Merge pull request #17 from firstsun-dev/worktree-snuggly-riding-sedgewick

fix: dashboard bulk-analyze false failures + codex/copilot message-id collision

Author: ClaudiaFang

cli/src/providers/codex.ts

@@ -231,7 +231,7 @@ function parseFormatA(content: string): ParsedSession | null {
     } : null;
 
     messages.push({
-      id: `codex-assistant-${messages.length}`,
+      id: `${sessionId}:assistant-${messages.length}`,
       sessionId: sessionId,
       type: 'assistant',
       content: text.slice(0, 10000),
@@ -306,7 +306,7 @@ function parseFormatA(content: string): ParsedSession | null {
         const msgText = (payload.message as string) || '';
         if (msgText && !isSystemContextMessage(msgText)) {
           messages.push({
-            id: (payload.id as string) || `codex-user-${messages.length}`,
+            id: `${sessionId}:${(payload.id as string) || `user-${messages.length}`}`,
             sessionId: sessionId,
             type: 'user',
             content: msgText.slice(0, 10000),
@@ -522,7 +522,7 @@ function parseFormatB(content: string): ParsedSession | null {
     if (currentToolCalls.length === 0 && !currentThinking) return;
 
     messages.push({
-      id: `codex-assistant-${messages.length}`,
+      id: `${sessionId}:assistant-${messages.length}`,
       sessionId: sessionId,
       type: 'assistant',
       content: '',
@@ -548,7 +548,7 @@ function parseFormatB(content: string): ParsedSession | null {
       const userContent = extractFormatBContent(item.content);
       if (userContent && !isSystemContextMessage(userContent)) {
         messages.push({
-          id: `codex-user-${messages.length}`,
+          id: `${sessionId}:user-${messages.length}`,
           sessionId: sessionId,
           type: 'user',
           content: userContent.slice(0, 10000),

cli/src/providers/copilot-cli.ts

@@ -188,7 +188,7 @@ function parseCopilotSession(filePath: string): ParsedSession | null {
       if (!text && currentToolCalls.length === 0) return;
 
       messages.push({
-        id: `copilot-assistant-${messages.length}`,
+        id: `${sessionId}:assistant-${messages.length}`,
         sessionId: sessionId,
         type: 'assistant',
         content: text.slice(0, 10000),
@@ -251,7 +251,7 @@ function parseCopilotSession(filePath: string): ParsedSession | null {
           const userContent = extractText(data);
           if (userContent) {
             messages.push({
-              id: (data.id as string) || `copilot-user-${messages.length}`,
+              id: `${sessionId}:${(data.id as string) || `user-${messages.length}`}`,
               sessionId: sessionId,
               type: 'user',
               content: userContent.slice(0, 10000),

dashboard/src/components/sessions/SessionListPanel.tsx

@@ -21,6 +21,7 @@ import { extractPQScore } from '@/lib/score-utils';
 import { SearchX, Terminal, EyeOff, CalendarDays } from 'lucide-react';
 import { useDeletedSessionCount } from '@/hooks/useSessions';
 import { useQueuedSessionIds } from '@/hooks/useAnalysisQueue';
+import { useAnalyzedSessionIds } from '@/hooks/useAnalyzedSessionIds';
 import { SaveFilterPopover } from '@/components/filters/SaveFilterPopover';
 import { SavedFiltersDropdown } from '@/components/filters/SavedFiltersDropdown';
 import { SourceToolSelect } from '@/components/filters/SourceToolSelect';
@@ -100,10 +101,11 @@ export function SessionListPanel({
 
   const { data: deletedCount = 0 } = useDeletedSessionCount(projectId);
   const queuedSessionIds = useQueuedSessionIds();
-  const analyzedSessionIds = useMemo(
-    () => new Set(insights.map((i) => i.session_id)),
-    [insights]
-  );
+  // Sourced from analysis_usage, not `insights` — insights has no safe row cap to
+  // rely on for "is this session analyzed" at scale (a single session's analysis
+  // produces 5-10+ insight rows), so a capped insights query would silently
+  // misclassify already-analyzed sessions as unanalyzed on large histories.
+  const { data: analyzedSessionIds = new Set<string>() } = useAnalyzedSessionIds();
 
   const insightCountsBySession = useMemo(() => {
     const map = new Map<string, Record<string, number>>();

dashboard/src/hooks/useAnalyzedSessionIds.ts

@@ -0,0 +1,15 @@
+import { useQuery } from '@tanstack/react-query';
+import { fetchAnalyzedSessionIds } from '@/lib/api';
+
+/**
+ * Session IDs with a completed session analysis, sourced from analysis_usage
+ * (one row per session) rather than the insights table — insights has no row
+ * cap safe to rely on for "is this session analyzed" checks at scale, since a
+ * single session can produce 5-10+ insight rows.
+ */
+export function useAnalyzedSessionIds() {
+  return useQuery({
+    queryKey: ['analyzedSessionIds'],
+    queryFn: () => fetchAnalyzedSessionIds().then((r) => new Set(r.sessionIds)),
+  });
+}

dashboard/src/lib/api.ts

@@ -552,6 +552,12 @@ export function fetchAnalysisQueue() {
   return request<AnalysisQueueStatus>('/analysis/queue');
 }
 
+// ── Analyzed session IDs ───────────────────────────────────────────────────────
+
+export function fetchAnalyzedSessionIds() {
+  return request<{ sessionIds: string[] }>('/analysis/analyzed-session-ids');
+}
+
 // ── Facets ─────────────────────────────────────────────────────────────────────
 
 export interface FacetRow {

dashboard/src/pages/DashboardPage.tsx

@@ -3,6 +3,7 @@ import { Link } from 'react-router';
 import { useDashboardStats, useDailyStats } from '@/hooks/useAnalytics';
 import { useSessions } from '@/hooks/useSessions';
 import { useInsights } from '@/hooks/useInsights';
+import { useAnalyzedSessionIds } from '@/hooks/useAnalyzedSessionIds';
 import { useProjects } from '@/hooks/useProjects';
 import { StatsHero } from '@/components/dashboard/StatsHero';
 import { DashboardActivityChart } from '@/components/dashboard/DashboardActivityChart';
@@ -34,19 +35,24 @@ export default function DashboardPage() {
   const { data: dailyStats = [], isLoading: dailyLoading, isError: dailyError, refetch: refetchDaily } = useDailyStats(range, effectiveHomeId);
   const { data: sessions = [], isLoading: sessionsLoading, isError: sessionsError, refetch: refetchSessions } = useSessions({ limit: 500, ...(homeId !== 'all' && { homeId }) });
   const { data: insights = [], isLoading: insightsLoading } = useInsights();
+  const { data: analyzedSessionIds, isLoading: analyzedIdsLoading } = useAnalyzedSessionIds();
   const { data: projects = [] } = useProjects();
 
-  const loading = statsLoading || sessionsLoading || insightsLoading || dailyLoading;
+  const loading = statsLoading || sessionsLoading || insightsLoading || dailyLoading || analyzedIdsLoading;
   const hasError = statsError || sessionsError || dailyError;
 
   const todayLabel = new Date().toLocaleDateString(undefined, {
     month: 'long',
     day: 'numeric',
   });
 
-  // Sessions not yet analyzed
-  const analyzedSessionIds = new Set(insights.map((i) => i.session_id));
-  const unanalyzedSessions = sessions.filter((s) => !analyzedSessionIds.has(s.id));
+  // Sessions not yet analyzed. Sourced from analysis_usage (via useAnalyzedSessionIds),
+  // not the insights list — insights has no safe row cap to rely on at scale (a single
+  // session's analysis produces 5-10+ insight rows), so a capped insights query would
+  // silently misclassify already-analyzed sessions as unanalyzed on large histories.
+  const unanalyzedSessions = analyzedSessionIds
+    ? sessions.filter((s) => !analyzedSessionIds.has(s.id))
+    : [];
 
   // Compute stats for hero — all from dashStats (range-filtered)
   const totalTokens = dashStats

server/src/routes/analysis.ts

@@ -11,6 +11,7 @@ import {
   AnalysisResultSchema,
   AnalysisUsageResponseSchema,
   AnalysisUsageQuerySchema,
+  AnalyzedSessionIdsResponseSchema,
   RecurringInsightResultSchema,
 } from '../schemas/analysis.js';
 import {
@@ -100,6 +101,32 @@ app.openapi(usageRoute, (c) => {
   }, 200);
 });
 
+// GET /api/analysis/analyzed-session-ids
+// Returns every session_id that has a completed 'session' analysis in analysis_usage.
+// Used by the dashboard to determine which sessions still need analysis — deliberately
+// sourced from analysis_usage (one row per session, PRIMARY KEY (session_id, analysis_type))
+// rather than the insights table, which has no such bound: a single session's analysis
+// produces 5-10+ insight rows, so a capped/paginated insights query silently truncates
+// on large histories and misclassifies already-analyzed sessions as unanalyzed.
+const analyzedSessionIdsRoute = createRoute({
+  method: 'get',
+  path: '/analyzed-session-ids',
+  responses: {
+    200: {
+      content: { 'application/json': { schema: AnalyzedSessionIdsResponseSchema } },
+      description: 'Session IDs with a completed session analysis',
+    },
+  },
+});
+
+app.openapi(analyzedSessionIdsRoute, (c) => {
+  const db = getDb();
+  const rows = db.prepare(
+    `SELECT session_id FROM analysis_usage WHERE analysis_type = 'session'`
+  ).all() as Array<{ session_id: string }>;
+  return c.json({ sessionIds: rows.map((r) => r.session_id) }, 200);
+});
+
 // POST /api/analysis/session
 // Body: { sessionId: string }
 // Fetches session + messages from SQLite, runs LLM analysis, saves insights, returns results.

server/src/schemas/analysis.ts

@@ -83,6 +83,12 @@ export const AnalysisUsageQuerySchema = z.object({
 
 export const SessionIdBodyResponseSchema = AnalysisResultSchema;
 
+export const AnalyzedSessionIdsResponseSchema = z
+  .object({
+    sessionIds: z.array(z.string()),
+  })
+  .openapi('AnalyzedSessionIdsResponse');
+
 /** Mirrors server/src/llm/recurring-insights.ts RecurringInsightResult. */
 export const RecurringInsightResultSchema = z
   .object({