deterministic Session N labels)
• Selecting a session from that dropdown resumes the persisted conversation; this keeps switchSession() resume-oriented rather than forcing a new thread
• Entering /new or /clear (exact match after trimming) in the Quick Chat composer clears the active thread target: direct/model targets use startFreshSession(...), while room targets call rooms.clearRoom(activeRoom.id).
• The + action opens an inline new-session chooser (inside the panel, not a modal) with Model selected by default and optional switch to Agent
• Submitting the inline chooser uses explicit fresh-session creation and immediately persists/selects the new thread, then refreshes the session dropdown list
• On every open, Quick Chat restores the most recently used non-archived session by latest activity (max(lastMessageAt, updatedAt)); only when no prior session exists does it fall back to the first agent / configured default model.
• Resume lookups still use targeted session queries instead of loading the full active-session list first
• Tool-call summaries in the floating quick-chat panel are intentionally condensed into a single-line header row (especially on small screens) so tool name + status stay scannable without multi-line wrapping
• On mobile viewports, opening Quick Chat auto-focuses the composer as soon as it is ready so the keyboard opens immediately
• FAB dragging uses pointer events with document-level move/up tracking and a 5px drag threshold so Android touch drags reposition reliably while short taps still open Quick Chat
• Quick Chat now mirrors full Chat tail behavior: if you scroll up, live updates stop auto-following and a Latest jump control appears until you jump back down.
• On mobile, Quick Chat re-anchors to the newest message whenever the panel is opened/reopened and when page visibility is restored, while still preserving the near-bottom gate so intentional scroll-away keeps Latest jump behavior.
• On mobile, Quick Chat bubbles are slightly wider while keeping compact tool-call summary layout and full-screen/safe-area behavior intact.

Mailbox View

Mailbox view shows inbox/outbox communication threads and unread state.

• Inbox renders one row per message (no sender-based collapsing)
• clicking a message in the Mail tab opens the task detail pane with full message content and conversation context
• reply rows in the mailbox modal can expand inline to show the replied-to message context for easier thread reading
• mailbox now includes an Approvals tab with pending and history filters (approved / denied / completed), approval detail context, and inline approve/deny actions for pending requests
• in the Agents tab, the agent selector now includes All agents, which shows one combined agent-to-agent stream (with sender + recipient labels); selecting a specific agent still shows Inbox/Outbox subtabs
• mailbox entry points now show pending-approval indicators: Header mailbox toggle dot, Header overflow mailbox badge, Mobile mailbox tab dot, and Mobile More → Mailbox badge
• approval lifecycle SSE events (approval:requested, approval:updated, approval:decided) trigger mailbox approvals refresh without manual reload
• when a task newly enters awaiting-approval, the app shows a persistent approval banner above project content with an Open Mailbox CTA; dismissals are remembered per approval item until that item advances or a different one arrives
• Visible message history/threading is driven by explicit message.metadata.replyTo.messageId links
• Separate top-level messages from the same sender remain independent in the inbox and detail pane

Interactive Terminal

Fusion embeds a terminal using xterm.js.

Features:

• Multiple terminal tabs
• PTY-backed shell sessions
• Mobile-aware virtual keyboard handling and auto-refit behavior
• Reopen/reconnect/session-recovery flows preserve single-keystroke input forwarding (no duplicate characters, no page refresh required)

Git Manager

Git manager centralizes repo operations in the dashboard.

Features:

• Branch/worktree visibility
• Commit and diff browsing
• Push/pull/fetch actions
• Pull with rebase option (split-button chooses between git pull and git pull --rebase)
• Remote editing controls
• Stash inspection (view stat + patch) before apply/pop/drop actions
• Remotes tab keeps "Recent commits on {remote}" in sync immediately after successful push/pull actions

Merge Advance Notice

Merge Advance Notice is a global banner (MergeAdvanceNotice) mounted in the main app chrome that appears when the integration branch advances.

When it appears:

• Reacts to task:merged SSE events
• Hydrates from GET /api/tasks/merge-advance-events
• Shows the latest merge-advance event for the current project

What it shows:

• Integration branch name and the new tip short SHA
• Advancing task ID and advance metadata from the event payload (advanceMode, refName, SHA details)
• Checkout-state warnings when your current worktree is dirty or has untracked files

How to react:

• Click Pull to run Smart Pull (POST /api/git/smart-pull), including the stash-conflict flow in StashConflictModal when needed
• Use the dismiss close button to hide the notice
• Treat dirty/untracked warnings as a hint that local changes may be auto-stashed during pull

Push follow-up (when shown):

• If the integration branch is ahead of origin, the banner can show push controls with ahead count
• Use Push to origin (or force-with-lease via Advanced) to publish the advanced branch tip
• If push is rejected (rejected-non-ff / sha-mismatch), the banner offers a Smart Pull retry path

Branch names are dynamic from merge/audit payloads; the banner is not hardcoded to main.

OAuth Re-login Banner

The global OAuth re-login banner now clears a provider row immediately after that provider successfully re-authenticates (from Settings → Authentication or Model Onboarding), instead of waiting for the next GET /auth/status poll interval.

Smart Pull

Smart Pull is a one-shot pull workflow that keeps local work safe while advancing your checked-out integration branch.

What it does:

• Calls POST /api/git/smart-pull
• If your worktree is clean, runs a fast-forward pull and returns kind: "clean-pull"
• If local changes exist, auto-stashes (including untracked files), runs git pull --ff-only, then restores the stash
• Returns kind: "stash-pull-pop" when stash → pull → pop succeeds cleanly
• Returns kind: "stash-pop-conflict" when stash restore conflicts, then opens StashConflictModal

Where it is triggered:

• From the merge-advance banner pull action (MergeAdvanceNotice)
• From any dashboard surface that invokes POST /api/git/smart-pull

When stash-pop-conflict occurs, StashConflictModal shows:

• Stash short SHA + stash label
• Per-file conflict list
• Per-file resolution actions (Keep mine / Keep incoming, backed by /api/git/stash-resolve choices ours/theirs)
• Stash actions: Drop stash (POST /api/git/stash-drop) and Restore from stash ref (POST /api/git/stash-restore)
• A stash-SHA copy button for sharing the conflict list/reference

After resolution:

• As each file is resolved, remainingConflicts shrinks; when empty, the modal can be closed and the branch stays at the advanced integration tip with resolved stash content applied
• Dropping the stash discards the saved local edits after conflicts are resolved
• Restoring from stash ref re-applies the stash and may reintroduce conflicts for manual handling

You may also see matching run-audit events in logs, including pull:fast-forward and stash:pop-conflict. goal:* run-audit events (goal:injection-applied, goal:injection-skipped, goal:retrieval-invoked) use the same timeline endpoint and are filterable with startTime/endTime query params.

Documents View

Documents view aggregates task documents and project markdown files.

Features:

• Group task documents by task ID (with revision history metadata)
• Search documents across tasks
• Open project markdown files with inline preview
• Jump directly from a document group to the owning task detail modal
• Toggle between raw text and rendered markdown using the Markdown/Plain button

Reports View

Reports View is available when the Reports plugin is installed and enabled.

Navigation:

• Desktop: Header → More views → Reports
• Mobile: More sheet → Reports

Features:

• Reports history list with filters for cadence, status, date range, title text, and agent
• Detail viewer with a sandboxed iframe preview backed by the report HTML preview endpoint
• Section quick-jump sidebar based on stable report section markers
• Compare drawer for side-by-side report comparisons with section-level diff groupings
• Standalone HTML download/export action for sharing a self-contained report file

For plugin internals (registration, API routes, rendering/export pipeline), see Reports plugin docs.

Markdown Rendering

Documents view supports toggling between raw text and formatted markdown when viewing document content:

• Raw mode (default): Shows markdown syntax as plain text (e.g., **bold**)
• Markdown mode: Renders markdown with proper formatting (e.g., bold, headings, lists, tables)

The toggle button is accessible with aria-pressed for screen readers. Toggle state is scoped per-document, so switching between documents resets the view to raw mode.

Todo View

Todo View is an experimental dashboard surface for managing per-project todo lists and turning items into planning or task workflows.

Available when experimentalFeatures.todoView is enabled.

Navigation:

• Desktop: Header → More views → Todos (single canonical desktop entry)
• Mobile: More sheet → Todos

For full behavior, API contracts, and storage details, use the canonical Todo View guide.

Research View

Research view is a standalone dashboard surface for creating and managing research runs.

Available when experimentalFeatures.researchView is enabled. The related Settings sections (Research Defaults and project Research) are also hidden until this flag is enabled.

Features:

• Create-run form with required query text and selectable provider options
• Searchable run history list with project-scoped state
• Selected-run reader with summary, citations, findings, and run event history
• Run lifecycle controls: cancel, retry, and refresh
• Export actions for supported formats (markdown, json, html as advertised by backend availability)
• Task-facing actions to create a new task from findings or attach findings to an existing task
• Graceful unavailable/setup messaging when research backend capability is disabled or not configured

Navigation:

• Desktop: Header → More views overflow menu
• Mobile: More sheet in MobileNavBar
• Research is intentionally not shown in the primary board/list/agents/missions/chat toggle row

For the full research workflow, provider setup, CLI commands, API reference, and agent integration, see the canonical Research guide.

Files Modal

The Files modal provides a workspace-aware file browser and editor.

• Source/text editing supports a Line # header toggle to show or hide line numbers in the editor gutter
• The line-number preference is saved per project and restored automatically when you switch projects

Memory View

Memory view provides a multi-file editor for project and daily memory files.

Available when the experimentalFeatures.memoryView toggle is enabled.

Agents View

Agent list and detail surfaces now surface pending approvals per agent:

• Agents list/board cards show a warning-colored pending-approval badge when pendingApprovalCount > 0
• Agent detail summary shows a matching pending-approval badge for the selected agent
• Approval SSE events refresh these indicators live (no page reload required)

Agents view is the control surface for runtime agents and team structure.

Navigation:

• Desktop: primary view toggle (Agents)
• Mobile: bottom nav tab (Agents)

Features:

• Switch between List, Board, and Org chart layouts
• Filter by role/state, include/exclude system agents, and inspect health/status
• Start, pause, stop, and trigger agent runs from the view and from detail panels
• In Agent detail, use the kebab Bulk agent actions button in the header utility cluster (next to Refresh and Close) to run project-wide lifecycle transitions for non-ephemeral agents in the current project — Pause All Agents targets agents in the active or running state, while Resume All Agents targets agents in the paused state only
• Bulk menu items stay disabled when nothing is eligible and show an inline hint (Loading eligible agents..., No active agents eligible, No paused agents eligible, or the current eligible count such as Pause 2 active/running agents)
• Bulk lifecycle flow: open Bulk agent actions, review the eligibility hint, confirm the modal, then use the success or partial-failure toast to verify paused/resumed counts plus skipped/failed agents
• Open agent detail tabs for runs, logs, read-only mail (agent inbox/outbox), settings/config, tasks, memory, and chain-of-command relationships
• Error indicator on agent list cards when an agent is in the error state and has a captured error (lastError); select it to open Agent Error Details
• Run-level error indicator in Agent detail → Runs when a run has captured stderr; select it to open the same Agent Error Details modal
• Agent Error Details shows full error text plus Copy and Report on GitHub actions
• Report on GitHub opens a pre-filled issue draft with available context from where you launched it (surface plus agent metadata, and run/task IDs when available on that view)
• Jump from agent activity to related task logs, and (when experimentalFeatures.agentOnboarding is enabled) launch AI Interview from the New Agent dialog (create mode) or Agent detail → Settings (edit mode)

For full lifecycle behavior, runtime/heartbeat settings, and budgets, see Agents guide.

Roadmaps View

Roadmaps view manages roadmap hierarchies (roadmaps, milestones, features) and planning handoff exports.

Available when experimentalFeatures.roadmap is enabled. Hidden when a plugin replaces Roadmaps navigation.

Navigation:

• Desktop: Header → More views → Roadmaps
• Mobile: More sheet (or promoted to a top tab when eligible based on mobile nav slot rules)

Features:

• Create, edit, archive/delete, and reorder roadmaps, milestones, and features
• Use inline editing plus drag/drop for milestone and feature organization
• Open roadmap export modal and copy mission/feature planning handoff payloads
• Feed roadmap output into mission/task planning workflows

For mission planning context and handoff structure, see Missions guide.

Goals View

Goals view is a strategic-goals surface backed by the Goals REST API.

No feature flag required. Current status: the GoalsView chunk is lazy-defined/prefetched in App.tsx, but it is not yet wired into the primary dashboard navigation.

What it shows:

• Header with active-goal count (N active goals) and an Add Goal action
• Goal cards with title, optional description, and Status: active|archived
• Empty state when no goals exist: No goals yet. Add one to begin tracking strategic outcomes.

Data behavior:

• Initial load: GET /api/goals (returns { goals })
• Create: inline Add Goal form posts title (required) + description (optional) to POST /api/goals
• Edit: per-card inline form patches title/description via PATCH /api/goals/:id
• Archive/unarchive: POST /api/goals/:id/archive and POST /api/goals/:id/unarchive

Active-goal cap behavior:

• Hard cap of 5 active goals (server-enforced)
• Warning banner appears when active goals are in the 3–5 range
• Cap violations (for create or unarchive) return HTTP 409 with code: ACTIVE_GOAL_LIMIT_EXCEEDED and are surfaced as inline goal errors

Source file: packages/dashboard/app/components/GoalsView.tsx

Evals View

Evals view is a dedicated dashboard surface for reviewing scheduled task-evaluation output.

Available when experimentalFeatures.evalsView is enabled.

Navigation:

• Desktop: Header → More views → Evals
• Mobile: More sheet → Evals

Features:

• Filter eval results by free-text query, run, and score range
• Review list summaries (task, eval/run identity, timestamps, and score)
• Drill into full rationale, category scores, evidence references, and suggested follow-ups
• Open Scheduled Evals settings directly when setup is disabled

Insights View

Insights view surfaces categorized project insights and lets you turn findings into work.

Available when experimentalFeatures.insights is enabled.

Navigation:

• Desktop: Header → More views → Insights
• Mobile: More sheet → Insights

Features:

• Category-based insight browser with run metadata and status indicators
• Manual insight generation plus refresh actions for latest insight runs
• Dismiss/archive/unarchive insight records as they age
• Create triage tasks from selected insights directly from the view

Reliability View

Reliability view summarizes in-review pipeline health so operators can spot bounce/merge instability trends without leaving the dashboard.

Navigation:

• Desktop: Header → More views → Reliability
• Mobile: More sheet → Reliability

Features:

• Headline 7-day in-review success rate (derived as 1 - inReviewFailureRate7d) with color thresholds: success for ≥95%, warning for ≥90%, error below 90%; shows Insufficient data when the metric is null
• Per-day in-review flow table showing tasks that entered in-review versus tasks bounced back to in-progress
• In-review duration percentiles (P50 and P95) plus sample count
• Merge-attempt distribution stats including mean, max, and histogram buckets
• Auto-refreshes every 60 seconds

For the backing API and windowDays query parameter, see architecture.md.

Dev Server View

Dev Server view manages detected dev server commands, preview URLs, and live logs for local development.

Available when experimentalFeatures.devServerView is enabled (devServer is treated as a legacy alias).

Navigation:

• Desktop: Header → More views → Dev Server
• Mobile: More sheet → Dev Server

Features:

• Detect candidate dev server commands and choose which command/session to run
• Start, stop, and restart the current server session
• Manage preview URLs with embedded preview and Open in new tab fallback
• Tail live logs, load older history, and refresh session status

For module-level behavior and API surfaces, see Dev Server modules.

Stash Recovery View

Stash Recovery view helps recover orphaned merger autostashes (fusion-merger-autostash:*) left behind when merge restore could not fully complete.

Navigation:

• Desktop: Header → More views → Stash Recovery
• Mobile: More sheet → Stash Recovery

Features:

• Lists orphaned stash entries grouped by source task ID (or Unknown source when unavailable)
• Surfaces provenance metadata from recovery events (sourcePhase, detectedByTaskId, detectedAt) to show where/when leftovers were captured and surfaced
• Inspect diff output for any orphaned stash before taking action
• Apply a stash to recover changes, or drop a stash with confirmation to permanently remove it

For API endpoints, see architecture.md.

Plugin Manager

Plugin management lives in Settings → Plugins → Fusion Plugins.

Features:

• Install bundled plugins or custom path-based plugins
• Enable/disable plugins, reload active plugins, and uninstall plugins
• Inspect plugin runtime state and transition feedback
• Edit and save plugin-defined settings schemas from the same panel

For full plugin lifecycle workflows (discovery, install, enable/disable, configure, update, uninstall, troubleshooting), see Plugin Management. For plugin-related settings and experimental toggles, see Settings reference.

Pi Extensions Manager

Pi extension management lives in Settings → Plugins → Pi Extensions.

Features:

• Add/remove Pi package sources (npm, git, or local)
• Reinstall the Fusion Pi package/skill bundle
• Enable/disable discovered extensions
• Manage extension, skill, prompt, and theme path lists in one place

For related global/project configuration behavior, see Settings reference.

Task Detail Modal

Inspect task definition, logs, review feedback, comments, documents, workflow outcomes, model overrides, and task routing from a single modal.

• The priority chip in task metadata is an inline picker: you can change priority directly without entering full edit mode.
• Execution mode has a read-mode inline lightning-bolt toggle for Fast mode on/off without opening the full edit form.
• These two metadata controls share matched sizing/alignment in read mode (including mobile wrapping) so they behave like a single polished control group.
• Task metadata also shows compact Created / Updated timestamps: recent values render as relative time (just now, Xm, Xh, Xd) and older values switch to short month/day dates; these stay grouped on one row across desktop and mobile widths for a compact metadata layout.
• Eligible existing tasks (triage, todo, in-progress, in-review) expose a GitHub tracking section directly in Task Detail, even when tracking is currently disabled.
• The GitHub tracking section now defaults to a compact summary row; use the disclosure arrow to expand linked-issue details plus tracking edit controls.
• In shared task edit/create forms, GitHub Tracking appears at the bottom of More options, after Workflow Steps.
• From this section you can explicitly enable/disable tracking and manage a per-task repo override (owner/repo). Clearing the override saves null and falls back to project/global defaults.
• In in-review, pull-request controls/status (including stall badges) are in a dedicated Pull Request tab instead of the Definition tab.
• The Review tab is separate from Comments: Review shows actionable PR/reviewer feedback and same-task revision controls, while Comments remains the general collaboration thread.
• Request revision in Review resumes work on the same task ID (no refinement task): in-progress tasks get steering injection, while in-review tasks are moved back to in-progress for the same branch/worktree revision pass.
• Review supports a manual Refresh action in-place: PR mode pulls latest GitHub review state/decision, while direct mode rehydrates reviewer-agent feedback from persisted task data (no GitHub call).
• In direct/non-PR auto-merge mode, Review renders normalized reviewer-agent feedback (verdict/step/timestamp/detail) with dedicated loading/error/empty states; it does not require users to read raw agent logs.

Identifying high-impact blockers

Use blocker fan-out signals on task cards and in the footer status bar to spot blockers with high downstream impact:

• Blocks N counts active downstream dependents in triage, todo, in-progress, or in-review.
• FN-3942 immediate signal: blockers with at least 5 active todo dependents (activeTodoCount >= 5) are marked High fan-out.
• FN-3954 escalation signal: a high-fan-out blocker is upgraded to Escalated only after it remains in in-progress/in-review past staleHighFanoutBlockerAgeThresholdMs (age source: columnMovedAt ?? updatedAt).
• Escalation payload surfaced in UI includes blocker ID, active todo downstream count, total active downstream count, and computed blocking age.
• Done and archived downstream tasks remain visible for debugging context but do not count toward the todo threshold.
• The badge tooltip shows active totals and, when escalated, the computed blocking age context.
• (stale) markers mean the dependent is blocked through blockedBy and matches stale conditions that clearStaleBlockedBy self-healing should clear automatically.
• Stale dependencies[] links are shown for awareness but are not auto-cleared by clearStaleBlockedBy.
• The executor footer summarizes the top escalated blocker (deterministic rank: highest todo fan-out, then highest active total, then oldest age, then stable task ID).

Recommended workflow: ordinary chains stay as Blocks N so noise stays low, high-fan-out blockers stand out immediately, and only long-lived high-impact blockers trigger explicit escalation.

Logs → Agent Log view

The Logs tab includes an Agent Log subview designed for debugging long-running and tool-heavy sessions:

• Full thinking, tool_result, and tool_error payloads are shown without entry-content truncation.
• Raw tool output is rendered as multiline blocks, preserving line breaks and indentation.
• The initial load fetches a recent page, then Load More progressively prepends older history.
• Live streaming appends new entries in chronological order while preserving your scroll position when loading older pages.
• The Markdown / Plain toggle lets you switch between formatted markdown and literal/raw text rendering.
• The Tools: On/Off toggle shows or hides tool-call rows (tool, tool_result, tool_error) so you can focus on narrative/thinking output when needed.
• Both display preferences persist across sessions via local storage (fn-agent-log-markdown and fn-agent-log-tool-output).

The Routing tab shows:

• effective node
• routing source (task override vs project default vs local)
• unavailable-node policy value
• per-task node override controls (locked while task is active)

Project-wide routing defaults are configured in Settings → Node Routing.

Node Dashboard

The Node Dashboard provides a mesh view of connected Fusion nodes. Each node can be a local instance or a remote headless node (fn serve).

Navigation:

• Desktop: Header node controls / overflow entry
• Mobile: MobileNavBar → More sheet → Nodes (shown only when experimentalFeatures.nodesView is enabled)

Local/Remote Node Switching

When remote nodes are available, the dashboard header displays a node status indicator:

• Local mode — Shows a green "Local" badge, indicating the dashboard is connected to the local Fusion instance
• Remote mode — Shows the remote node name with its connection status (online/offline/connecting)

Click the chevron next to the status indicator to open the node selector dropdown:

• Local — Switch back to viewing the local Fusion instance
• Remote nodes — Select a remote node to view its tasks, projects, and status

Remote Node Onboarding Discovery

When adding a remote node in the Nodes view, onboarding now discovers projects directly from the target node before the node is registered.

1. Enter the remote URL (and API key when required)
2. Click Discover Remote Projects
3. Fusion calls the remote node's /api/projects endpoint and shows discovered projects (name, path, status)
4. For selected local projects, Fusion only auto-prefills a node path when there is exactly one discovered project with the same name
5. If discovery fails, onboarding shows an inline error and does not prefill remote mappings for that attempt
6. If discovery succeeds with zero projects, onboarding shows an explicit empty state

This keeps remote path mappings anchored to remote-authoritative data instead of local guesses.

How Node Switching Works

1. The node selector appears in the header when remote nodes are registered in the mesh
2. Selecting a remote node routes all API calls through the proxy endpoint (/api/proxy/:nodeId/...)
3. Task data (projects, tasks) is fetched from the remote node and displayed in the dashboard
4. SSE events from the remote node are streamed via the proxy and update the dashboard in real-time
5. Selecting "Local" returns to the local Fusion instance with full local data

Benefits of Remote Node Viewing

• Monitor task progress across distributed teams
• View task status on remote headless nodes without direct SSH access
• Compare project health across multiple Fusion instances
• Stay informed about remote agent activity and task completion

Node Status Indicators

Status	Color	Meaning
Online	Green	Node is connected and responsive
Offline	Red	Node is unreachable or shut down
Connecting	Yellow (pulsing)	Connection attempt in progress

Project availability and path visibility

Node and project surfaces now use per-node project mappings (nodeMappings) instead of a single project.nodeId assumption.

• Node cards / counts include only projects with an available: true mapping for that node.
• Node Details modal lists one row per project available on the selected node and shows:
  • project name
  • project ID
  • configured path for that node
• Project node filter in the Projects view is built from available mappings and uses canonical node-name resolution (Node.name → mapping name → source node name → node ID).
• Project cards show node availability as compact Node → /path rows:
  • up to 3 rows inline
  • +N more summary when additional mappings exist
  • single-node projects still show the configured path clearly
• Mappings marked available: false are excluded from node counts, node filter options, node detail project rows, and project-card availability summaries.

Persistence

The selected node persists across browser sessions via localStorage. If the selected remote node is unregistered, the dashboard automatically falls back to local mode.

Native shell connection flow

If you use Fusion from a native shell (mobile app or desktop shell in remote mode), dashboard startup is gated by shell onboarding until a connection is selected.

For the canonical workflow (first-run onboarding, QR/manual setup, saved profiles, and desktop local/remote handoff), see Native Shell Connection Guide.

Remote Access (Settings)

Dashboard remote controls live in Settings → Remote Access.

From this section, operators can:

• Configure Tailscale and Cloudflare provider fields
• Activate the current provider
• Start/stop tunnel lifecycle manually
• Generate login URLs / QR payloads using persistent or short-lived token mode

For setup prerequisites, security caveats for tokenized URLs/QR links, and troubleshooting, use the canonical Remote Access runbook.

Skills API

The Skills API provides endpoints for managing execution skills. Skills are toggled via project-scoped settings in .fusion/settings.json.

GET /api/skills/discovered

List all discovered skills with their enabled state.

Response: 200 OK

{
  "skills": [
    {
      "id": "npm%3A%40example%2Fskill::skills/foo/SKILL.md",
      "name": "foo/SKILL.md",
      "path": "/path/to/skills/foo/SKILL.md",
      "relativePath": "skills/foo/SKILL.md",
      "enabled": true,
      "metadata": {
        "source": "npm:@example/skill",
        "scope": "project",
        "origin": "package"
      }
    }
  ]
}

Skill ID Format: encodeURIComponent(metadata.source) + "::" + relativePath

• Top-level skills use source: "*"
• Package skills use the package source identifier

Error Response: 404 Not Found

{
  "error": "Skills adapter not configured",
  "code": "adapter_not_configured"
}

GET /api/skills/:id/content

Fetch a skill's SKILL.md content and supplementary file metadata.

Response: 200 OK

{
  "content": {
    "name": "foo/SKILL.md",
    "skillMd": "# Foo Skill\n...",
    "files": [
      {
        "name": "examples",
        "relativePath": "skills/foo/examples",
        "type": "directory"
      },
      {
        "name": "example.ts",
        "relativePath": "skills/foo/examples/example.ts",
        "type": "file"
      }
    ]
  }
}

Error Responses:

• 400 Bad Request — invalid encoded skill ID (code: "invalid_skill_id")
• 404 Not Found — skill not found (code: "skill_not_found") or adapter missing (code: "adapter_not_configured")

PATCH /api/skills/execution

Toggle a skill's enabled/disabled state.

Request Body:

{
  "skillId": "npm%3A%40example%2Fskill::skills/foo/SKILL.md",
  "enabled": true
}

Response: 200 OK

{
  "success": true,
  "skillId": "npm%3A%40example%2Fskill::skills/foo/SKILL.md",
  "enabled": true,
  "persistence": {
    "scope": "project",
    "targetFile": "/path/to/.fusion/settings.json",
    "settingsPath": "packages[].skills",
    "pattern": "+skills/foo/SKILL.md"
  }
}

Toggle Semantics:

• Top-level skills (origin: "top-level"): Mutate settings.skills
  • Enable: ensures +<relativePath> exists, removes -<relativePath>
  • Disable: ensures -<relativePath> exists, removes +<relativePath>
• Package skills (origin: "package"): Mutate settings.packages[].skills for the matching metadata.source
  • If the package entry is a string, it's converted to an object { source: <same>, skills: [] }
  • Other package fields (extensions, prompts, themes) are preserved

Error Responses:

• 400 Bad Request — Invalid request body

  { "error": "skillId is required", "code": "invalid_body" }
  

• 404 Not Found — Adapter not configured

  { "error": "Skills adapter not configured", "code": "adapter_not_configured" }
  

GET /api/skills/catalog

Fetch the skills.sh catalog with optional authentication.

Query Parameters:

• limit (optional): Number of results (default 20, max 100)
• q (optional): Search query string

Response: 200 OK

{
  "entries": [
    {
      "id": "example-skill",
      "slug": "example-skill",
      "name": "Example Skill",
      "description": "An example skill",
      "tags": ["utility"],
      "installs": 100,
      "installation": {
        "installed": true,
        "matchingSkillIds": ["npm%3A%40example%2Fskill::skills/example/SKILL.md"],
        "matchingPaths": ["skills/example/SKILL.md"]
      }
    }
  ],
  "auth": {
    "mode": "unauthenticated",
    "tokenPresent": false,
    "fallbackUsed": false
  }
}

Authentication Flow:

1. If SKILLS_SH_TOKEN env var is present, use authenticated request
2. If authenticated request returns 400/401/403, retry without authentication (fallback mode)
3. If no token, use unauthenticated request directly

Unauthenticated Short-Query Behavior:

• Public skills.sh /api/search requests are only sent when q has at least 2 characters
• For omitted, empty, or 1-character queries, the API returns 200 with { entries: [] }
• This applies both to direct unauthenticated mode and authenticated-to-unauthenticated fallback mode, preventing upstream 400 Bad Request responses during initial load

Auth Mode Values:

• authenticated — Request made with token
• unauthenticated — Request made without token (no token available)
• fallback-unauthenticated — Initial authenticated request failed with 401/403, retried without token

Error Response: 502 Bad Gateway

{
  "error": "Upstream request timed out",
  "code": "upstream_timeout"
}

Possible error codes:

• upstream_timeout — Request timed out
• upstream_http_error — Upstream returned an error status
• upstream_invalid_payload — Upstream returned invalid response format

Agent Import

The Agent Import feature allows you to import agents from Agent Companies packages. When importing agents from companies.sh or local directories, Fusion now also persists any skill definitions from the package.

Launch Points

You can open Agent Import from:

• Agents view → Controls popup → Import
• Agent Detail header → Import (opens directly to the companies.sh browse catalog)

How It Works

1. Select Source: Choose to import from:

   • The companies.sh catalog (browse and search)
   • A local directory containing AGENTS.md files
   • A single manifest file (.md or .txt)
   • Paste manifest content directly

2. Preview: Review the agents and skills that will be imported before confirming

3. Import: Upon confirmation:

   • Agents are created in Fusion's agent store
   • Skills are persisted to skills/imported/{companySlug}/{skillSlug}/SKILL.md
   • Each skill's SKILL.md contains YAML frontmatter with skill metadata and the instruction body

Skill Persistence

Skills from Agent Companies packages are persisted to the project-local skills directory:

{projectRoot}/
  skills/
    imported/
      {companySlug}/          # slugified company name or "unknown-company"
        {skillSlug}/          # slugified skill name
          SKILL.md            # skill manifest with frontmatter + instructions

Collision Handling: If a SKILL.md file already exists at the target path, the import skips that skill (does not overwrite). This prevents accidental data loss.

Path Safety: All path segments are slugified to prevent directory traversal attacks. Special characters are removed and whitespace is normalized to hyphens.

Import Result

The import result shows:

Agents:

• Number of agents created
• Number of agents skipped (already exist)
• Number of errors (import failures)

Skills:

• Number of skills imported (written to disk)
• Number of skills skipped (already exist)
• Number of skill errors (write failures)

API Response

The POST /api/agents/import endpoint returns skill import results:

{
  "companyName": "Example Co",
  "companySlug": "example-co",
  "created": [{ "id": "agent-1", "name": "CEO" }],
  "skipped": [],
  "errors": [],
  "skillsCount": 3,
  "skills": {
    "imported": [
      { "name": "review", "path": "skills/imported/example-co/review/SKILL.md" },
      { "name": "strategy", "path": "skills/imported/example-co/strategy/SKILL.md" }
    ],
    "skipped": [],
    "errors": []
  }
}

The skills object contains detailed import outcomes for each skill from the package.

Styling Guide

The dashboard's CSS is split into a global stylesheet (packages/dashboard/app/styles.css) and per-component files (packages/dashboard/app/components/ComponentName.css). Each ComponentName.tsx imports its stylesheet at the top.

Rule: New CSS for a component goes in app/components/ComponentName.css, NOT styles.css. Only design tokens, primitives (.btn, .card, .modal, .form-input), and cross-component @media overrides belong in the global file.

PR tab note: PrPanel cards use tokenized .pr-card grid spacing (padding + gap) and boxed token-based hint callouts for empty/loading states. Manual PR merges now show in-progress feedback (Merging… button state + status hint) until the merge call resolves.

The index.html shell is templated server-side: the server injects a per-user <link rel="modulepreload"> for the last-used taskView chunk, sourced from Vite's dist/client/.vite/manifest.json and kb:<projectId>:kb-dashboard-task-view in localStorage.

Design tokens

styles.css is the source of truth for tokens (--space-*, --radius-*, --shadow-*, --transition-*, --font-*, --header-height, --mobile-nav-height, --standalone-bottom-gap, --overlay-padding-top) and color variables (--bg, --surface, --card, --text, --text-muted, status colors --triage/--todo/--in-progress/--in-review/--done, semantic --color-success/--color-error/--color-warning/--color-info, status backgrounds --status-*-bg).

Always reference tokens. Never hardcode pixels, hex, or rgba() in component CSS — the only exception is inside :root/theme blocks where tokens are defined. For translucent backgrounds use color-mix(in srgb, var(--color) X%, transparent), not rgba().

Theme system

Dark/light modes via data-theme; 54 color themes via data-color-theme (lazy-loaded from app/public/theme-data.css).

• Base tokens (--bg, --surface, etc.) — redefine in :root, [data-theme="light"], and every theme block.
• Semantic tokens (--autopilot-pulse, --event-error-text, --badge-mission-*, --fab-*) — :root + [data-theme="light"] only; no per-color-theme overrides.
• Status tokens (--triage, --todo, etc.) — redefine per theme block.

status-colors-theme.test.ts iterates all theme blocks to catch regressions.

Component classes

Reuse existing primitives from styles.css:

• Buttons: .btn, .btn-primary, .btn-danger, .btn-warning, .btn-sm, .btn-icon, .btn-icon--active, .btn-badge. All inherit :focus-visible via --focus-ring-strong and :active via transform: scale(0.97).
• Modals: .modal-overlay[.open], .modal, .modal-lg, .modal-header, .modal-close, .modal-actions, .modal-actions-left/right. Overlay pads top with --overlay-padding-top.
• Forms: .form-group, .input, .select, .checkbox-label, .form-error. Inputs in .form-group get focus styles automatically.
• Cards: .card, .card-header, .card-id, .card-title, .card-meta, .card-status-badge--{triage,todo,in-progress,in-review,done,archived}.
• Utility: .touch-target (44px min), .visually-hidden.

Don't create parallel button/form variants — add states (:hover, :focus-visible, :active) to the existing primitives.

Mobile responsive

Breakpoints: 768px (primary mobile), 1024px (tablet min-width: 769px and max-width: 1024px), 640px (compact), 480px (xs). Mobile overrides go in @media (max-width: 768px) blocks at the bottom of styles.css after base styles.

Bottom spacing: --mobile-nav-height (44px) + env(safe-area-inset-bottom, 0px) + --standalone-bottom-gap (0/8px PWA). All bottom-positioned mobile elements compose those. When the soft keyboard opens, the mobile nav bar stays pinned to page bottom cross-platform; the executor footer keyboard-collapse pin is iOS-only. On Android (interactive-widget=resizes-content), the footer keeps its stacked position above the nav bar to avoid overlap after keyboard dismiss.

Touch targets: Standing button-freeze directive supersedes per-button touch-target guidance. For non-button elements, primary controls (nav bar, FAB, tab action rows, modal CTAs, list-row tap targets, form controls) must be ≥36px on mobile. Secondary controls inside a card/list-row where the row itself is the tap target stay compact (24–28px or small chips).

Safe area: max(var(--space-md), env(safe-area-inset-left, 0px)) for notch-aware horizontal padding.

Secrets management in Settings

Manage project and global secrets directly inside Settings → Project → Secrets. This section embeds the existing Secrets UI in the settings content panel so you no longer need a footer "Manage secrets" link to leave the modal.

Lazy-Loaded Heavy Views

These 19 views are lazy-loaded via React.lazy() with <Suspense fallback={null}>. prefetchLazyViews() warms chunks once on mount via requestIdleCallback. Do not make these eager.

• AgentsView
• NodesView
• ChatView
• MemoryView
• DevServerView
• SecretsView
• InsightsView
• DocumentsView
• SkillsView
• ResearchView
• ReliabilityView
• EvalsView
• TodoView
• GoalsView
• StashRecoveryView
• SetupWizardModal
• PluginManager
• PiExtensionsManager
• AgentDetailView

When adding or removing entries, update packages/dashboard/app/__tests__/lazy-loaded-views-docs.test.ts (expected set + count).

CSS testing

Use packages/dashboard/app/test/cssFixture.ts:

import { loadAllAppCss, loadAllAppCssBaseOnly } from "../test/cssFixture";
const allCss = await loadAllAppCss();          // styles.css + all component .css
const baseOnly = await loadAllAppCssBaseOnly(); // strips @media/@supports

Never directly readFileSync('../styles.css') — an ESLint rule (no-restricted-syntax in eslint.config.mjs) bans this and points at cssFixture.ts. vitest.config.ts has test.css: { include: [/.+/] } so component CSS imports inject into jsdom for getComputedStyle assertions.

File browser editor & autosize textarea

• FileEditor.tsx is CodeMirror 6-only (no <textarea> fallback). Language resolution: packages/dashboard/app/utils/codemirror-language.ts.
• For chat-style composer fields use packages/dashboard/app/hooks/useAutosizeTextarea.ts. Pattern: height = "auto" then clamp scrollHeight to min/max in useLayoutEffect. Pair with resize: none; keep overflow-y: hidden while under the max-height cap and switch to overflow-y: auto only after content exceeds the cap.

File-path links

Reuse packages/dashboard/app/utils/filePathLinkify.tsx and FileBrowserContext. Wrap plain text with linkifyFilePaths(...), mixed JSX with linkifyReactChildren(...). Mount under FileBrowserProvider and route clicks through its openFile(path, { workspace?, line?, col? }).

Common pitfalls

• --surface-hover undefined — reference with a fallback (var(--surface-hover, rgba(0,0,0,0.03))) or define explicitly.
• BEM specificity — when a container state class and an element modifier target the same node, the container can win. Use :not(.modifier) to scope.
• CSS @media detection — track brace depth to confirm a rule is mobile-scoped; don't scan backwards for the nearest @media. Many components are global even if visually mobile-only.
• Mobile board scroll-snap (FN-001) — scroll-snap-type: x mandatory on mobile .board causes iOS Safari to compress the viewport when switching from ListView. Use x proximity + overflow-anchor: none.
• lucide-react icon adds — update vi.mock("lucide-react") test mocks immediately; missing exports cascade.
• .spin is global — don't redefine the generic spin keyframes in component CSS.

Integration Branch Push to Origin

The merge-advance notice includes an explicit Push to origin action for the dynamically resolved integration branch.

• The branch name is resolved from project settings, then origin/HEAD, then fallback; UI copy and API behavior must remain dynamic.
• Push status probes compute ahead/behind counts and disable push when there is no origin, no upstream tracking ref, the branch is not ahead, or a Fusion merge lock is active.
• The mutating route performs a TOCTOU merge-lock recheck immediately before building push argv.
• Standard push is git push origin refs/heads/<branch>:refs/heads/<branch> with no plain --force path.
• Advanced mode enables opt-in --force-with-lease=refs/heads/<branch>:<localSha> only.
• Non-fast-forward and lease-stale failures surface actionable messaging with Smart Pull.
• Every attempt records mutationType: "push:origin" run-audit metadata: integrationBranch, remote, localSha, remoteSha, aheadCount, behindCount, forceWithLease, outcome, optional stderrPreview, and durationMs.
• Push remains explicit user authorization only through dashboard HTTP routes (no scheduler/heartbeat auto-push).