Fix #3: Migrate dashboard UI to React 18 + Vite SPA

[rekpero]

Closes #3

Summary

• React 18 + Vite 5 frontend in frontend/ with Tailwind CSS 3, React Query v5, Lucide React, and date-fns
• Full component tree preserving the existing dark color palette:
  • Layout: Header, WorkspaceSwitcher, TabNav
  • Metrics: MetricsBar, MetricCard (7-card grid)
  • Agents: ActiveAgents, AgentCard, AgentLogViewer, AgentStatusBadge
  • Issues: IssueQueue, IssueStatusBadge with Retry button for needs_human
  • PRs: PRTracker, ReviewThreads
  • Modals: AddWorkspaceModal, WorkspaceSettingsModal, EnvEditor
  • UI primitives: Card, Badge, Button, Modal, Spinner, EmptyState
• WorkspaceContext persists selectedWorkspaceId to localStorage
• SPA catch-all route added to dashboard.py (registered after all /api/* routes and static mount) so React client-side navigation works on direct URL loads
• build-ui command added to run.sh (cd frontend && npm install && npm run build), integrated into the install flow
• Vite config sets build.outDir = '../orchestrator/static' and emptyOutDir = true — build artifacts go directly into orchestrator/static/
• .gitignore updated with frontend/node_modules/ and frontend/dist/

Test plan

• cd frontend && npm run dev — hot-reload dev server at http://localhost:5173 proxying /api to FastAPI
• npm run build — outputs orchestrator/static/index.html + assets/
• Start orchestrator with python -m orchestrator.main → open http://localhost:8420 — React app renders
• All 7 metric cards show correct values (or — when empty)
• Agent cards expand with log viewer, only fetching logs when expanded
• Issue queue shows Retry button for needs_human issues
• Add workspace modal creates a new workspace and selects it
• Workspace settings modal opens for selected workspace
• http://localhost:8420/ in a fresh tab loads the React app (SPA catch-all works)

🤖 Generated with Claude Code

[rekpero]

🔴 Claude BugBot Analysis

Found 6 potential bugs in this PR.

high: 3 | medium: 2 | low: 1

The PR introduces a React frontend dashboard with several concrete bugs: a log viewer that ignores its pagination cursor and re-fetches all events on every poll, a PR link pointing to the wrong URL, a settings form that can overwrite workspace data with empty strings, silent error swallowing in the env editor, a potential interval leak on agent status transitions, and a backend SPA fallback that can 500 when the frontend bundle is absent.

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

medium: 1 | low: 2

All six previously reported bugs have been fixed in this revision. Three new bugs were found: log events are replaced rather than accumulated on each poll (losing history), log polling runs unconditionally even for terminal agents, and workspace update API errors are silently discarded with no user feedback.

[rekpero]

🔴 Claude BugBot Analysis

Found 4 potential bugs in this PR.

medium: 2 | low: 2

All three previously reported bugs (log event accumulation, unconditional polling, and silent update errors) are now fixed. Four new bugs were found: completed-agent logs are never fetched due to enabled: isRunning fully disabling the query; the since cursor in the React Query key causes spurious loading states on each poll; the delete-workspace mutation silently ignores errors; and the updateError state is not cleared when switching workspaces.

[rekpero]

🔴 Claude BugBot Analysis

Found 1 potential bug in this PR.

low: 1

All four previously reported bugs are fixed. One new low-severity issue was introduced by the fix for bug PRRT_kwDORZvToM50UOIh: the setUpdateError(null) call inside useEffect([workspace]) can clear an active error message whenever the workspace's server data changes (not just when the user switches to a different workspace).

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

medium: 1 | low: 2

Three bugs found: a stale-closure issue in useAgentLogs where the since cursor is missing from the React Query key (causing the incremental log viewer to never fetch new events past the initial window), a file-upload handler that destructively replaces all env rows instead of merging, and a missing cache invalidation after saving env vars despite useQueryClient being called. The previously reported WorkspaceSettingsModal bug (effect clearing errors on any workspace refetch) is fixed — the dependency is now correctly [workspace?.id].

[rekpero]

🔴 Claude BugBot Analysis

Found 2 potential bugs in this PR.

medium: 1 | low: 1

Two new bugs found: PRTracker uses a non-unique React list key (pr_number alone) that can cause incorrect DOM reconciliation in multi-repo or all-workspace views; EnvEditor's post-save cache invalidation targets a query key that no hook is registered under, making it a no-op. All three previously reported bugs (missing since in queryKey, file-upload replacing rows instead of merging, and unused queryClient) have been fixed in this PR.

[rekpero]

🔴 Claude BugBot Analysis

Found 5 potential bugs in this PR.

high: 1 | medium: 1 | low: 3

Five bugs found across four files: a high-severity polling feedback loop caused by including a mutable cursor in a React Query key, a medium-severity fetch race condition from missing effect cleanup, and three low-severity issues (index-based React key, mismatched-quote stripping in env parsing, and a missing preventDefault on a hash link). Both previously reported open bugs are resolved: the PR list key is now a correct composite key, and the EnvEditor has been rewritten without the stale invalidateQueries call.

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

medium: 1 | low: 2

All five previously-reported bugs have been fixed. Three new issues were found: a stale closure in the log-polling hook means the since cursor is never advanced so every poll re-fetches all logs from the start; env-editor rows use array index as React key risking stale UI after row deletion; and the workspace settings form won't reinitialize if workspace data is updated while the modal is open.

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

medium: 1 | low: 2

All three previously reported bugs have been fixed in this revision (stale cursor closure via useRef, array-index React keys via crypto.randomUUID(), and incomplete useEffect dependency array). Three new bugs were found: the SPA catch-all route silently swallows unknown API 404s by returning HTML with HTTP 200, a shared mutation isPending flag incorrectly disables all Retry buttons in IssueQueue when any single update is in-flight, and the WorkspaceSettingsModal's confirmDelete state persists across open/close cycles causing the delete confirmation prompt to appear immediately on re-open.

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

medium: 2 | low: 1

The PR fixes the three previously reported bugs (SPA 200-on-API-path, shared Retry button loading state, and confirmDelete state leak) and introduces several new features. Three new bugs were found: a narrow edge case in the SPA guard where /api without a trailing slash bypasses the API-path check and serves HTML instead of a 404, a state accumulation issue where agent log entries from a prior agentId bleed into the newly selected agent's view, and a file input not being cleared after upload which prevents re-uploading the same file.

[rekpero]

🔴 Claude BugBot Analysis

Found 1 potential bug in this PR.

high: 1

All three previously reported bugs are fixed in this PR. One new high-severity bug is introduced: the SPA catch-all route intercepts browser requests for Vite-built assets at /assets/... (no static mount covers that path) and returns index.html instead of the JS/CSS bundles, breaking the frontend in production.

[rekpero]

🔴 Claude BugBot Analysis

Found 2 potential bugs in this PR.

high: 1 | medium: 1

The PR adds an /assets static mount and SPA catch-all to serve a Vite-built React frontend. Two bugs were found: the new /assets StaticFiles mount will crash the server on startup if the frontend hasn't been built, and a race condition in the agent-log cursor reset can cause the wrong cursor to be used on the first fetch when switching between agents. The previously-reported SPA catch-all/Vite asset interception bug has been fixed.

[rekpero]

🔴 Claude BugBot Analysis

Found 2 potential bugs in this PR.

medium: 2

Two new bugs were found: the root / route in the backend can 500 when the frontend is not built (unlike the new SPA fallback which gracefully returns 404), and the agent list pagination offset is not reset when the user switches workspaces. Both previously reported bugs are now fixed: the /assets StaticFiles mount no longer crashes the server at startup (check_dir=False was added), and the agent-log cursor is now reset synchronously during render (not in a racing useEffect) when agentId changes.

[rekpero]

🔴 Claude BugBot Analysis

Found 1 potential bug in this PR.

medium: 1

Both previously reported bugs (missing file-existence check in index() and missing offset reset on workspace switch) have been fixed. One new medium-severity bug was found: synchronous ref mutation during render in useAgents.js violates React's render-purity contract and can cause incorrect log-cursor behavior under React StrictMode, which is active in this application.

[rekpero]

🔴 Claude BugBot Analysis

Found 1 potential bug in this PR.

medium: 1

The StrictMode render-purity bug is fixed, but a new race condition remains: TanStack Query schedules its initial queryFn call as a microtask (before React's useEffect macrotasks run), so the cursor is not guaranteed to be 0 for the first fetch after an agent switch, contrary to the code comment. A related effect-ordering issue in AgentLogViewer can also allow stale cached data to overwrite the cursor reset when revisiting a previously-viewed agent.

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

medium: 2 | low: 1

All three previously reported bugs have been fixed. Three new bugs were found: a wrong-file content restoration on discard after switching files in EnvEditor, an unthrottled polling loop in usePlanning when the API returns null, and misleading agent IDs containing the literal string 'None' in the resume path of agent_pool.

[rekpero]

🔴 Claude BugBot Analysis

Found 4 potential bugs in this PR.

medium: 2 | low: 2

All three previously-reported bugs are confirmed fixed in this PR. Four new bugs were identified: a path traversal guard edge case in the dashboard, a potentially blocking os.waitpid call in the agent monitor thread, orphan subprocesses from process-group-unaware SIGTERM, and a race window in the polling re-entry guard.

[rekpero]

🔴 Claude BugBot Analysis

Found 2 potential bugs in this PR.

medium: 2

Two new medium-severity bugs introduced in the reattach/restart infrastructure: _monitor_pid uses os.kill instead of os.killpg when timing out a reattached agent (orphaning child processes), and _monitor_agent incorrectly fires the completion callback for externally-stopped agents. All four previously-reported bugs have been fixed in this PR.

[rekpero]

🔴 Claude BugBot Analysis

Found 1 potential bug in this PR.

medium: 1

Both previously-reported bugs (orphaned child processes on timeout kill, and _on_agent_complete firing for externally-stopped agents) are fixed. One new medium-severity bug remains: the newly-added _monitor_pid method completes its DB writes for reattached agents without ever invoking _on_agent_complete, so callback-driven downstream automation (e.g. dispatching a fix_review agent after a reattached implement agent creates a PR) never fires for agents that survived an orchestrator restart.

[rekpero]

🔴 Claude BugBot Analysis

Found 1 potential bug in this PR.

medium: 1

The previously reported _monitor_pid/_on_agent_complete bug is fixed via the new _on_reattach_complete callback mechanism. One new medium-severity bug was found: stopPolling() in usePlanning.js fails to reset pollActiveRef.current, which can permanently block poll() from rescheduling after a cancelGeneration() failure, leaving the UI frozen in a 'generating' state.

[rekpero]

🔴 Claude BugBot Analysis

Found 2 potential bugs in this PR.

medium: 2

The previously reported stopPolling/pollActiveRef bug is fixed. Two new medium-severity bugs exist in usePlanning.js: poll() silently exits without clearing generating state when its retry budget is exhausted (both in the null-data and catch paths), and resumeSession does not reset generating when session data fails to load, leaving the UI permanently stuck in a generating spinner in both scenarios.

[rekpero]

🔴 Claude BugBot Analysis

Found 2 potential bugs in this PR.

low: 1 | medium: 1

Both previously reported generating-spinner bugs in usePlanning.js are fixed. Two new bugs were found: a race condition in IssueQueue where a single retryingIssue string value causes a second concurrent retry's onSettled to clear the first row's spinner prematurely, and a logic error in agent_pool.py where failed reattached fix_review agents set the issue status to in_progress instead of needs_human/pending, permanently stranding those issues.

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

low: 3

Both previously reported bugs are fixed: IssueQueue now uses a Set for retryingIssues (concurrent retry spinners no longer interfere), and the reattached fix_review agent failure path now correctly sets status to needs_human. Three new low-severity bugs were found: a race condition in usePlanning.js that can create a duplicate polling loop, an ID mismatch in EnvEditor that unmounts/remounts inputs after save, and a silent scroll listener registration failure in PlannerModal on first open.

[rekpero]

🔴 Claude BugBot Analysis

Found 3 potential bugs in this PR.

medium: 2 | low: 1

All three previously reported bugs are fixed. Three new bugs were found: two in usePlanning.js where poll() cancels backoff timers on every recursive tick and sendMessage() lacks a stopPolling() guard, and one in ReviewThreads.jsx where custom sentinel tags can inadvertently match and alter user-authored comment content.

[rekpero]

🔴 Claude BugBot Analysis

Found 4 potential bugs in this PR.

low: 4

All three previously reported bugs (poll() backoff cancellation, sendMessage() missing stopPolling(), and sentinel tag collisions in formatCommentBody) have been fixed. Four new low-severity issues were found: an over-broad severity-prefix strip in formatCommentBody affecting all comment types, a stale event flush window in createIssue's poll interval, a React concurrent-mode-unsafe ref mutation during render in useAgents, and an unbounded module-level set in planner.py.

[rekpero]

🔴 Claude BugBot Analysis

Found 5 potential bugs in this PR.

medium: 2 | low: 3

The PR fixes all four previously reported bugs (formatCommentBody regex, createIssue cancelled-flag race, ref-mutation-in-render, and _issue_created_keys unbounded growth). Five new bugs were found: a file-unlink-while-open resource handling issue and a _stopped_agent_ids leak in error paths (both medium), plus three low-severity issues involving waitpid exit-code misreporting, log-tailer thread termination on transient DB errors, and a cursor-reset race in AgentLogViewer.

[rekpero]

🔴 Claude BugBot Analysis

Found 1 potential bug in this PR.

low: 1

All five previously-reported bugs have been fixed in this PR. One new bug was introduced: the new _tail_log method (for freshly-spawned agents) lacks the try/except guard around db.update_agent that was correctly applied to the parallel _tail_log_file method — a transient DB error will kill the tailer thread.

[rekpero]

🔴 Claude BugBot Analysis

Found 4 potential bugs in this PR.

medium: 3 | low: 1

The PR fixes the previously-reported unprotected db.update_agent bug in _tail_log (both calls are now wrapped in try/except), but introduces analogous unprotected db.insert_event calls in the same methods (_tail_log and _tail_log_file) that can still kill the tailer thread; additionally, a misleading WNOHANG comment in _monitor_pid and a race condition leaving old agent records stuck in 'running' state in restart_agent were found.

[rekpero]

🔴 Claude BugBot Analysis

Found 4 potential bugs in this PR.

medium: 1 | low: 3

All four previously reported bugs are fixed in this revision. Four new issues were found: a missing unmark_externally_stopped call in restart_agent when the dispatch call throws (medium), an incorrect unmark call when db.update_issue fails after finish_agent already committed (low), a TOCTOU deduplication race in planner.py between two separate lock acquisitions (low), and accumulating phantom blank rows in the EnvEditor on repeated paste/upload (low).

[rekpero]

🟢 Claude BugBot Analysis

All four previously reported bugs are fixed in this PR. The restart_agent endpoint now correctly separates finish_agent and update_issue into independent try/except blocks (fixing PRRT_kwDORZvToM50Ytgv), wraps each dispatch_* call in try/except that calls unmark_externally_stopped on exception (fixing PRRT_kwDORZvToM50Ytgu), the planner dedup check and insertion are now in the same lock block (fixing PRRT_kwDORZvToM50Ytgw), and EnvEditor retains at most one blank row during paste/upload (fixing PRRT_kwDORZvToM50Ytgx). No new bugs were identified in the added code.

No bugs were detected in this PR.