@bahulam/code 0.1.6 → 0.1.8

package/src/ui/slash-commands.mjs

@@ -52,12 +52,12 @@ export const COMMANDS = {
   '/review':      'Code review agent',
   '/architect':   'Feature architect agent',
   '/safety':      'Show safety guardrail status',
-  '/auto':        'Session autopilot: auto-approve routine tools (dangerous still prompts)',
-  '/approvals':   'List or edit session approval grants',
-  '/revoke':      'Revoke auto-approvals',
+  '/auto':        'Session autopilot: `/auto` routine-only · `/auto full` everything · `/auto off`',
+  '/approvals':   'List or edit session approval grants (`/approvals clear` revokes all)',
   '/resume':      'Resume a previous session',
   '/sessions':    'List resumable sessions',
   '/logout':      'Sign out and clear credentials',
+  '/watch':       'Toggle live agent activity tracking panel',
   '/exit':        'Exit CLI',
 };
 
@@ -118,7 +118,6 @@ export const HELP_GROUPS = [
       ['/settings logout', 'Sign out'],
       ['/settings whoami', 'Current user'],
       ['/settings quiet|verbose|surgical', 'Verbosity'],
-      ['/settings revoke', 'Revoke auto-approvals'],
     ],
   },
   {
@@ -138,6 +137,7 @@ export const HELP_GROUPS = [
     title: 'Agents',
     summary: 'specialist modes',
     commands: [
+      ['/watch', 'Toggle live agent tracking panel'],
       ['/agents', 'List built-in and local agents'],
       ['/agents create <name>', 'Create .bahulam/agents/<name>.yaml'],
       ['/agents edit <name>', 'Open local agent YAML'],
@@ -207,7 +207,6 @@ export const LEGACY_COMMAND_HINTS = {
   '/quiet':      '/settings quiet',
   '/verbose':    '/settings verbose',
   '/surgical':   '/settings surgical',
-  '/revoke':     '/settings revoke',
 };
 
 // Namespaced subcommands. `/status metrics` -> `/stats`, etc.
@@ -235,7 +234,6 @@ export const NAMESPACED_COMMANDS = {
     quiet:    '/quiet',
     verbose:  '/verbose',
     surgical: '/surgical',
-    revoke:   '/revoke',
   },
 };
 

package/src/ui/spinner.mjs

@@ -16,7 +16,7 @@
  *
  * Behavior:
  *   - 120ms per frame.
- *   - Suppressed entirely when stdout is not a TTY or when KEPLER_PLAIN=1.
+ *   - Suppressed entirely when stdout is not a TTY or when BAHULAM_PLAIN=1.
  *     The `start`/`stop` API is still safe to call (no-op).
  *   - Color follows the current orbit (orchestrated by the caller via the
  *     palette token). Default: brand.primary.

package/src/ui/term.mjs

@@ -34,9 +34,7 @@ function readEnv() {
     NO_COLOR: env.NO_COLOR,
     FORCE_COLOR: env.FORCE_COLOR,
     BAHULAM_PLAIN: env.BAHULAM_PLAIN,
-    KEPLER_PLAIN: env.KEPLER_PLAIN,
     BAHULAM_TTY_MODE: env.BAHULAM_TTY_MODE,
-    KEPLER_TTY_MODE: env.KEPLER_TTY_MODE,
     COLORTERM: (env.COLORTERM || '').toLowerCase(),
     TERM: (env.TERM || '').toLowerCase(),
     TERM_PROGRAM: env.TERM_PROGRAM || '',
@@ -48,7 +46,6 @@ function detectColorLevel(env, isTTY) {
   // Hard opt-out (https://no-color.org). Honored even on TTYs.
   if (env.NO_COLOR !== undefined && env.NO_COLOR !== '') return 'none';
   if (env.BAHULAM_PLAIN === '1') return 'none';
-  if (env.KEPLER_PLAIN === '1') return 'none';
 
   // Hard opt-in. FORCE_COLOR=1|2|3 maps to ansi16|ansi256|truecolor.
   // FORCE_COLOR with no value or =true falls through to detection.
@@ -84,7 +81,6 @@ function detectColorLevel(env, isTTY) {
 
 function detectUnicode(env) {
   if (env.BAHULAM_PLAIN === '1') return false;
-  if (env.KEPLER_PLAIN === '1') return false;
   // Most modern terminals on macOS/Linux handle UTF-8.
   // Windows ConEmu / older terminals are the main holdouts; conservative
   // fallback when LANG and LC_* are missing or explicitly POSIX.
@@ -95,8 +91,8 @@ function detectUnicode(env) {
 }
 
 function detectTtyMode(env) {
-  const raw = String(env.BAHULAM_TTY_MODE || env.KEPLER_TTY_MODE || '').trim().toLowerCase();
-  if (env.BAHULAM_PLAIN === '1' || env.KEPLER_PLAIN === '1') return 'plain';
+  const raw = String(env.BAHULAM_TTY_MODE || '').trim().toLowerCase();
+  if (env.BAHULAM_PLAIN === '1') return 'plain';
   if (raw === 'stable' || raw === 'simple' || raw === 'static' || raw === 'transcript') return 'stable';
   return 'rich';
 }

package/src/ui/tool-card.mjs

@@ -78,12 +78,15 @@ const paintShellAdapter = {
  *
  * @returns {{ text: string, tone: 'success'|'warn'|'danger'|'dim' }}
  */
-export function summarizeResult(tool, data) {
+export function summarizeResult(tool, data, args = {}) {
   if (!data) return { text: '', tone: 'dim' };
 
   if (data._blocked) {
     return { text: firstOutputLine(data) || 'blocked', tone: 'danger' };
   }
+  if (isNoChangeEditResult(tool, data)) {
+    return { text: 'no changes', tone: 'warn' };
+  }
   if (data._observation_timeout) {
     const ms = data._observation_timeout_ms;
     const duration = typeof ms === 'number' ? formatDuration(ms) : '';
@@ -111,7 +114,8 @@ export function summarizeResult(tool, data) {
 
   switch (tool) {
     case 'read_file': {
-      const lines = data._total_lines || lineCount(data.output || data.output_preview);
+      const lines = readFileLineCount(data, args);
+      if (lines == null) return { text: 'read', tone: 'success' };
       return { text: `${lines} line${lines === 1 ? '' : 's'}`, tone: 'success' };
     }
     case 'read_files':
@@ -136,6 +140,9 @@ export function summarizeResult(tool, data) {
     case 'write_file':
     case 'write_project': {
       const delta = diffDelta(data);
+      if (tool === 'edit_file' && delta === '+0 −0') {
+        return { text: 'no changes', tone: 'warn' };
+      }
       if (delta) return { text: delta, tone: 'success' };
       return { text: 'updated', tone: 'success' };
     }
@@ -159,7 +166,7 @@ export function summarizeResult(tool, data) {
         // Multi-row preview: first N rows + a "+ M more" tail so long
         // outputs (e.g. `ls`) surface their scale instead of collapsing
         // to a single first line with no hint that more exists.
-        const { preview, remaining } = outputPreviewRows(data, shellPreviewRows());
+        const { preview, remaining } = outputPreviewRows(data, shellPreviewRows(data));
         if (!preview) return { text: 'ok', tone: 'success' };
         if (remaining === 0) return { text: preview, tone: 'success' };
         const tail = paint.text.dim(`+ ${remaining} more row${remaining === 1 ? '' : 's'}`);
@@ -434,9 +441,13 @@ function outputPreviewRows(data, n, perRow = 200) {
 // Default rows shown in the shell result preview. Overridable via env for
 // power users; keep it small — the result line rides above every command
 // and eats vertical space in a long session.
-function shellPreviewRows() {
+function shellPreviewRows(data = {}) {
   const raw = parseInt(process.env.BAHULAM_SHELL_PREVIEW_ROWS ?? '', 10);
-  return Number.isFinite(raw) && raw >= 1 ? raw : 2;
+  if (Number.isFinite(raw) && raw >= 1) return raw;
+  const command = String(data?.args?.command || data?.command || '');
+  if (/^\s*git\s+(diff|show)\b/i.test(command)) return 8;
+  if (/^\s*git\s+status\b/i.test(command)) return 8;
+  return 2;
 }
 
 function lineCount(s) {
@@ -444,6 +455,64 @@ function lineCount(s) {
   return String(s).split('\n').filter(Boolean).length;
 }
 
+function physicalLineCount(text) {
+  const value = String(text ?? '').replace(/\r\n?/g, '\n');
+  if (!value) return 0;
+  const lines = value.split('\n');
+  if (lines[lines.length - 1] === '') lines.pop();
+  return lines.length;
+}
+
+function readFileLineCount(data = {}, args = {}) {
+  const explicit = explicitReadLineCount(data);
+  if (explicit != null) return explicit;
+
+  const text = firstReadTextPayload(data);
+  if (text != null) return physicalLineCount(text);
+
+  const range = requestedRangeLineCount(args || data?.args || {});
+  if (range != null) return range;
+
+  return null;
+}
+
+function explicitReadLineCount(data = {}) {
+  if (!data || typeof data !== 'object') return null;
+  for (const key of ['_total_lines', 'line_count', 'lines_count', 'total_lines', 'lineCount', 'totalLines']) {
+    const value = data[key];
+    if (value == null || value === '') continue;
+    const number = Number(value);
+    if (Number.isFinite(number) && number >= 0) return Math.floor(number);
+  }
+  if (Array.isArray(data.lines)) return data.lines.length;
+  if (data.result && typeof data.result === 'object' && data.result !== data) {
+    return explicitReadLineCount(data.result);
+  }
+  return null;
+}
+
+function firstReadTextPayload(data = {}) {
+  if (!data || typeof data !== 'object') return null;
+  for (const key of ['content', 'text', 'output', 'output_preview', 'message']) {
+    if (typeof data[key] === 'string') return data[key];
+  }
+  if (data.result && typeof data.result === 'object' && data.result !== data) {
+    return firstReadTextPayload(data.result);
+  }
+  return null;
+}
+
+function requestedRangeLineCount(args = {}) {
+  const start = Number(args.start_line ?? args.startLine);
+  const end = Number(args.end_line ?? args.endLine);
+  if (Number.isFinite(start) && Number.isFinite(end) && start > 0 && end >= start) {
+    return Math.floor(end - start + 1);
+  }
+  const limit = Number(args.limit ?? args.max_lines ?? args.maxLines);
+  if (Number.isFinite(limit) && limit >= 0) return Math.floor(limit);
+  return null;
+}
+
 function countMatches(data) {
   if (typeof data?.match_count === 'number') return data.match_count;
   return lineCount(data?.output);
@@ -470,6 +539,12 @@ function diffDelta(data) {
   return `+${a} −${r}`;
 }
 
+function isNoChangeEditResult(tool, data = {}) {
+  if (tool !== 'edit_file') return false;
+  if (data._no_change || data.no_change) return true;
+  return data.success !== false && data.lines_added === 0 && data.lines_removed === 0;
+}
+
 function tone(text, t) {
   switch (t) {
     case 'success': return paint.state.success(text);
@@ -573,7 +648,7 @@ export function formatCard({ tool, args, result, durationMs, indent, columns, cw
   const cols = columns || term().columns || 120;
   const head = formatCardHead(tool, args, { indent, columns: cols, cwd });
 
-  const summary = summarizeResult(tool, result);
+  const summary = summarizeResult(tool, result, args);
   const duration = formatDuration(durationMs ?? result?.duration_ms ?? (result?.duration_s != null ? result.duration_s * 1000 : null));
 
   if (!summary.text && !duration) return head;

package/pulse/app/activity/page.tsx

@@ -1,190 +0,0 @@
-'use client'
-
-import useSWR from 'swr'
-import { TopBar } from '@/components/layout/top-bar'
-import { ActivityHeatmap } from '@/components/overview/activity-heatmap'
-import { PeakHoursChart } from '@/components/overview/peak-hours-chart'
-import { DayOfWeekChart } from '@/components/activity/day-of-week-chart'
-import { UsageOverTimeChart } from '@/components/overview/usage-over-time-chart'
-import type { DailyActivity } from '@/types/claude'
-import { Card, CardContent, CardDescription, CardHeader, CardTitle } from '@/components/ui/card'
-import { Skeleton } from '@/components/ui/skeleton'
-import { Alert, AlertDescription } from '@/components/ui/alert'
-import { AlertTriangle, Flame, CalendarDays, BarChart3, Clock, Zap, TrendingUp, Star } from 'lucide-react'
-
-const fetcher = (url: string) =>
-  fetch(url).then(r => { if (!r.ok) throw new Error(`API error ${r.status}`); return r.json() })
-
-interface ActivityData {
-  daily_activity: DailyActivity[]
-  hour_counts: Array<{ hour: number; count: number }>
-  dow_counts: Array<{ day: string; count: number }>
-  streaks: { current: number; longest: number }
-  most_active_day: string
-  most_active_day_msgs: number
-  total_active_days: number
-}
-
-function StatTile({
-  label,
-  value,
-  sub,
-  icon: Icon,
-  color,
-}: {
-  label: string
-  value: string | number
-  sub: string
-  icon: React.ElementType
-  color: string
-}) {
-  return (
-    <Card className="gap-0">
-      <CardHeader className="pb-2">
-        <CardDescription className="flex items-center gap-2 text-sm font-medium">
-          <Icon className="h-4 w-4 shrink-0 text-muted-foreground" />
-          {label}
-        </CardDescription>
-        <CardTitle className="text-3xl font-bold tabular-nums leading-none" style={{ color }}>
-          {value}
-        </CardTitle>
-      </CardHeader>
-      <CardContent className="pt-0">
-        <p className="text-xs text-muted-foreground">{sub}</p>
-      </CardContent>
-    </Card>
-  )
-}
-
-export default function ActivityPage() {
-  const { data, error, isLoading } = useSWR<ActivityData>('/api/activity', fetcher, { refreshInterval: 5_000 })
-
-  const hourCounts = data
-    ? Object.fromEntries(data.hour_counts.map(h => [String(h.hour), h.count]))
-    : {}
-
-  return (
-    <div className="flex flex-col min-h-screen">
-      <TopBar title="Activity" subtitle="Patterns, streaks, and peak hours" />
-      <div className="w-full space-y-4 p-4 md:p-6">
-
-        {error && (
-          <Alert variant="destructive">
-            <AlertTriangle className="h-4 w-4" />
-            <AlertDescription>Error loading data: {String(error)}</AlertDescription>
-          </Alert>
-        )}
-
-        {isLoading && (
-          <>
-            <div className="grid grid-cols-2 md:grid-cols-4 gap-4">
-              {Array.from({ length: 4 }).map((_, i) => <Skeleton key={i} className="h-28 rounded-xl" />)}
-            </div>
-            <div className="grid grid-cols-1 gap-4 md:grid-cols-2 md:items-stretch">
-              <Skeleton className="h-72 min-h-72 rounded-xl md:min-h-80" />
-              <Skeleton className="h-56 min-h-56 rounded-xl md:min-h-80" />
-            </div>
-            <div className="grid grid-cols-1 gap-4 md:grid-cols-2 md:items-stretch">
-              <Skeleton className="h-64 rounded-xl md:min-h-72" />
-              <Skeleton className="h-56 rounded-xl md:min-h-72" />
-            </div>
-          </>
-        )}
-
-        {data && (
-          <>
-            {/* Stat tiles */}
-            <div className="grid grid-cols-2 md:grid-cols-4 gap-4">
-              <StatTile
-                label="Current Streak"
-                value={data.streaks.current}
-                sub="consecutive days"
-                icon={Flame}
-                color="#f97316"
-              />
-              <StatTile
-                label="Longest Streak"
-                value={data.streaks.longest}
-                sub="personal best"
-                icon={Zap}
-                color="var(--viz-sky)"
-              />
-              <StatTile
-                label="Active Days"
-                value={data.total_active_days}
-                sub="total days with activity"
-                icon={TrendingUp}
-                color="#a78bfa"
-              />
-              <StatTile
-                label="Most Active Day"
-                value={data.most_active_day ? data.most_active_day.slice(5) : '—'}
-                sub={data.most_active_day_msgs ? `${data.most_active_day_msgs.toLocaleString()} messages` : 'no data'}
-                icon={Star}
-                color="#34d399"
-              />
-            </div>
-
-            {/* Row 1: Activity calendar | Peak hours */}
-            <div className="grid grid-cols-1 gap-4 md:grid-cols-2 md:items-stretch">
-              <Card className="flex h-full min-h-0 min-w-0 flex-col gap-2 py-4">
-                <CardHeader className="space-y-1 pb-0">
-                  <CardTitle className="flex items-center gap-2 text-base font-semibold">
-                    <CalendarDays className="h-4 w-4 shrink-0 text-muted-foreground" />
-                    Activity Calendar
-                  </CardTitle>
-                  <CardDescription>GitHub-style contribution heatmap</CardDescription>
-                </CardHeader>
-                <CardContent className="flex flex-1 flex-col pt-3 pb-0">
-                  <ActivityHeatmap data={data.daily_activity} />
-                </CardContent>
-              </Card>
-
-              <Card className="flex h-full min-h-0 min-w-0 flex-col gap-2 py-4">
-                <CardHeader className="space-y-1 pb-0">
-                  <CardTitle className="flex items-center gap-2 text-base font-semibold">
-                    <Clock className="h-4 w-4 shrink-0 text-muted-foreground" />
-                    Peak Hours
-                  </CardTitle>
-                  <CardDescription>Activity by hour of day</CardDescription>
-                </CardHeader>
-                <CardContent className="flex flex-1 flex-col justify-between pt-3 pb-0">
-                  <PeakHoursChart hourCounts={hourCounts} />
-                </CardContent>
-              </Card>
-            </div>
-
-            {/* Row 2: Usage over time | Day of week */}
-            <div className="grid grid-cols-1 gap-4 md:grid-cols-2 md:items-stretch">
-              <Card className="flex h-full min-h-0 min-w-0 flex-col gap-2 py-4">
-                <CardHeader className="space-y-1 pb-0">
-                  <CardTitle className="flex items-center gap-2 text-base font-semibold">
-                    <BarChart3 className="h-4 w-4 shrink-0 text-muted-foreground" />
-                    Usage Over Time
-                  </CardTitle>
-                  <CardDescription>Messages and sessions over the last 90 days</CardDescription>
-                </CardHeader>
-                <CardContent className="flex flex-1 flex-col pt-3 pb-0">
-                  <UsageOverTimeChart data={data.daily_activity} days={90} />
-                </CardContent>
-              </Card>
-
-              <Card className="flex h-full min-h-0 min-w-0 flex-col gap-2 py-4">
-                <CardHeader className="space-y-1 pb-0">
-                  <CardTitle className="flex items-center gap-2 text-base font-semibold">
-                    <CalendarDays className="h-4 w-4 shrink-0 text-muted-foreground" />
-                    Day of Week
-                  </CardTitle>
-                  <CardDescription>Which days you use Bahulam Code most</CardDescription>
-                </CardHeader>
-                <CardContent className="flex flex-1 flex-col justify-between pt-3 pb-0">
-                  <DayOfWeekChart data={data.dow_counts} />
-                </CardContent>
-              </Card>
-            </div>
-          </>
-        )}
-      </div>
-    </div>
-  )
-}

package/pulse/app/api/activity/route.ts

@@ -1,138 +0,0 @@
-import { NextResponse } from 'next/server'
-import { getSessions, readStatsCache } from '@/lib/claude-reader'
-import type { DailyActivity, SessionMeta } from '@/types/claude'
-
-export const dynamic = 'force-dynamic'
-
-function computeStreaks(dates: Set<string>): { current: number; longest: number } {
-  const sorted = [...dates].sort()
-  if (sorted.length === 0) return { current: 0, longest: 0 }
-
-  let longest = 1
-  let streak = 1
-  for (let i = 1; i < sorted.length; i++) {
-    const prev = new Date(sorted[i - 1])
-    const curr = new Date(sorted[i])
-    const diff = (curr.getTime() - prev.getTime()) / 86_400_000
-    if (diff === 1) {
-      streak++
-      if (streak > longest) longest = streak
-    } else {
-      streak = 1
-    }
-  }
-
-  // Current streak (from today backwards)
-  const today = new Date().toISOString().slice(0, 10)
-  let current = 0
-  const d = new Date(today)
-  while (dates.has(d.toISOString().slice(0, 10))) {
-    current++
-    d.setDate(d.getDate() - 1)
-  }
-
-  return { current, longest }
-}
-
-function computeDailyActivityFromSessions(sessions: SessionMeta[]): DailyActivity[] {
-  const byDate = new Map<string, { messages: number; sessions: number; tools: number }>()
-
-  for (const s of sessions) {
-    if (!s.start_time) continue
-    const date = s.start_time.slice(0, 10)
-    if (!/^\d{4}-\d{2}-\d{2}$/.test(date)) continue
-
-    const existing = byDate.get(date) ?? { messages: 0, sessions: 0, tools: 0 }
-    existing.messages += (s.user_message_count ?? 0) + (s.assistant_message_count ?? 0)
-    existing.sessions += 1
-    existing.tools += Object.values(s.tool_counts ?? {}).reduce((sum, count) => sum + count, 0)
-    byDate.set(date, existing)
-  }
-
-  return Array.from(byDate.entries())
-    .map(([date, { messages, sessions: sessionCount, tools }]) => ({
-      date,
-      messageCount: messages,
-      sessionCount,
-      toolCallCount: tools,
-    }))
-    .sort((a, b) => a.date.localeCompare(b.date))
-}
-
-function mergeDailyActivity(fromStats: DailyActivity[], fromSessions: DailyActivity[]): DailyActivity[] {
-  const byDate = new Map<string, DailyActivity>()
-  for (const d of fromStats) byDate.set(d.date, d)
-  for (const d of fromSessions) byDate.set(d.date, d)
-  return Array.from(byDate.values()).sort((a, b) => a.date.localeCompare(b.date))
-}
-
-function computeHourCounts(sessions: SessionMeta[]): Array<{ hour: number; count: number }> {
-  const hourCounts = Array.from({ length: 24 }, () => 0)
-
-  for (const s of sessions) {
-    if (s.user_message_timestamps?.length) {
-      for (const ts of s.user_message_timestamps) {
-        const d = new Date(ts)
-        if (!isNaN(d.getTime())) hourCounts[d.getHours()]++
-      }
-      continue
-    }
-
-    if (s.message_hours?.length) {
-      for (const hour of s.message_hours) {
-        if (Number.isInteger(hour) && hour >= 0 && hour < 24) hourCounts[hour]++
-      }
-      continue
-    }
-
-    const d = new Date(s.start_time)
-    if (!isNaN(d.getTime())) hourCounts[d.getHours()]++
-  }
-
-  return hourCounts.map((count, hour) => ({ hour, count }))
-}
-
-export async function GET() {
-  const [stats, sessions] = await Promise.all([readStatsCache(), getSessions()])
-  const dailyFromSessions = computeDailyActivityFromSessions(sessions)
-  const dailyActivity = stats
-    ? mergeDailyActivity(stats.dailyActivity ?? [], dailyFromSessions)
-    : dailyFromSessions
-
-  // Day-of-week counts from session timestamps
-  const dowCounts: number[] = [0, 0, 0, 0, 0, 0, 0] // Sun=0..Sat=6
-  const activeDates = new Set<string>()
-
-  for (const s of sessions) {
-    if (!s.start_time) continue
-    const d = new Date(s.start_time)
-    if (isNaN(d.getTime())) continue
-    dowCounts[d.getDay()]++
-    activeDates.add(s.start_time.slice(0, 10))
-  }
-
-  const streaks = computeStreaks(activeDates)
-
-  // Most active day
-  let mostActiveDay = ''
-  let mostActiveMsgs = 0
-  for (const da of dailyActivity) {
-    if (da.messageCount > mostActiveMsgs) {
-      mostActiveMsgs = da.messageCount
-      mostActiveDay = da.date
-    }
-  }
-
-  return NextResponse.json({
-    daily_activity: dailyActivity,
-    hour_counts: computeHourCounts(sessions),
-    dow_counts: dowCounts.map((count, i) => ({
-      day: ['Sun', 'Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat'][i],
-      count,
-    })),
-    streaks,
-    most_active_day: mostActiveDay,
-    most_active_day_msgs: mostActiveMsgs,
-    total_active_days: activeDates.size,
-  })
-}

package/pulse/app/api/benchmark/route.ts

@@ -1,113 +0,0 @@
-import { NextResponse } from 'next/server'
-import { readFileSync } from 'fs'
-import { join } from 'path'
-
-export const dynamic = 'force-dynamic'
-
-interface BenchmarkResult {
-  instance_id: string
-  repo: string
-  base_commit: string
-  test_patch: string
-  resolved: boolean
-  test_result: {
-    result: string[]
-    exit_code: number
-  }
-  metadata: {
-    agent_class: string
-    model_name: string
-    max_iterations: number
-    eval_history: Array<{
-      timestamp: string
-      action: string
-      observation: string
-    }>
-    submission: string
-    instance_id: string
-    predict_output: string
-    model_patch: string
-    test_result: {
-      result: string[]
-      exit_code: number
-    }
-  }
-}
-
-interface BenchmarkData {
-  results: BenchmarkResult[]
-}
-
-export async function GET(request: Request) {
-  try {
-    const { searchParams } = new URL(request.url)
-    const run = searchParams.get('run') || 'swebench-v4-flash-300'
-    const limit = parseInt(searchParams.get('limit') || '50', 10)
-    const offset = parseInt(searchParams.get('offset') || '0', 10)
-
-    // Load benchmark results from file
-    const resultsPath = join(
-      process.cwd(),
-      '..',
-      'benchmark',
-      'results',
-      'runs',
-      run,
-      'harness-results.json'
-    )
-
-    let data: BenchmarkData
-    try {
-      const fileContent = readFileSync(resultsPath, 'utf-8')
-      data = JSON.parse(fileContent)
-    } catch (error) {
-      return NextResponse.json(
-        { error: `Benchmark run "${run}" not found` },
-        { status: 404 }
-      )
-    }
-
-    // Calculate statistics
-    const results = data.results || []
-    const totalTests = results.length
-    const resolvedTests = results.filter((r) => r.resolved).length
-    const passRate = totalTests > 0 ? (resolvedTests / totalTests) * 100 : 0
-
-    // Group by repo
-    const byRepo = new Map<string, number>()
-    const byRepoResolved = new Map<string, number>()
-    for (const result of results) {
-      const repo = result.repo || 'unknown'
-      byRepo.set(repo, (byRepo.get(repo) || 0) + 1)
-      if (result.resolved) {
-        byRepoResolved.set(repo, (byRepoResolved.get(repo) || 0) + 1)
-      }
-    }
-
-    // Paginate results
-    const paginatedResults = results.slice(offset, offset + limit)
-
-    return NextResponse.json({
-      run,
-      stats: {
-        totalTests,
-        resolvedTests,
-        passRate: parseFloat(passRate.toFixed(2)),
-        byRepo: Object.fromEntries(byRepo),
-        byRepoResolved: Object.fromEntries(byRepoResolved),
-      },
-      pagination: {
-        limit,
-        offset,
-        total: totalTests,
-      },
-      results: paginatedResults,
-    })
-  } catch (error) {
-    console.error('Benchmark API error:', error)
-    return NextResponse.json(
-      { error: 'Failed to load benchmark data' },
-      { status: 500 }
-    )
-  }
-}