- Resize the dashboard TUI splash logo to scale with the terminal: a Colossal-font variant kicks in on terminals ≥70 cols × 16 rows; gradient palette is now strictly white → blueBright → blue (no cyan/magenta)
- Lower-clip TUI Kanban cards to a single-line title and constrain each column with overflow hidden so cards no longer bleed past the header row
- Center the TerminalModal vertically by overriding overlay alignment via :has() and trimming the modal height
- Match GitHub Import modal width to the file browser modal (90vw / 1600px max)
- Render the Usage indicator as a top-right popover on desktop instead of a centered modal (mobile keeps the full-screen sheet)
- Widen SetupWizard / Model Onboarding modals (540→880px, 560→880px) and replace the washed-out done step indicator background with a solid success swatch
- Drop the duplicate mailbox icon button from the desktop header (view switch covers it; overflow menu still surfaces it on compact)
- Loosen agent health detection: bump the heartbeat grace multiplier 2× → 4× and the staleness floor 60s → 5min so transient ticks don't flag agents Unresponsive
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Batch 2 of runMaintenance() runs ~10 recover passes back-to-back, each
calling listTasks({ column: ... }) without slim. On busy boards this
materializes every task's activity log into memory ~10× per cycle,
walking the dashboard heap toward the 8 GB V8 limit until OOM. The
archive pass at line 610 already had this fix; extend it to the in-progress
and in-review recover passes that only read steps / paused / worktree /
mergeDetails / postReviewFixCount — all included in the slim projection.
Triage recovers are left non-slim because hasLatestSpecReviewApproval
scans task.log to find the most recent spec review; the triage column
is small so the memory cost is bounded.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Move misclassified .mesh-topology, .connect-node, .nodes-view-topology, .node-status-indicator, and pulse-warning keyframe from AgentReflectionsTab.css to NodesView.css; relocate node-related mobile @media rules from SettingsSyncLog.css so NodesView renders standalone
- Add missing CSS import on GitManagerModal so its .gm-* styles load without depending on ScriptsModal being mounted
- Bump desktop max-widths for FileBrowser, Terminal, GitManager, GitHubImport, and WorkflowStepManager modals; raise selector specificity to .modal.X so the base .modal { width: 480px } can't win the cascade
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Add a new /api/git/diff/file endpoint with path and staged query handling, including validation and untracked-file diff support
- Introduce fetchGitFileDiff API client wiring and update GitManagerModal to load diffs for selected staged/unstaged file rows
- Improve file-row UX with active/focus styling, keyboard activation, and guarded async diff loading/error states
- Expand dashboard route and modal tests to cover per-file diff behavior, query validation, and stage/unstage interaction isolation
The global SSE broadcast called res.write() without checking the return
value, so a paused or backgrounded client would silently accumulate
every store event for every entity (tasks, missions, plugins, agents,
chat, ...) into res.outputData until the dashboard process OOMed.
Add a 4 MB writableLength threshold; when exceeded, tear down the
connection so the OS releases the buffer.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The global SSE broadcast called res.write() without checking the return
value, so a paused or backgrounded client would silently accumulate
every store event for every entity (tasks, missions, plugins, agents,
chat, ...) into res.outputData until the dashboard process OOMed.
Add a 4 MB writableLength threshold; when exceeded, tear down the
connection so the OS releases the buffer.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Treat the completion view as an effective final step so all prior indicators/connectors remain marked done
- Mark the first-task step as completed before tracking onboarding completion and clear it from skipped state
- Replace hardcoded success colors in setup wizard stepper done styles with theme-aware done status tokens
- Add regression assertions verifying all step indicators and connectors stay in done state on the completion screen
- Render the compact project switch trigger for tablet layouts by gating on compact mode instead of mobile-only mode
- Extend ProjectSelector compact-switch styles to the tablet media range so the trigger and dropdown render correctly beside the logo
- Update tablet header control regression coverage to expect the compact switch affordance and absence of the desktop selector trigger
- Add Header tablet interaction tests covering Escape key, outside click, and project selection closing behavior for the compact switch dropdown
- Add cross-browser scrollbar styling for settings sidebar and content panes using existing design tokens
- Extend settings mobile CSS tests to assert the new scrollbar rules for WebKit and Firefox-compatible properties
- Refactor dashboard TUI app tests to render via React.createElement for compatibility with current test tooling
- Tidy mission e2e mock ID generators to satisfy formatting/quality gate checks
- Switch the DirectoryPicker Select action to shared btn btn-primary classes with explicit button type
- Move select-button padding to component-scoped DirectoryPicker.css tokens and remove hardcoded !important overrides
- Extend DirectoryPicker tests to assert the expected class contract before click behavior
- Resolve merge conflicts while preserving current provider grouping and onboarding flow
- Render GitHub login instructions in the onboarding connect CTA while auth is pending
- Add a dedicated test id for the GitHub onboarding instruction message to satisfy regression coverage
A rebase preserved two copies of the eager `import "./components/ChatView.css"`
in App.tsx (one at the top of the imports block, one inline among the
lazy view declarations). Keep the top-of-file copy, remove the inline
duplicate.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- ChatView mobile: hide the top sidebar header's "New Chat" button on
<=768px and show the existing footer button as a pinned full-width
action with safe-area padding. Both buttons existed in JSX; only
CSS was needed.
- styles.css: comments compose/edit textareas use var(--surface) so
they remain visible against the task detail modal's --card panel
(in light theme --bg and --card both resolve to #ffffff).
- Includes resolved-conflict changes from fusion/fn-2469 in
ModelOnboardingModal and SettingsModal (and the related test).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- ChatView mobile: add @media (max-width: 768px) so the session-list
sidebar goes full-width and the thread is hidden when the sidebar
is visible (and vice-versa). The component already had the state,
back-button (ChevronLeft), and visibility toggling — only the CSS
was missing.
- TaskComments compose/edit textareas now use var(--surface) for
background. The base .spec-editor-feedback uses var(--bg), which
in light theme resolves to #ffffff — the same as the task detail
modal's --card panel — making the comment box invisible.
- ChatView CSS is now imported eagerly from App.tsx so it bundles
into the main CSS file. Previously the lazy ChatView JS chunk
loaded its CSS via an async <link> tag, producing a brief flash
of unstyled chat UI on first render.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- ChatView mobile: add @media (max-width: 768px) so the session-list
sidebar goes full-width and the thread is hidden when the sidebar
is visible (and vice-versa). The component already had the state,
back-button (ChevronLeft), and visibility toggling — only the CSS
was missing.
- TaskComments compose/edit textareas now use var(--surface) for
background. The base .spec-editor-feedback uses var(--bg), which
in light theme resolves to #ffffff — the same as the task detail
modal's --card panel — making the comment box invisible.
- ChatView CSS is now imported eagerly from App.tsx so it bundles
into the main CSS file. Previously the lazy ChatView JS chunk
loaded its CSS via an async <link> tag, producing a brief flash
of unstyled chat UI on first render.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The AgentsView mobile header was wrapping onto two rows because
DocumentsView.css carried leaked agents-view-* rules forcing
flex-wrap: wrap on the header/controls/title and width: 100% on the
primary-actions group. Removed those stale overrides so AgentsView's
own mobile rules can take effect.
Header changes:
- Drop the Refresh button.
- Reorder: Bot icon, view-toggle, then primary-actions on the far right
(Controls icon left of New Agent +). Controls is now icon-only with
no text label on every viewport.
- Bump the Bot icon to 24px (was 20) and give its mobile container a
32x32 footprint to align with the view-toggle pill.
- Pin primary-actions to the right via space-between on the controls
flex container; everything stays on one row at <=320px.
Layout: move AgentMetricsBar (stats cards) above the agent collection
so stats sit on top of the list/board/tree/org views.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Lower the multi-panel grid threshold from 100×24 to 80×20 so standard
terminals render the full layout instead of falling back to single-pane.
- Agents view stacks list above detail when cols < 80; Tab focus model
preserved.
- Board view: press [n] to open a centered create-task popup. Title is
edited via ink-text-input; submit calls TaskStore.createTask, refresh
the task list on success.
- System panel: shows the tokenized auth URL alongside the token, says
"no auth" plainly when --no-auth was used, and prompts with
"Press [Enter] to open in browser". Enter (while System is focused)
spawns the platform's URL opener (open / start / xdg-open).
- Tighten log-row layout: each row is now a single Text with inline
spans + fixed-width prefix slot so columns align and the highlight
bar spans the full row.
- Default status section is "logs" so the live feed is what users see
when they run `fn`.
- Drop the ASCII circle brand mark from the splash; FUSION block
letters + tagline + spinner only, pinned to the top-left.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Default activeSection is now "logs" so the live log feed is what the
user sees first when running `fn`.
- Splash drops the ASCII circle brand mark; only the FUSION block-letter
word and tagline remain. The splash always pins to the top-left rather
than centering, so it doesn't shift on resize. Compact terminals show
plain "FUSION" text instead of block letters.
- Log rows render as a single Text with inline-colored spans (marker,
timestamp, level, prefix, message). Prefix sits in a fixed 14-char
slot so the message column aligns across rows, and the selection
highlight now spans the full row width.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mostly mechanical cleanup left over from the earlier test-consolidation pass:
- Update import paths to ../../ for mocks now that test files moved deeper
- Simplify mock setup (drop usePluginUiSlots inline mock, etc.)
- Move engine ipc + runtimes tests into __tests__/ subdirs
- Move dashboard utils tests into __tests__/ subdir
- Refresh fusion-plugin-hermes-runtime/dist artifacts
build-exe.test.ts: spawn-import fix from a parallel branch (resolved during
worktree merge of the CSS extraction work).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Split app/styles.css from ~40k lines down to ~4.5k. Created 56 co-located
component CSS files in app/components/, each imported by its owning .tsx.
The remainder of styles.css holds genuinely global rules (design tokens,
.btn/.card/.modal/.form-input primitives, cross-component @media overrides).
- Lazy-load 13 heavy views (AgentsView, RoadmapsView, NodesView, etc.) via
React.lazy + Suspense; prefetch all chunks on idle so first navigation is
instant. Initial JS bundle: 1.58 MB → 1.16 MB (-26%). Initial CSS bundle:
635 kB → 471 kB (-26%); the rest splits into 13 per-view chunks.
- Add app/test/cssFixture.ts exposing loadAllAppCss() + loadAllAppCssBaseOnly()
so CSS regression tests load the full per-component bundle (mirroring Vite
source order). Migrate 30+ tests off direct readFileSync('../styles.css').
- Enable test.css: { include: [/.+/] } in vitest.config.ts so component CSS
imports actually inject styles in jsdom (fixes getComputedStyle assertions).
- Add ESLint rule (no-restricted-syntax) banning direct styles.css reads in
dashboard test files; points at loadAllAppCss() instead.
- Restore lost utility classes (.text-muted, .text-secondary, .text-dim,
.form-input) and rescue dropped chat tool-call rules into QuickChatFAB.css.
- Mobile fixes along the way: scroll containment for view containers
(min-height:0 + -webkit-overflow-scrolling), QuickChatFAB full-screen on
mobile (with safe-area-inset for iOS home bar), AgentsView single-row
header layout, ActivityLogModal close button on right, model-combobox
z-index above the mobile quick-chat panel.
- Bug fix: SkillsView toggle was display:none which hid the input from the
accessibility tree; replaced with the visually-hidden pattern so screen
readers + getByRole still find the checkbox.
- Bug fix: standalone Delete button in TaskDetailModal for triage-column
tasks (Actions dropdown is hidden in triage state, so previously no way
to delete a freshly-created task without status change first).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The selection arrow on the dashboard TUI's logs panel was rendering for
the wrong row: the list was reversed (newest-first) but selectedLogIndex
is in chronological order, and the viewport was anchored to the buffer's
tail. On a long log buffer the cursor pointed at the oldest entry, which
was always offscreen.
Switch the display to chronological order (oldest top, newest bottom —
matches tail/less/k9s), compute the viewport from the cursor so the
highlighted row is always visible, and surface "↑ N more / ↓ N more"
hints when entries scroll out of view. addLog now keeps the cursor
pinned to the tail when the user hasn't navigated away, so live logs
follow the latest event.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace the legacy ANSI-based DashboardTUI with an Ink/React rewrite
under packages/cli/src/commands/dashboard-tui/, delete the standalone
@fusion/tui package, and make `fn` (no args) launch the dashboard.
The new TUI keeps the existing 5-panel status mode (system, logs,
utilities, stats, settings) but adds an interactive mode (b/a/g) with
three views: a kanban board with project picker and per-task detail,
an agents list+detail with state management, and a settings editor.
Bordered focus-aware panels, solid-background help overlay, static
all-blue FUSION splash that adapts to small terminals. DashboardTUI
and DashboardLogSink public API are unchanged so dashboard.ts only
needed import-path updates plus interactiveData/loadingStatus wiring.
Also adds zod to @fusion/dashboard to satisfy a peer dep introduced
by pi-coding-agent 0.70.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Move all co-located *.test.* files into sibling __tests__/ directories so the
layout is consistent across packages (159 renames + content-rewrite moves).
Updates relative imports, vi.mock specifiers, and __dirname/import.meta.url
path resolutions where tests read fixtures from disk.
- Drop tracked tsc-emit alongside engine .ts sources (auth-storage/logger/
skill-resolver/context-limit-detector/pi.{js,d.ts,*.map}). These were
accidentally committed in a merge and the stale pi.js was masking a real
test-mock vs source mismatch (tests imported "../pi.js" and vite preferred
the stale build over pi.ts).
- Add packages/engine/.gitignore to block future src/*.{js,d.ts,map}.
- Refactor plugin pi-module seams (openclaw/paperclip/hermes) to ESM-import
createFnAgent / promptWithFallback / describeModel from @fusion/engine
instead of require()-ing packages/engine/src/pi.js. Adds @fusion/engine to
the two plugin package.jsons that were missing it; exports describeModel
from the engine public API.
- Fix engine test mocks now that they run against current pi.ts: add
ModelRegistry.create static to mocks in pi.test.ts and pi-create-fn-agent
.test.ts; switch three boundary-result toEqual assertions to toMatchObject
so the new content/isError fields don't trip exact-match comparison.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Gate unified progress display in ListView to tasks that are executing or in-progress
- Gate TaskCard progress bar and steps toggle behind the same active-execution visibility rule
- Update ListView and mobile list-card tests to cover hidden progress for non-executing todo tasks
- Expand TaskCard and mobile board tests to assert progress/toggle visibility for executing vs queued states
- Add ntfy base URL to global settings schema/types with persistence coverage and regression tests
- Extend dashboard settings API/routes and Settings modal UI to edit and save a custom ntfy server
- Update notifier runtime to use configured ntfy base URL when sending notifications
- Document the new setting and include a changeset for @runfusion/fusion
Drops @fusion/pi-claude-cli from runtime dependencies so
`pnpm install -g @runfusion/fusion` no longer 404s on the unpublishable
private workspace package. The pi extension is staged into
dist/pi-claude-cli/ at build time and resolved relative to the running
module, with the workspace require.resolve preserved as a dev fallback.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Wire priority through task creation and update API payloads for dashboard clients
- Add priority selection to TaskForm, New Task modal, and inline create card with default/reset behavior
- Support priority editing and display in Task Detail modal plus non-default priority badges on task cards
- Extend dashboard styles and component tests to cover priority selectors, rendering, and mobile behavior
- Regenerate fusion-plugin-hermes-runtime build artifacts and manifest metadata for runtime packaging
- Refactor Hermes runtime source into dedicated pi-module, runtime-adapter, and shared type modules
- Expand Hermes and engine plugin-runner tests to validate cross-runtime compatibility behavior
- Update getting-started, settings reference, and Hermes README docs to reflect the current runtime integration guidance
- Split New Agent step-0 into Presets and Custom tabs with presets-first default behavior
- Refine tab layout and styling in dashboard styles to address review feedback and improve responsive presentation
- Expand NewAgentDialog and AgentsView tests to cover tab switching flows and related route behavior
- Update agent documentation to reflect the revised new-agent dialog flow
Agents working on a task that depends on other tasks (e.g. documentation
alignment tasks needing the sibling tasks' specs) were repeatedly
rejected by the worktree boundary when reading .fusion/tasks/FN-NNNN/PROMPT.md,
which also contributed to the malformed-tool-result crash we just fixed.
Add a read-only exception to isWorktreeAllowedPath: the read/glob/grep
tools may access .fusion/tasks/*/PROMPT.md and .fusion/tasks/*/task.json
at the project root. Writes and bash cwd remain restricted.
Update the system-prompt boundary docs (executor.ts) so agents know the
exception exists and stop burning turns re-trying rejected reads.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
wrapToolsWithBoundary returned a bare {ok:false,error} object when an
agent tried to read/write/bash outside the worktree. pi-coding-agent
wraps tool returns into a toolResult message whose content field it
expects to be an array of content blocks; a bare object leaves content
undefined, which later crashes downstream with "Cannot read properties
of undefined (reading 'filter')" — the failure we've been chasing on
FN-2479 and similar.
Return { content:[{type:"text",text:...}], isError:true, ok:false, error:... }
so pi records a valid toolResult block while existing callers that
inspect .ok / .error still work.
Diagnostic evidence: transcript tail for the failing task showed three
consecutive `read` toolResults with content=array(len=0) (normalized
from undefined by our earlier guard) immediately before the assistant
message with stopReason="error".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Add QuickEntryBox state for a Fast toggle and include executionMode="fast" in create payloads when enabled
- Reset fast mode during form reset flows so follow-up creates default back to standard mode
- Add regression tests for fast toggle pressed state and payload behavior across Enter and Save submission paths
- Add tests ensuring fast mode is omitted when inactive and cleared after successful create, Plan, and Subtask flows
pi-coding-agent swallows its own exceptions onto session.state.errorMessage
without preserving a stack, so fusion sees only "Cannot read properties
of undefined (reading 'filter')" with no indication which message is
malformed. When promptSessionAndCheck rethrows an error that matches the
generic TypeError shape, dump the last few state.messages (role, content
type, toolName, stopReason) so the offending message can be identified
next run.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Move secondary agent sections below the collection area to support the agents-first layout
- Add a popup controls panel in AgentsView with updated styling and responsive behavior
- Expand dashboard tests to cover popup controls, CSS classes, and the new mobile/section ordering
- Resolve engine logger lint issues and update agents documentation/README for the new layout
When a branch contained commits already on main (duplicate cherry-picks),
the merger's local squash collapsed to an empty commit. The merger then
recorded that empty commit's SHA on mergeDetails.commitSha. The actual
content landed later on main as a different SHA via PR merge, but the
task kept pointing at the orphaned empty commit.
Symptom: TaskCard showed "N files changed" (falling back to
task.modifiedFiles), but the Changes tab in the modal showed nothing
because the API hit `git diff sha^..sha` on the empty commit and
returned no files.
Two fixes:
1. merger.ts: detect empty squash commits and skip storing commitSha,
logging clearly. recoverInterruptedMergingTasks → findLandedTaskCommit
already exists to backfill the right SHA when the real commit lands;
a missing commitSha is a known fallback path the UI already handles.
2. TaskChangesTab.tsx: when the API returns no files for a done task,
fall back to task.modifiedFiles (paths only, no patches) with a clear
note. Mirrors the existing 3-tier fallback in TaskCard.tsx:1090-1124
so card and modal always agree.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
pi-coding-agent's _getUserMessageText calls content.filter(...) on user
messages; if a message lands in state.messages with content === undefined
(string or array expected), the library throws
"Cannot read properties of undefined (reading 'filter')", which gets
caught and stored on session.state.errorMessage and rethrown without a
stack. Fusion's existing message-content guard already normalized
assistant/toolResult messages — extend it to user messages as well, and
sweep state.messages once when the guard is installed so content loaded
from a session file is repaired before the first event fires.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Add severity filter controls to the dashboard log viewer and apply filtering to rendered entries
- Extend dashboard-tui log streaming command to support severity filtering in output
- Update and stabilize dashboard and settings modal tests covering filter behavior and loading readiness
- Document the severity filter usage in CLI reference and add a changeset for @runfusion/fusion
The FN-2370 auto-resolved squash (7c1a1c36c) reverted three commits' worth
of refinements to the claude-cli provider because the branch contained
rebased duplicates of commits already on main. The auto-resolver picked
the older side line-by-line and dropped the newer.
Restored:
- /api/models filter logic (was inverted; emptied every model picker)
- Claude Opus 4.7 catalog entry in pi-claude-cli
- Provider card status text and toast messages (no longer claim a restart
is needed — the extension is always-loaded now)
- POST /api/auth/claude-cli returns restartRequired: false
Prevention:
- Regression tests on the /api/models useClaudeCli filter
- scripts/audit-squash-merge.mjs flags duplicate-cherry-pick risk and
touched-file overlap on any squash commit
- AGENTS.md documents the rebase-before-squash rule and requires the
merging agent to run the audit and triage every flagged item itself
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Resolve workflow step names by preferring lookup entries, then workflow result names, then raw step IDs
- Shorten task card workflow phase badges to concise pre-merge/post-merge labels while preserving tooltip context
- Expand TaskCard coverage for phase badge rendering and workflow name fallback behavior when lookup values are blank or missing
- Add semantic neutral surface tier tokens in root and light-theme blocks for subtle, muted, emphasis, and stronger hover states
- Replace Git Manager light-theme hardcoded neutral rgba backgrounds with surface tokens and token-based color-mix accent states
- Add dedicated Git Manager theme styling tests to guard token usage and prevent regressions to raw rgba values
- Expand status color/theme token tests and stabilize SettingsModal navigation selectors used by related test coverage
Add formatError() helper that extracts both message and full stack from
unknown caught values, and use it at every status:"failed" catch site in
executor, agent-heartbeat, and triage. Stack traces now land in
store.logEntry outcome (persisted to task.log/activityLog) and in stderr
logger output, so failures like "Cannot read properties of undefined
(reading 'filter')" can be diagnosed without re-running.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Codify the patterns that the recent dashboard test trim removed so
they don't grow back: CSS-class permutation tests, field-presence
tests that duplicate payload-roundtrip tests, React.memo tautologies,
and structural CSS assertions. Keep FN-linked regression tests,
integration tests, and lean core/engine units unconditionally.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
