insights.ts

Full source

import { execFileSync } from 'child_process'
import { diffLines } from 'diff'
import { constants as fsConstants } from 'fs'
import {
  copyFile,
  mkdir,
  mkdtemp,
  readdir,
  readFile,
  rm,
  unlink,
  writeFile,
} from 'fs/promises'
import { tmpdir } from 'os'
import { extname, join } from 'path'
import type { Command } from '../commands.js'
import { queryWithModel } from '../services/api/claude.js'
import {
  AGENT_TOOL_NAME,
  LEGACY_AGENT_TOOL_NAME,
} from '../tools/AgentTool/constants.js'
import type { LogOption } from '../types/logs.js'
import { getClaudeConfigHomeDir } from '../utils/envUtils.js'
import { toError } from '../utils/errors.js'
import { execFileNoThrow } from '../utils/execFileNoThrow.js'
import { logError } from '../utils/log.js'
import { extractTextContent } from '../utils/messages.js'
import { getDefaultOpusModel } from '../utils/model/model.js'
import {
  getProjectsDir,
  getSessionFilesWithMtime,
  getSessionIdFromLog,
  loadAllLogsFromSessionFile,
} from '../utils/sessionStorage.js'
import { jsonParse, jsonStringify } from '../utils/slowOperations.js'
import { countCharInString } from '../utils/stringUtils.js'
import { asSystemPrompt } from '../utils/systemPromptType.js'
import { escapeXmlAttr as escapeHtml } from '../utils/xml.js'

// Model for facet extraction and summarization (Opus - best quality)
function getAnalysisModel(): string {
  return getDefaultOpusModel()
}

// Model for narrative insights (Opus - best quality)
function getInsightsModel(): string {
  return getDefaultOpusModel()
}

// ============================================================================
// Homespace Data Collection
// ============================================================================

type RemoteHostInfo = {
  name: string
  sessionCount: number
}

/* eslint-disable custom-rules/no-process-env-top-level */
const getRunningRemoteHosts: () => Promise<string[]> =
  process.env.USER_TYPE === 'ant'
    ? async () => {
        const { stdout, code } = await execFileNoThrow(
          'coder',
          ['list', '-o', 'json'],
          { timeout: 30000 },
        )
        if (code !== 0) return []
        try {
          const workspaces = jsonParse(stdout) as Array<{
            name: string
            latest_build?: { status?: string }
          }>
          return workspaces
            .filter(w => w.latest_build?.status === 'running')
            .map(w => w.name)
        } catch {
          return []
        }
      }
    : async () => []

const getRemoteHostSessionCount: (hs: string) => Promise<number> =
  process.env.USER_TYPE === 'ant'
    ? async (homespace: string) => {
        const { stdout, code } = await execFileNoThrow(
          'ssh',
          [
            `${homespace}.coder`,
            'find /root/.claude/projects -name "*.jsonl" 2>/dev/null | wc -l',
          ],
          { timeout: 30000 },
        )
        if (code !== 0) return 0
        return parseInt(stdout.trim(), 10) || 0
      }
    : async () => 0

const collectFromRemoteHost: (
  hs: string,
  destDir: string,
) => Promise<{ copied: number; skipped: number }> =
  process.env.USER_TYPE === 'ant'
    ? async (homespace: string, destDir: string) => {
        const result = { copied: 0, skipped: 0 }

        // Create temp directory
        const tempDir = await mkdtemp(join(tmpdir(), 'claude-hs-'))

        try {
          // SCP the projects folder
          const scpResult = await execFileNoThrow(
            'scp',
            ['-rq', `${homespace}.coder:/root/.claude/projects/`, tempDir],
            { timeout: 300000 },
          )
          if (scpResult.code !== 0) {
            // SCP failed
            return result
          }

          const projectsDir = join(tempDir, 'projects')
          let projectDirents: Awaited<ReturnType<typeof readdir>>
          try {
            projectDirents = await readdir(projectsDir, { withFileTypes: true })
          } catch {
            return result
          }

          // Merge into destination (parallel per project directory)
          await Promise.all(
            projectDirents.map(async dirent => {
              const projectName = dirent.name
              const projectPath = join(projectsDir, projectName)

              // Skip if not a directory
              if (!dirent.isDirectory()) return

              const destProjectName = `${projectName}__${homespace}`
              const destProjectPath = join(destDir, destProjectName)

              try {
                await mkdir(destProjectPath, { recursive: true })
              } catch {
                // Directory may already exist
              }

              // Copy session files (skip existing)
              let files: Awaited<ReturnType<typeof readdir>>
              try {
                files = await readdir(projectPath, { withFileTypes: true })
              } catch {
                return
              }
              await Promise.all(
                files.map(async fileDirent => {
                  const fileName = fileDirent.name
                  if (!fileName.endsWith('.jsonl')) return

                  const srcFile = join(projectPath, fileName)
                  const destFile = join(destProjectPath, fileName)

                  try {
                    await copyFile(srcFile, destFile, fsConstants.COPYFILE_EXCL)
                    result.copied++
                  } catch {
                    // EEXIST from COPYFILE_EXCL means dest already exists
                    result.skipped++
                  }
                }),
              )
            }),
          )
        } finally {
          try {
            await rm(tempDir, { recursive: true, force: true })
          } catch {
            // Ignore cleanup errors
          }
        }

        return result
      }
    : async () => ({ copied: 0, skipped: 0 })

const collectAllRemoteHostData: (destDir: string) => Promise<{
  hosts: RemoteHostInfo[]
  totalCopied: number
  totalSkipped: number
}> =
  process.env.USER_TYPE === 'ant'
    ? async (destDir: string) => {
        const rHosts = await getRunningRemoteHosts()
        const result: RemoteHostInfo[] = []
        let totalCopied = 0
        let totalSkipped = 0

        // Collect from all hosts in parallel (SCP per host can take seconds)
        const hostResults = await Promise.all(
          rHosts.map(async hs => {
            const sessionCount = await getRemoteHostSessionCount(hs)
            if (sessionCount > 0) {
              const { copied, skipped } = await collectFromRemoteHost(
                hs,
                destDir,
              )
              return { name: hs, sessionCount, copied, skipped }
            }
            return { name: hs, sessionCount, copied: 0, skipped: 0 }
          }),
        )

        for (const hr of hostResults) {
          result.push({ name: hr.name, sessionCount: hr.sessionCount })
          totalCopied += hr.copied
          totalSkipped += hr.skipped
        }

        return { hosts: result, totalCopied, totalSkipped }
      }
    : async () => ({ hosts: [], totalCopied: 0, totalSkipped: 0 })
/* eslint-enable custom-rules/no-process-env-top-level */

// ============================================================================
// Types
// ============================================================================

type SessionMeta = {
  session_id: string
  project_path: string
  start_time: string
  duration_minutes: number
  user_message_count: number
  assistant_message_count: number
  tool_counts: Record<string, number>
  languages: Record<string, number>
  git_commits: number
  git_pushes: number
  input_tokens: number
  output_tokens: number
  first_prompt: string
  summary?: string
  // New stats
  user_interruptions: number
  user_response_times: number[]
  tool_errors: number
  tool_error_categories: Record<string, number>
  uses_task_agent: boolean
  uses_mcp: boolean
  uses_web_search: boolean
  uses_web_fetch: boolean
  // Additional stats
  lines_added: number
  lines_removed: number
  files_modified: number
  message_hours: number[]
  user_message_timestamps: string[] // ISO timestamps for multi-clauding detection
}

type SessionFacets = {
  session_id: string
  underlying_goal: string
  goal_categories: Record<string, number>
  outcome: string
  user_satisfaction_counts: Record<string, number>
  claude_helpfulness: string
  session_type: string
  friction_counts: Record<string, number>
  friction_detail: string
  primary_success: string
  brief_summary: string
  user_instructions_to_claude?: string[]
}

type AggregatedData = {
  total_sessions: number
  total_sessions_scanned?: number
  sessions_with_facets: number
  date_range: { start: string; end: string }
  total_messages: number
  total_duration_hours: number
  total_input_tokens: number
  total_output_tokens: number
  tool_counts: Record<string, number>
  languages: Record<string, number>
  git_commits: number
  git_pushes: number
  projects: Record<string, number>
  goal_categories: Record<string, number>
  outcomes: Record<string, number>
  satisfaction: Record<string, number>
  helpfulness: Record<string, number>
  session_types: Record<string, number>
  friction: Record<string, number>
  success: Record<string, number>
  session_summaries: Array<{
    id: string
    date: string
    summary: string
    goal?: string
  }>
  // New aggregated stats
  total_interruptions: number
  total_tool_errors: number
  tool_error_categories: Record<string, number>
  user_response_times: number[]
  median_response_time: number
  avg_response_time: number
  sessions_using_task_agent: number
  sessions_using_mcp: number
  sessions_using_web_search: number
  sessions_using_web_fetch: number
  // Additional stats from Python reference
  total_lines_added: number
  total_lines_removed: number
  total_files_modified: number
  days_active: number
  messages_per_day: number
  message_hours: number[] // Hour of day for each user message (for time of day chart)
  // Multi-clauding stats (matching Python reference)
  multi_clauding: {
    overlap_events: number
    sessions_involved: number
    user_messages_during: number
  }
}

// ============================================================================
// Constants
// ============================================================================

const EXTENSION_TO_LANGUAGE: Record<string, string> = {
  '.ts': 'TypeScript',
  '.tsx': 'TypeScript',
  '.js': 'JavaScript',
  '.jsx': 'JavaScript',
  '.py': 'Python',
  '.rb': 'Ruby',
  '.go': 'Go',
  '.rs': 'Rust',
  '.java': 'Java',
  '.md': 'Markdown',
  '.json': 'JSON',
  '.yaml': 'YAML',
  '.yml': 'YAML',
  '.sh': 'Shell',
  '.css': 'CSS',
  '.html': 'HTML',
}

// Label map for cleaning up category names (matching Python reference)
const LABEL_MAP: Record<string, string> = {
  // Goal categories
  debug_investigate: 'Debug/Investigate',
  implement_feature: 'Implement Feature',
  fix_bug: 'Fix Bug',
  write_script_tool: 'Write Script/Tool',
  refactor_code: 'Refactor Code',
  configure_system: 'Configure System',
  create_pr_commit: 'Create PR/Commit',
  analyze_data: 'Analyze Data',
  understand_codebase: 'Understand Codebase',
  write_tests: 'Write Tests',
  write_docs: 'Write Docs',
  deploy_infra: 'Deploy/Infra',
  warmup_minimal: 'Cache Warmup',
  // Success factors
  fast_accurate_search: 'Fast/Accurate Search',
  correct_code_edits: 'Correct Code Edits',
  good_explanations: 'Good Explanations',
  proactive_help: 'Proactive Help',
  multi_file_changes: 'Multi-file Changes',
  handled_complexity: 'Multi-file Changes',
  good_debugging: 'Good Debugging',
  // Friction types
  misunderstood_request: 'Misunderstood Request',
  wrong_approach: 'Wrong Approach',
  buggy_code: 'Buggy Code',
  user_rejected_action: 'User Rejected Action',
  claude_got_blocked: 'Claude Got Blocked',
  user_stopped_early: 'User Stopped Early',
  wrong_file_or_location: 'Wrong File/Location',
  excessive_changes: 'Excessive Changes',
  slow_or_verbose: 'Slow/Verbose',
  tool_failed: 'Tool Failed',
  user_unclear: 'User Unclear',
  external_issue: 'External Issue',
  // Satisfaction labels
  frustrated: 'Frustrated',
  dissatisfied: 'Dissatisfied',
  likely_satisfied: 'Likely Satisfied',
  satisfied: 'Satisfied',
  happy: 'Happy',
  unsure: 'Unsure',
  neutral: 'Neutral',
  delighted: 'Delighted',
  // Session types
  single_task: 'Single Task',
  multi_task: 'Multi Task',
  iterative_refinement: 'Iterative Refinement',
  exploration: 'Exploration',
  quick_question: 'Quick Question',
  // Outcomes
  fully_achieved: 'Fully Achieved',
  mostly_achieved: 'Mostly Achieved',
  partially_achieved: 'Partially Achieved',
  not_achieved: 'Not Achieved',
  unclear_from_transcript: 'Unclear',
  // Helpfulness
  unhelpful: 'Unhelpful',
  slightly_helpful: 'Slightly Helpful',
  moderately_helpful: 'Moderately Helpful',
  very_helpful: 'Very Helpful',
  essential: 'Essential',
}

// Lazy getters: getClaudeConfigHomeDir() is memoized and reads process.env.
// Calling it at module scope would populate the memoize cache before
// entrypoints can set CLAUDE_CONFIG_DIR, breaking all 150+ other callers.
function getDataDir(): string {
  return join(getClaudeConfigHomeDir(), 'usage-data')
}
function getFacetsDir(): string {
  return join(getDataDir(), 'facets')
}
function getSessionMetaDir(): string {
  return join(getDataDir(), 'session-meta')
}

const FACET_EXTRACTION_PROMPT = `Analyze this Claude Code session and extract structured facets.

CRITICAL GUIDELINES:

1. **goal_categories**: Count ONLY what the USER explicitly asked for.
   - DO NOT count Claude's autonomous codebase exploration
   - DO NOT count work Claude decided to do on its own
   - ONLY count when user says "can you...", "please...", "I need...", "let's..."

2. **user_satisfaction_counts**: Base ONLY on explicit user signals.
   - "Yay!", "great!", "perfect!" → happy
   - "thanks", "looks good", "that works" → satisfied
   - "ok, now let's..." (continuing without complaint) → likely_satisfied
   - "that's not right", "try again" → dissatisfied
   - "this is broken", "I give up" → frustrated

3. **friction_counts**: Be specific about what went wrong.
   - misunderstood_request: Claude interpreted incorrectly
   - wrong_approach: Right goal, wrong solution method
   - buggy_code: Code didn't work correctly
   - user_rejected_action: User said no/stop to a tool call
   - excessive_changes: Over-engineered or changed too much

4. If very short or just warmup, use warmup_minimal for goal_category

SESSION:
`

// ============================================================================
// Helper Functions
// ============================================================================

function getLanguageFromPath(filePath: string): string | null {
  const ext = extname(filePath).toLowerCase()
  return EXTENSION_TO_LANGUAGE[ext] || null
}

function extractToolStats(log: LogOption): {
  toolCounts: Record<string, number>
  languages: Record<string, number>
  gitCommits: number
  gitPushes: number
  inputTokens: number
  outputTokens: number
  // New stats
  userInterruptions: number
  userResponseTimes: number[]
  toolErrors: number
  toolErrorCategories: Record<string, number>
  usesTaskAgent: boolean
  usesMcp: boolean
  usesWebSearch: boolean
  usesWebFetch: boolean
  // Additional stats
  linesAdded: number
  linesRemoved: number
  filesModified: Set<string>
  messageHours: number[]
  userMessageTimestamps: string[] // ISO timestamps for multi-clauding detection
} {
  const toolCounts: Record<string, number> = {}
  const languages: Record<string, number> = {}
  let gitCommits = 0
  let gitPushes = 0
  let inputTokens = 0
  let outputTokens = 0

  // New stats
  let userInterruptions = 0
  const userResponseTimes: number[] = []
  let toolErrors = 0
  const toolErrorCategories: Record<string, number> = {}
  let usesTaskAgent = false

  // Additional stats
  let linesAdded = 0
  let linesRemoved = 0
  const filesModified = new Set<string>()
  const messageHours: number[] = []
  const userMessageTimestamps: string[] = [] // For multi-clauding detection
  let usesMcp = false
  let usesWebSearch = false
  let usesWebFetch = false
  let lastAssistantTimestamp: string | null = null

  for (const msg of log.messages) {
    // Get message timestamp for response time calculation
    const msgTimestamp = (msg as { timestamp?: string }).timestamp

    if (msg.type === 'assistant' && msg.message) {
      // Track timestamp for response time calculation
      if (msgTimestamp) {
        lastAssistantTimestamp = msgTimestamp
      }

      const usage = (
        msg.message as {
          usage?: { input_tokens?: number; output_tokens?: number }
        }
      ).usage
      if (usage) {
        inputTokens += usage.input_tokens || 0
        outputTokens += usage.output_tokens || 0
      }

      const content = msg.message.content
      if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'tool_use' && 'name' in block) {
            const toolName = block.name as string
            toolCounts[toolName] = (toolCounts[toolName] || 0) + 1

            // Check for special tool usage
            if (
              toolName === AGENT_TOOL_NAME ||
              toolName === LEGACY_AGENT_TOOL_NAME
            )
              usesTaskAgent = true
            if (toolName.startsWith('mcp__')) usesMcp = true
            if (toolName === 'WebSearch') usesWebSearch = true
            if (toolName === 'WebFetch') usesWebFetch = true

            const input = (block as { input?: Record<string, unknown> }).input

            if (input) {
              const filePath = (input.file_path as string) || ''
              if (filePath) {
                const lang = getLanguageFromPath(filePath)
                if (lang) {
                  languages[lang] = (languages[lang] || 0) + 1
                }
                // Track files modified by Edit/Write tools
                if (toolName === 'Edit' || toolName === 'Write') {
                  filesModified.add(filePath)
                }
              }

              if (toolName === 'Edit') {
                const oldString = (input.old_string as string) || ''
                const newString = (input.new_string as string) || ''
                for (const change of diffLines(oldString, newString)) {
                  if (change.added) linesAdded += change.count || 0
                  if (change.removed) linesRemoved += change.count || 0
                }
              }

              // Track lines from Write tool (all added)
              if (toolName === 'Write') {
                const writeContent = (input.content as string) || ''
                if (writeContent) {
                  linesAdded += countCharInString(writeContent, '\n') + 1
                }
              }

              const command = (input.command as string) || ''
              if (command.includes('git commit')) gitCommits++
              if (command.includes('git push')) gitPushes++
            }
          }
        }
      }
    }

    // Check user messages
    if (msg.type === 'user' && msg.message) {
      const content = msg.message.content

      // Check if this is an actual human message (has text) vs just tool_result
      // matching Python reference logic
      let isHumanMessage = false
      if (typeof content === 'string' && content.trim()) {
        isHumanMessage = true
      } else if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'text' && 'text' in block) {
            isHumanMessage = true
            break
          }
        }
      }

      // Only track message hours and response times for actual human messages
      if (isHumanMessage) {
        // Track message hour for time-of-day analysis and timestamp for multi-clauding
        if (msgTimestamp) {
          try {
            const msgDate = new Date(msgTimestamp)
            const hour = msgDate.getHours() // Local hour 0-23
            messageHours.push(hour)
            // Collect timestamp for multi-clauding detection (matching Python)
            userMessageTimestamps.push(msgTimestamp)
          } catch {
            // Skip invalid timestamps
          }
        }

        // Calculate response time (time from last assistant message to this user message)
        // Only count gaps > 2 seconds (real user think time, not tool results)
        if (lastAssistantTimestamp && msgTimestamp) {
          const assistantTime = new Date(lastAssistantTimestamp).getTime()
          const userTime = new Date(msgTimestamp).getTime()
          const responseTimeSec = (userTime - assistantTime) / 1000
          // Only count reasonable response times (2s-1 hour) matching Python
          if (responseTimeSec > 2 && responseTimeSec < 3600) {
            userResponseTimes.push(responseTimeSec)
          }
        }
      }

      // Process tool results (for error tracking)
      if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'tool_result' && 'content' in block) {
            const isError = (block as { is_error?: boolean }).is_error

            // Count and categorize tool errors (matching Python reference logic)
            if (isError) {
              toolErrors++
              const resultContent = (block as { content?: string }).content
              let category = 'Other'
              if (typeof resultContent === 'string') {
                const lowerContent = resultContent.toLowerCase()
                if (lowerContent.includes('exit code')) {
                  category = 'Command Failed'
                } else if (
                  lowerContent.includes('rejected') ||
                  lowerContent.includes("doesn't want")
                ) {
                  category = 'User Rejected'
                } else if (
                  lowerContent.includes('string to replace not found') ||
                  lowerContent.includes('no changes')
                ) {
                  category = 'Edit Failed'
                } else if (lowerContent.includes('modified since read')) {
                  category = 'File Changed'
                } else if (
                  lowerContent.includes('exceeds maximum') ||
                  lowerContent.includes('too large')
                ) {
                  category = 'File Too Large'
                } else if (
                  lowerContent.includes('file not found') ||
                  lowerContent.includes('does not exist')
                ) {
                  category = 'File Not Found'
                }
              }
              toolErrorCategories[category] =
                (toolErrorCategories[category] || 0) + 1
            }
          }
        }
      }

      // Check for interruptions (matching Python reference)
      if (typeof content === 'string') {
        if (content.includes('[Request interrupted by user')) {
          userInterruptions++
        }
      } else if (Array.isArray(content)) {
        for (const block of content) {
          if (
            block.type === 'text' &&
            'text' in block &&
            (block.text as string).includes('[Request interrupted by user')
          ) {
            userInterruptions++
            break
          }
        }
      }
    }
  }

  return {
    toolCounts,
    languages,
    gitCommits,
    gitPushes,
    inputTokens,
    outputTokens,
    // New stats
    userInterruptions,
    userResponseTimes,
    toolErrors,
    toolErrorCategories,
    usesTaskAgent,
    usesMcp,
    usesWebSearch,
    usesWebFetch,
    // Additional stats
    linesAdded,
    linesRemoved,
    filesModified,
    messageHours,
    userMessageTimestamps,
  }
}

function hasValidDates(log: LogOption): boolean {
  return (
    !Number.isNaN(log.created.getTime()) &&
    !Number.isNaN(log.modified.getTime())
  )
}

function logToSessionMeta(log: LogOption): SessionMeta {
  const stats = extractToolStats(log)
  const sessionId = getSessionIdFromLog(log) || 'unknown'
  const startTime = log.created.toISOString()
  const durationMinutes = Math.round(
    (log.modified.getTime() - log.created.getTime()) / 1000 / 60,
  )

  let userMessageCount = 0
  let assistantMessageCount = 0
  for (const msg of log.messages) {
    if (msg.type === 'assistant') assistantMessageCount++
    // Only count user messages that have actual text content (human messages)
    // not just tool_result messages (matching Python reference)
    if (msg.type === 'user' && msg.message) {
      const content = msg.message.content
      let isHumanMessage = false
      if (typeof content === 'string' && content.trim()) {
        isHumanMessage = true
      } else if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'text' && 'text' in block) {
            isHumanMessage = true
            break
          }
        }
      }
      if (isHumanMessage) {
        userMessageCount++
      }
    }
  }

  return {
    session_id: sessionId,
    project_path: log.projectPath || '',
    start_time: startTime,
    duration_minutes: durationMinutes,
    user_message_count: userMessageCount,
    assistant_message_count: assistantMessageCount,
    tool_counts: stats.toolCounts,
    languages: stats.languages,
    git_commits: stats.gitCommits,
    git_pushes: stats.gitPushes,
    input_tokens: stats.inputTokens,
    output_tokens: stats.outputTokens,
    first_prompt: log.firstPrompt || '',
    summary: log.summary,
    // New stats
    user_interruptions: stats.userInterruptions,
    user_response_times: stats.userResponseTimes,
    tool_errors: stats.toolErrors,
    tool_error_categories: stats.toolErrorCategories,
    uses_task_agent: stats.usesTaskAgent,
    uses_mcp: stats.usesMcp,
    uses_web_search: stats.usesWebSearch,
    uses_web_fetch: stats.usesWebFetch,
    // Additional stats
    lines_added: stats.linesAdded,
    lines_removed: stats.linesRemoved,
    files_modified: stats.filesModified.size,
    message_hours: stats.messageHours,
    user_message_timestamps: stats.userMessageTimestamps,
  }
}

/**
 * Deduplicate conversation branches within the same session.
 *
 * When a session file has multiple leaf messages (from retries or branching),
 * loadAllLogsFromSessionFile produces one LogOption per leaf. Each branch
 * shares the same root message, so its duration overlaps with sibling
 * branches. This keeps only the branch with the most user messages
 * (tie-break by longest duration) per session_id.
 */
export function deduplicateSessionBranches(
  entries: Array<{ log: LogOption; meta: SessionMeta }>,
): Array<{ log: LogOption; meta: SessionMeta }> {
  const bestBySession = new Map<string, { log: LogOption; meta: SessionMeta }>()
  for (const entry of entries) {
    const id = entry.meta.session_id
    const existing = bestBySession.get(id)
    if (
      !existing ||
      entry.meta.user_message_count > existing.meta.user_message_count ||
      (entry.meta.user_message_count === existing.meta.user_message_count &&
        entry.meta.duration_minutes > existing.meta.duration_minutes)
    ) {
      bestBySession.set(id, entry)
    }
  }
  return [...bestBySession.values()]
}

function formatTranscriptForFacets(log: LogOption): string {
  const lines: string[] = []
  const meta = logToSessionMeta(log)

  lines.push(`Session: ${meta.session_id.slice(0, 8)}`)
  lines.push(`Date: ${meta.start_time}`)
  lines.push(`Project: ${meta.project_path}`)
  lines.push(`Duration: ${meta.duration_minutes} min`)
  lines.push('')

  for (const msg of log.messages) {
    if (msg.type === 'user' && msg.message) {
      const content = msg.message.content
      if (typeof content === 'string') {
        lines.push(`[User]: ${content.slice(0, 500)}`)
      } else if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'text' && 'text' in block) {
            lines.push(`[User]: ${(block.text as string).slice(0, 500)}`)
          }
        }
      }
    } else if (msg.type === 'assistant' && msg.message) {
      const content = msg.message.content
      if (Array.isArray(content)) {
        for (const block of content) {
          if (block.type === 'text' && 'text' in block) {
            lines.push(`[Assistant]: ${(block.text as string).slice(0, 300)}`)
          } else if (block.type === 'tool_use' && 'name' in block) {
            lines.push(`[Tool: ${block.name}]`)
          }
        }
      }
    }
  }

  return lines.join('\n')
}

const SUMMARIZE_CHUNK_PROMPT = `Summarize this portion of a Claude Code session transcript. Focus on:
1. What the user asked for
2. What Claude did (tools used, files modified)
3. Any friction or issues
4. The outcome

Keep it concise - 3-5 sentences. Preserve specific details like file names, error messages, and user feedback.

TRANSCRIPT CHUNK:
`

async function summarizeTranscriptChunk(chunk: string): Promise<string> {
  try {
    const result = await queryWithModel({
      systemPrompt: asSystemPrompt([]),
      userPrompt: SUMMARIZE_CHUNK_PROMPT + chunk,
      signal: new AbortController().signal,
      options: {
        model: getAnalysisModel(),
        querySource: 'insights',
        agents: [],
        isNonInteractiveSession: true,
        hasAppendSystemPrompt: false,
        mcpTools: [],
        maxOutputTokensOverride: 500,
      },
    })

    const text = extractTextContent(result.message.content)
    return text || chunk.slice(0, 2000)
  } catch {
    // On error, just return truncated chunk
    return chunk.slice(0, 2000)
  }
}

async function formatTranscriptWithSummarization(
  log: LogOption,
): Promise<string> {
  const fullTranscript = formatTranscriptForFacets(log)

  // If under 30k chars, use as-is
  if (fullTranscript.length <= 30000) {
    return fullTranscript
  }

  // For long transcripts, split into chunks and summarize in parallel
  const CHUNK_SIZE = 25000
  const chunks: string[] = []

  for (let i = 0; i < fullTranscript.length; i += CHUNK_SIZE) {
    chunks.push(fullTranscript.slice(i, i + CHUNK_SIZE))
  }

  // Summarize all chunks in parallel
  const summaries = await Promise.all(chunks.map(summarizeTranscriptChunk))

  // Combine summaries with session header
  const meta = logToSessionMeta(log)
  const header = [
    `Session: ${meta.session_id.slice(0, 8)}`,
    `Date: ${meta.start_time}`,
    `Project: ${meta.project_path}`,
    `Duration: ${meta.duration_minutes} min`,
    `[Long session - ${chunks.length} parts summarized]`,
    '',
  ].join('\n')

  return header + summaries.join('\n\n---\n\n')
}

async function loadCachedFacets(
  sessionId: string,
): Promise<SessionFacets | null> {
  const facetPath = join(getFacetsDir(), `${sessionId}.json`)
  try {
    const content = await readFile(facetPath, { encoding: 'utf-8' })
    const parsed: unknown = jsonParse(content)
    if (!isValidSessionFacets(parsed)) {
      // Delete corrupted cache file so it gets re-extracted next run
      try {
        await unlink(facetPath)
      } catch {
        // Ignore deletion errors
      }
      return null
    }
    return parsed
  } catch {
    return null
  }
}

async function saveFacets(facets: SessionFacets): Promise<void> {
  try {
    await mkdir(getFacetsDir(), { recursive: true })
  } catch {
    // Directory may already exist
  }
  const facetPath = join(getFacetsDir(), `${facets.session_id}.json`)
  await writeFile(facetPath, jsonStringify(facets, null, 2), {
    encoding: 'utf-8',
    mode: 0o600,
  })
}

async function loadCachedSessionMeta(
  sessionId: string,
): Promise<SessionMeta | null> {
  const metaPath = join(getSessionMetaDir(), `${sessionId}.json`)
  try {
    const content = await readFile(metaPath, { encoding: 'utf-8' })
    return jsonParse(content)
  } catch {
    return null
  }
}

async function saveSessionMeta(meta: SessionMeta): Promise<void> {
  try {
    await mkdir(getSessionMetaDir(), { recursive: true })
  } catch {
    // Directory may already exist
  }
  const metaPath = join(getSessionMetaDir(), `${meta.session_id}.json`)
  await writeFile(metaPath, jsonStringify(meta, null, 2), {
    encoding: 'utf-8',
    mode: 0o600,
  })
}

async function extractFacetsFromAPI(
  log: LogOption,
  sessionId: string,
): Promise<SessionFacets | null> {
  try {
    // Use summarization for long transcripts
    const transcript = await formatTranscriptWithSummarization(log)

    // Build prompt asking for JSON directly (no tool use)
    const jsonPrompt = `${FACET_EXTRACTION_PROMPT}${transcript}

RESPOND WITH ONLY A VALID JSON OBJECT matching this schema:
{
  "underlying_goal": "What the user fundamentally wanted to achieve",
  "goal_categories": {"category_name": count, ...},
  "outcome": "fully_achieved|mostly_achieved|partially_achieved|not_achieved|unclear_from_transcript",
  "user_satisfaction_counts": {"level": count, ...},
  "claude_helpfulness": "unhelpful|slightly_helpful|moderately_helpful|very_helpful|essential",
  "session_type": "single_task|multi_task|iterative_refinement|exploration|quick_question",
  "friction_counts": {"friction_type": count, ...},
  "friction_detail": "One sentence describing friction or empty",
  "primary_success": "none|fast_accurate_search|correct_code_edits|good_explanations|proactive_help|multi_file_changes|good_debugging",
  "brief_summary": "One sentence: what user wanted and whether they got it"
}`

    const result = await queryWithModel({
      systemPrompt: asSystemPrompt([]),
      userPrompt: jsonPrompt,
      signal: new AbortController().signal,
      options: {
        model: getAnalysisModel(),
        querySource: 'insights',
        agents: [],
        isNonInteractiveSession: true,
        hasAppendSystemPrompt: false,
        mcpTools: [],
        maxOutputTokensOverride: 4096,
      },
    })

    const text = extractTextContent(result.message.content)

    // Parse JSON from response
    const jsonMatch = text.match(/\{[\s\S]*\}/)
    if (!jsonMatch) return null

    const parsed: unknown = jsonParse(jsonMatch[0])
    if (!isValidSessionFacets(parsed)) return null
    const facets: SessionFacets = { ...parsed, session_id: sessionId }
    return facets
  } catch (err) {
    logError(new Error(`Facet extraction failed: ${toError(err).message}`))
    return null
  }
}

/**
 * Detects multi-clauding (using multiple Claude sessions concurrently).
 * Uses a sliding window to find the pattern: session1 -> session2 -> session1
 * within a 30-minute window.
 */
export function detectMultiClauding(
  sessions: Array<{
    session_id: string
    user_message_timestamps: string[]
  }>,
): {
  overlap_events: number
  sessions_involved: number
  user_messages_during: number
} {
  const OVERLAP_WINDOW_MS = 30 * 60000
  const allSessionMessages: Array<{ ts: number; sessionId: string }> = []

  for (const session of sessions) {
    for (const timestamp of session.user_message_timestamps) {
      try {
        const ts = new Date(timestamp).getTime()
        allSessionMessages.push({ ts, sessionId: session.session_id })
      } catch {
        // Skip invalid timestamps
      }
    }
  }

  allSessionMessages.sort((a, b) => a.ts - b.ts)

  const multiClaudeSessionPairs = new Set<string>()
  const messagesDuringMulticlaude = new Set<string>()

  // Sliding window: sessionLastIndex tracks the most recent index for each session
  let windowStart = 0
  const sessionLastIndex = new Map<string, number>()

  for (let i = 0; i < allSessionMessages.length; i++) {
    const msg = allSessionMessages[i]!

    // Shrink window from the left
    while (
      windowStart < i &&
      msg.ts - allSessionMessages[windowStart]!.ts > OVERLAP_WINDOW_MS
    ) {
      const expiring = allSessionMessages[windowStart]!
      if (sessionLastIndex.get(expiring.sessionId) === windowStart) {
        sessionLastIndex.delete(expiring.sessionId)
      }
      windowStart++
    }

    // Check if this session appeared earlier in the window (pattern: s1 -> s2 -> s1)
    const prevIndex = sessionLastIndex.get(msg.sessionId)
    if (prevIndex !== undefined) {
      for (let j = prevIndex + 1; j < i; j++) {
        const between = allSessionMessages[j]!
        if (between.sessionId !== msg.sessionId) {
          const pair = [msg.sessionId, between.sessionId].sort().join(':')
          multiClaudeSessionPairs.add(pair)
          messagesDuringMulticlaude.add(
            `${allSessionMessages[prevIndex]!.ts}:${msg.sessionId}`,
          )
          messagesDuringMulticlaude.add(`${between.ts}:${between.sessionId}`)
          messagesDuringMulticlaude.add(`${msg.ts}:${msg.sessionId}`)
          break
        }
      }
    }

    sessionLastIndex.set(msg.sessionId, i)
  }

  const sessionsWithOverlaps = new Set<string>()
  for (const pair of multiClaudeSessionPairs) {
    const [s1, s2] = pair.split(':')
    if (s1) sessionsWithOverlaps.add(s1)
    if (s2) sessionsWithOverlaps.add(s2)
  }

  return {
    overlap_events: multiClaudeSessionPairs.size,
    sessions_involved: sessionsWithOverlaps.size,
    user_messages_during: messagesDuringMulticlaude.size,
  }
}

function aggregateData(
  sessions: SessionMeta[],
  facets: Map<string, SessionFacets>,
): AggregatedData {
  const result: AggregatedData = {
    total_sessions: sessions.length,
    sessions_with_facets: facets.size,
    date_range: { start: '', end: '' },
    total_messages: 0,
    total_duration_hours: 0,
    total_input_tokens: 0,
    total_output_tokens: 0,
    tool_counts: {},
    languages: {},
    git_commits: 0,
    git_pushes: 0,
    projects: {},
    goal_categories: {},
    outcomes: {},
    satisfaction: {},
    helpfulness: {},
    session_types: {},
    friction: {},
    success: {},
    session_summaries: [],
    // New stats
    total_interruptions: 0,
    total_tool_errors: 0,
    tool_error_categories: {},
    user_response_times: [],
    median_response_time: 0,
    avg_response_time: 0,
    sessions_using_task_agent: 0,
    sessions_using_mcp: 0,
    sessions_using_web_search: 0,
    sessions_using_web_fetch: 0,
    // Additional stats
    total_lines_added: 0,
    total_lines_removed: 0,
    total_files_modified: 0,
    days_active: 0,
    messages_per_day: 0,
    message_hours: [],
    // Multi-clauding stats (matching Python reference)
    multi_clauding: {
      overlap_events: 0,
      sessions_involved: 0,
      user_messages_during: 0,
    },
  }

  const dates: string[] = []
  const allResponseTimes: number[] = []
  const allMessageHours: number[] = []

  for (const session of sessions) {
    dates.push(session.start_time)
    result.total_messages += session.user_message_count
    result.total_duration_hours += session.duration_minutes / 60
    result.total_input_tokens += session.input_tokens
    result.total_output_tokens += session.output_tokens
    result.git_commits += session.git_commits
    result.git_pushes += session.git_pushes

    // New stats aggregation
    result.total_interruptions += session.user_interruptions
    result.total_tool_errors += session.tool_errors
    for (const [cat, count] of Object.entries(session.tool_error_categories)) {
      result.tool_error_categories[cat] =
        (result.tool_error_categories[cat] || 0) + count
    }
    allResponseTimes.push(...session.user_response_times)
    if (session.uses_task_agent) result.sessions_using_task_agent++
    if (session.uses_mcp) result.sessions_using_mcp++
    if (session.uses_web_search) result.sessions_using_web_search++
    if (session.uses_web_fetch) result.sessions_using_web_fetch++

    // Additional stats aggregation
    result.total_lines_added += session.lines_added
    result.total_lines_removed += session.lines_removed
    result.total_files_modified += session.files_modified
    allMessageHours.push(...session.message_hours)

    for (const [tool, count] of Object.entries(session.tool_counts)) {
      result.tool_counts[tool] = (result.tool_counts[tool] || 0) + count
    }

    for (const [lang, count] of Object.entries(session.languages)) {
      result.languages[lang] = (result.languages[lang] || 0) + count
    }

    if (session.project_path) {
      result.projects[session.project_path] =
        (result.projects[session.project_path] || 0) + 1
    }

    const sessionFacets = facets.get(session.session_id)
    if (sessionFacets) {
      // Goal categories
      for (const [cat, count] of safeEntries(sessionFacets.goal_categories)) {
        if (count > 0) {
          result.goal_categories[cat] =
            (result.goal_categories[cat] || 0) + count
        }
      }

      // Outcomes
      result.outcomes[sessionFacets.outcome] =
        (result.outcomes[sessionFacets.outcome] || 0) + 1

      // Satisfaction counts
      for (const [level, count] of safeEntries(
        sessionFacets.user_satisfaction_counts,
      )) {
        if (count > 0) {
          result.satisfaction[level] = (result.satisfaction[level] || 0) + count
        }
      }

      // Helpfulness
      result.helpfulness[sessionFacets.claude_helpfulness] =
        (result.helpfulness[sessionFacets.claude_helpfulness] || 0) + 1

      // Session types
      result.session_types[sessionFacets.session_type] =
        (result.session_types[sessionFacets.session_type] || 0) + 1

      // Friction counts
      for (const [type, count] of safeEntries(sessionFacets.friction_counts)) {
        if (count > 0) {
          result.friction[type] = (result.friction[type] || 0) + count
        }
      }

      // Success factors
      if (sessionFacets.primary_success !== 'none') {
        result.success[sessionFacets.primary_success] =
          (result.success[sessionFacets.primary_success] || 0) + 1
      }
    }

    if (result.session_summaries.length < 50) {
      result.session_summaries.push({
        id: session.session_id.slice(0, 8),
        date: session.start_time.split('T')[0] || '',
        summary: session.summary || session.first_prompt.slice(0, 100),
        goal: sessionFacets?.underlying_goal,
      })
    }
  }

  dates.sort()
  result.date_range.start = dates[0]?.split('T')[0] || ''
  result.date_range.end = dates[dates.length - 1]?.split('T')[0] || ''

  // Calculate response time stats
  result.user_response_times = allResponseTimes
  if (allResponseTimes.length > 0) {
    const sorted = [...allResponseTimes].sort((a, b) => a - b)
    result.median_response_time = sorted[Math.floor(sorted.length / 2)] || 0
    result.avg_response_time =
      allResponseTimes.reduce((a, b) => a + b, 0) / allResponseTimes.length
  }

  // Calculate days active and messages per day
  const uniqueDays = new Set(dates.map(d => d.split('T')[0]))
  result.days_active = uniqueDays.size
  result.messages_per_day =
    result.days_active > 0
      ? Math.round((result.total_messages / result.days_active) * 10) / 10
      : 0

  // Store message hours for time-of-day chart
  result.message_hours = allMessageHours

  result.multi_clauding = detectMultiClauding(sessions)

  return result
}

// ============================================================================
// Parallel Insights Generation (6 sections)
// ============================================================================

type InsightSection = {
  name: string
  prompt: string
  maxTokens: number
}

// Sections that run in parallel first
const INSIGHT_SECTIONS: InsightSection[] = [
  {
    name: 'project_areas',
    prompt: `Analyze this Claude Code usage data and identify project areas.

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "areas": [
    {"name": "Area name", "session_count": N, "description": "2-3 sentences about what was worked on and how Claude Code was used."}
  ]
}

Include 4-5 areas. Skip internal CC operations.`,
    maxTokens: 8192,
  },
  {
    name: 'interaction_style',
    prompt: `Analyze this Claude Code usage data and describe the user's interaction style.

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "narrative": "2-3 paragraphs analyzing HOW the user interacts with Claude Code. Use second person 'you'. Describe patterns: iterate quickly vs detailed upfront specs? Interrupt often or let Claude run? Include specific examples. Use **bold** for key insights.",
  "key_pattern": "One sentence summary of most distinctive interaction style"
}`,
    maxTokens: 8192,
  },
  {
    name: 'what_works',
    prompt: `Analyze this Claude Code usage data and identify what's working well for this user. Use second person ("you").

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "intro": "1 sentence of context",
  "impressive_workflows": [
    {"title": "Short title (3-6 words)", "description": "2-3 sentences describing the impressive workflow or approach. Use 'you' not 'the user'."}
  ]
}

Include 3 impressive workflows.`,
    maxTokens: 8192,
  },
  {
    name: 'friction_analysis',
    prompt: `Analyze this Claude Code usage data and identify friction points for this user. Use second person ("you").

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "intro": "1 sentence summarizing friction patterns",
  "categories": [
    {"category": "Concrete category name", "description": "1-2 sentences explaining this category and what could be done differently. Use 'you' not 'the user'.", "examples": ["Specific example with consequence", "Another example"]}
  ]
}

Include 3 friction categories with 2 examples each.`,
    maxTokens: 8192,
  },
  {
    name: 'suggestions',
    prompt: `Analyze this Claude Code usage data and suggest improvements.

## CC FEATURES REFERENCE (pick from these for features_to_try):
1. **MCP Servers**: Connect Claude to external tools, databases, and APIs via Model Context Protocol.
   - How to use: Run \`claude mcp add <server-name> -- <command>\`
   - Good for: database queries, Slack integration, GitHub issue lookup, connecting to internal APIs

2. **Custom Skills**: Reusable prompts you define as markdown files that run with a single /command.
   - How to use: Create \`.claude/skills/commit/SKILL.md\` with instructions. Then type \`/commit\` to run it.
   - Good for: repetitive workflows - /commit, /review, /test, /deploy, /pr, or complex multi-step workflows

3. **Hooks**: Shell commands that auto-run at specific lifecycle events.
   - How to use: Add to \`.claude/settings.json\` under "hooks" key.
   - Good for: auto-formatting code, running type checks, enforcing conventions

4. **Headless Mode**: Run Claude non-interactively from scripts and CI/CD.
   - How to use: \`claude -p "fix lint errors" --allowedTools "Edit,Read,Bash"\`
   - Good for: CI/CD integration, batch code fixes, automated reviews

5. **Task Agents**: Claude spawns focused sub-agents for complex exploration or parallel work.
   - How to use: Claude auto-invokes when helpful, or ask "use an agent to explore X"
   - Good for: codebase exploration, understanding complex systems

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "claude_md_additions": [
    {"addition": "A specific line or block to add to CLAUDE.md based on workflow patterns. E.g., 'Always run tests after modifying auth-related files'", "why": "1 sentence explaining why this would help based on actual sessions", "prompt_scaffold": "Instructions for where to add this in CLAUDE.md. E.g., 'Add under ## Testing section'"}
  ],
  "features_to_try": [
    {"feature": "Feature name from CC FEATURES REFERENCE above", "one_liner": "What it does", "why_for_you": "Why this would help YOU based on your sessions", "example_code": "Actual command or config to copy"}
  ],
  "usage_patterns": [
    {"title": "Short title", "suggestion": "1-2 sentence summary", "detail": "3-4 sentences explaining how this applies to YOUR work", "copyable_prompt": "A specific prompt to copy and try"}
  ]
}

IMPORTANT for claude_md_additions: PRIORITIZE instructions that appear MULTIPLE TIMES in the user data. If user told Claude the same thing in 2+ sessions (e.g., 'always run tests', 'use TypeScript'), that's a PRIME candidate - they shouldn't have to repeat themselves.

IMPORTANT for features_to_try: Pick 2-3 from the CC FEATURES REFERENCE above. Include 2-3 items for each category.`,
    maxTokens: 8192,
  },
  {
    name: 'on_the_horizon',
    prompt: `Analyze this Claude Code usage data and identify future opportunities.

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "intro": "1 sentence about evolving AI-assisted development",
  "opportunities": [
    {"title": "Short title (4-8 words)", "whats_possible": "2-3 ambitious sentences about autonomous workflows", "how_to_try": "1-2 sentences mentioning relevant tooling", "copyable_prompt": "Detailed prompt to try"}
  ]
}

Include 3 opportunities. Think BIG - autonomous workflows, parallel agents, iterating against tests.`,
    maxTokens: 8192,
  },
  ...(process.env.USER_TYPE === 'ant'
    ? [
        {
          name: 'cc_team_improvements',
          prompt: `Analyze this Claude Code usage data and suggest product improvements for the CC team.

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "improvements": [
    {"title": "Product/tooling improvement", "detail": "3-4 sentences describing the improvement", "evidence": "3-4 sentences with specific session examples"}
  ]
}

Include 2-3 improvements based on friction patterns observed.`,
          maxTokens: 8192,
        },
        {
          name: 'model_behavior_improvements',
          prompt: `Analyze this Claude Code usage data and suggest model behavior improvements.

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "improvements": [
    {"title": "Model behavior change", "detail": "3-4 sentences describing what the model should do differently", "evidence": "3-4 sentences with specific examples"}
  ]
}

Include 2-3 improvements based on friction patterns observed.`,
          maxTokens: 8192,
        },
      ]
    : []),
  {
    name: 'fun_ending',
    prompt: `Analyze this Claude Code usage data and find a memorable moment.

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "headline": "A memorable QUALITATIVE moment from the transcripts - not a statistic. Something human, funny, or surprising.",
  "detail": "Brief context about when/where this happened"
}

Find something genuinely interesting or amusing from the session summaries.`,
    maxTokens: 8192,
  },
]

type InsightResults = {
  at_a_glance?: {
    whats_working?: string
    whats_hindering?: string
    quick_wins?: string
    ambitious_workflows?: string
  }
  project_areas?: {
    areas?: Array<{ name: string; session_count: number; description: string }>
  }
  interaction_style?: {
    narrative?: string
    key_pattern?: string
  }
  what_works?: {
    intro?: string
    impressive_workflows?: Array<{ title: string; description: string }>
  }
  friction_analysis?: {
    intro?: string
    categories?: Array<{
      category: string
      description: string
      examples?: string[]
    }>
  }
  suggestions?: {
    claude_md_additions?: Array<{
      addition: string
      why: string
      where?: string
      prompt_scaffold?: string
    }>
    features_to_try?: Array<{
      feature: string
      one_liner: string
      why_for_you: string
      example_code?: string
    }>
    usage_patterns?: Array<{
      title: string
      suggestion: string
      detail?: string
      copyable_prompt?: string
    }>
  }
  on_the_horizon?: {
    intro?: string
    opportunities?: Array<{
      title: string
      whats_possible: string
      how_to_try?: string
      copyable_prompt?: string
    }>
  }
  cc_team_improvements?: {
    improvements?: Array<{
      title: string
      detail: string
      evidence?: string
    }>
  }
  model_behavior_improvements?: {
    improvements?: Array<{
      title: string
      detail: string
      evidence?: string
    }>
  }
  fun_ending?: {
    headline?: string
    detail?: string
  }
}

async function generateSectionInsight(
  section: InsightSection,
  dataContext: string,
): Promise<{ name: string; result: unknown }> {
  try {
    const result = await queryWithModel({
      systemPrompt: asSystemPrompt([]),
      userPrompt: section.prompt + '\n\nDATA:\n' + dataContext,
      signal: new AbortController().signal,
      options: {
        model: getInsightsModel(),
        querySource: 'insights',
        agents: [],
        isNonInteractiveSession: true,
        hasAppendSystemPrompt: false,
        mcpTools: [],
        maxOutputTokensOverride: section.maxTokens,
      },
    })

    const text = extractTextContent(result.message.content)

    if (text) {
      // Parse JSON from response
      const jsonMatch = text.match(/\{[\s\S]*\}/)
      if (jsonMatch) {
        try {
          return { name: section.name, result: jsonParse(jsonMatch[0]) }
        } catch {
          return { name: section.name, result: null }
        }
      }
    }
    return { name: section.name, result: null }
  } catch (err) {
    logError(new Error(`${section.name} failed: ${toError(err).message}`))
    return { name: section.name, result: null }
  }
}

async function generateParallelInsights(
  data: AggregatedData,
  facets: Map<string, SessionFacets>,
): Promise<InsightResults> {
  // Build data context string
  const facetSummaries = Array.from(facets.values())
    .slice(0, 50)
    .map(f => `- ${f.brief_summary} (${f.outcome}, ${f.claude_helpfulness})`)
    .join('\n')

  const frictionDetails = Array.from(facets.values())
    .filter(f => f.friction_detail)
    .slice(0, 20)
    .map(f => `- ${f.friction_detail}`)
    .join('\n')

  const userInstructions = Array.from(facets.values())
    .flatMap(f => f.user_instructions_to_claude || [])
    .slice(0, 15)
    .map(i => `- ${i}`)
    .join('\n')

  const dataContext = jsonStringify(
    {
      sessions: data.total_sessions,
      analyzed: data.sessions_with_facets,
      date_range: data.date_range,
      messages: data.total_messages,
      hours: Math.round(data.total_duration_hours),
      commits: data.git_commits,
      top_tools: Object.entries(data.tool_counts)
        .sort((a, b) => b[1] - a[1])
        .slice(0, 8),
      top_goals: Object.entries(data.goal_categories)
        .sort((a, b) => b[1] - a[1])
        .slice(0, 8),
      outcomes: data.outcomes,
      satisfaction: data.satisfaction,
      friction: data.friction,
      success: data.success,
      languages: data.languages,
    },
    null,
    2,
  )

  const fullContext =
    dataContext +
    '\n\nSESSION SUMMARIES:\n' +
    facetSummaries +
    '\n\nFRICTION DETAILS:\n' +
    frictionDetails +
    '\n\nUSER INSTRUCTIONS TO CLAUDE:\n' +
    (userInstructions || 'None captured')

  // Run sections in parallel first (excluding at_a_glance)
  const results = await Promise.all(
    INSIGHT_SECTIONS.map(section =>
      generateSectionInsight(section, fullContext),
    ),
  )

  // Combine results
  const insights: InsightResults = {}
  for (const { name, result } of results) {
    if (result) {
      ;(insights as Record<string, unknown>)[name] = result
    }
  }

  // Build rich context from generated sections for At a Glance
  const projectAreasText =
    (
      insights.project_areas as {
        areas?: Array<{ name: string; description: string }>
      }
    )?.areas
      ?.map(a => `- ${a.name}: ${a.description}`)
      .join('\n') || ''

  const bigWinsText =
    (
      insights.what_works as {
        impressive_workflows?: Array<{ title: string; description: string }>
      }
    )?.impressive_workflows
      ?.map(w => `- ${w.title}: ${w.description}`)
      .join('\n') || ''

  const frictionText =
    (
      insights.friction_analysis as {
        categories?: Array<{ category: string; description: string }>
      }
    )?.categories
      ?.map(c => `- ${c.category}: ${c.description}`)
      .join('\n') || ''

  const featuresText =
    (
      insights.suggestions as {
        features_to_try?: Array<{ feature: string; one_liner: string }>
      }
    )?.features_to_try
      ?.map(f => `- ${f.feature}: ${f.one_liner}`)
      .join('\n') || ''

  const patternsText =
    (
      insights.suggestions as {
        usage_patterns?: Array<{ title: string; suggestion: string }>
      }
    )?.usage_patterns
      ?.map(p => `- ${p.title}: ${p.suggestion}`)
      .join('\n') || ''

  const horizonText =
    (
      insights.on_the_horizon as {
        opportunities?: Array<{ title: string; whats_possible: string }>
      }
    )?.opportunities
      ?.map(o => `- ${o.title}: ${o.whats_possible}`)
      .join('\n') || ''

  // Now generate "At a Glance" with access to other sections' outputs
  const atAGlancePrompt = `You're writing an "At a Glance" summary for a Claude Code usage insights report for Claude Code users. The goal is to help them understand their usage and improve how they can use Claude better, especially as models improve.

Use this 4-part structure:

1. **What's working** - What is the user's unique style of interacting with Claude and what are some impactful things they've done? You can include one or two details, but keep it high level since things might not be fresh in the user's memory. Don't be fluffy or overly complimentary. Also, don't focus on the tool calls they use.

2. **What's hindering you** - Split into (a) Claude's fault (misunderstandings, wrong approaches, bugs) and (b) user-side friction (not providing enough context, environment issues -- ideally more general than just one project). Be honest but constructive.

3. **Quick wins to try** - Specific Claude Code features they could try from the examples below, or a workflow technique if you think it's really compelling. (Avoid stuff like "Ask Claude to confirm before taking actions" or "Type out more context up front" which are less compelling.)

4. **Ambitious workflows for better models** - As we move to much more capable models over the next 3-6 months, what should they prepare for? What workflows that seem impossible now will become possible? Draw from the appropriate section below.

Keep each section to 2-3 not-too-long sentences. Don't overwhelm the user. Don't mention specific numerical stats or underlined_categories from the session data below. Use a coaching tone.

RESPOND WITH ONLY A VALID JSON OBJECT:
{
  "whats_working": "(refer to instructions above)",
  "whats_hindering": "(refer to instructions above)",
  "quick_wins": "(refer to instructions above)",
  "ambitious_workflows": "(refer to instructions above)"
}

SESSION DATA:
${fullContext}

## Project Areas (what user works on)
${projectAreasText}

## Big Wins (impressive accomplishments)
${bigWinsText}

## Friction Categories (where things go wrong)
${frictionText}

## Features to Try
${featuresText}

## Usage Patterns to Adopt
${patternsText}

## On the Horizon (ambitious workflows for better models)
${horizonText}`

  const atAGlanceSection: InsightSection = {
    name: 'at_a_glance',
    prompt: atAGlancePrompt,
    maxTokens: 8192,
  }

  const atAGlanceResult = await generateSectionInsight(atAGlanceSection, '')
  if (atAGlanceResult.result) {
    insights.at_a_glance = atAGlanceResult.result as {
      whats_working?: string
      whats_hindering?: string
      quick_wins?: string
      ambitious_workflows?: string
    }
  }

  return insights
}

// Escape HTML but render **bold** as <strong>
function escapeHtmlWithBold(text: string): string {
  const escaped = escapeHtml(text)
  return escaped.replace(/\*\*(.+?)\*\*/g, '<strong>$1</strong>')
}

// Fixed orderings for specific charts (matching Python reference)
const SATISFACTION_ORDER = [
  'frustrated',
  'dissatisfied',
  'likely_satisfied',
  'satisfied',
  'happy',
  'unsure',
]

const OUTCOME_ORDER = [
  'not_achieved',
  'partially_achieved',
  'mostly_achieved',
  'fully_achieved',
  'unclear_from_transcript',
]

function generateBarChart(
  data: Record<string, number>,
  color: string,
  maxItems = 6,
  fixedOrder?: string[],
): string {
  let entries: [string, number][]

  if (fixedOrder) {
    // Use fixed order, only including items that exist in data
    entries = fixedOrder
      .filter(key => key in data && (data[key] ?? 0) > 0)
      .map(key => [key, data[key] ?? 0] as [string, number])
  } else {
    // Sort by count descending
    entries = Object.entries(data)
      .sort((a, b) => b[1] - a[1])
      .slice(0, maxItems)
  }

  if (entries.length === 0) return '<p class="empty">No data</p>'

  const maxVal = Math.max(...entries.map(e => e[1]))
  return entries
    .map(([label, count]) => {
      const pct = (count / maxVal) * 100
      // Use LABEL_MAP if available, otherwise clean up underscores and title case
      const cleanLabel =
        LABEL_MAP[label] ||
        label.replace(/_/g, ' ').replace(/\b\w/g, c => c.toUpperCase())
      return `<div class="bar-row">
        <div class="bar-label">${escapeHtml(cleanLabel)}</div>
        <div class="bar-track"><div class="bar-fill" style="width:${pct}%;background:${color}"></div></div>
        <div class="bar-value">${count}</div>
      </div>`
    })
    .join('\n')
}

function generateResponseTimeHistogram(times: number[]): string {
  if (times.length === 0) return '<p class="empty">No response time data</p>'

  // Create buckets (matching Python reference)
  const buckets: Record<string, number> = {
    '2-10s': 0,
    '10-30s': 0,
    '30s-1m': 0,
    '1-2m': 0,
    '2-5m': 0,
    '5-15m': 0,
    '>15m': 0,
  }

  for (const t of times) {
    if (t < 10) buckets['2-10s'] = (buckets['2-10s'] ?? 0) + 1
    else if (t < 30) buckets['10-30s'] = (buckets['10-30s'] ?? 0) + 1
    else if (t < 60) buckets['30s-1m'] = (buckets['30s-1m'] ?? 0) + 1
    else if (t < 120) buckets['1-2m'] = (buckets['1-2m'] ?? 0) + 1
    else if (t < 300) buckets['2-5m'] = (buckets['2-5m'] ?? 0) + 1
    else if (t < 900) buckets['5-15m'] = (buckets['5-15m'] ?? 0) + 1
    else buckets['>15m'] = (buckets['>15m'] ?? 0) + 1
  }

  const maxVal = Math.max(...Object.values(buckets))
  if (maxVal === 0) return '<p class="empty">No response time data</p>'

  return Object.entries(buckets)
    .map(([label, count]) => {
      const pct = (count / maxVal) * 100
      return `<div class="bar-row">
        <div class="bar-label">${label}</div>
        <div class="bar-track"><div class="bar-fill" style="width:${pct}%;background:#6366f1"></div></div>
        <div class="bar-value">${count}</div>
      </div>`
    })
    .join('\n')
}

function generateTimeOfDayChart(messageHours: number[]): string {
  if (messageHours.length === 0) return '<p class="empty">No time data</p>'

  // Group into time periods
  const periods = [
    { label: 'Morning (6-12)', range: [6, 7, 8, 9, 10, 11] },
    { label: 'Afternoon (12-18)', range: [12, 13, 14, 15, 16, 17] },
    { label: 'Evening (18-24)', range: [18, 19, 20, 21, 22, 23] },
    { label: 'Night (0-6)', range: [0, 1, 2, 3, 4, 5] },
  ]

  const hourCounts: Record<number, number> = {}
  for (const h of messageHours) {
    hourCounts[h] = (hourCounts[h] || 0) + 1
  }

  const periodCounts = periods.map(p => ({
    label: p.label,
    count: p.range.reduce((sum, h) => sum + (hourCounts[h] || 0), 0),
  }))

  const maxVal = Math.max(...periodCounts.map(p => p.count)) || 1

  const barsHtml = periodCounts
    .map(
      p => `
      <div class="bar-row">
        <div class="bar-label">${p.label}</div>
        <div class="bar-track"><div class="bar-fill" style="width:${(p.count / maxVal) * 100}%;background:#8b5cf6"></div></div>
        <div class="bar-value">${p.count}</div>
      </div>`,
    )
    .join('\n')

  return `<div id="hour-histogram">${barsHtml}</div>`
}

function getHourCountsJson(messageHours: number[]): string {
  const hourCounts: Record<number, number> = {}
  for (const h of messageHours) {
    hourCounts[h] = (hourCounts[h] || 0) + 1
  }
  return jsonStringify(hourCounts)
}

function generateHtmlReport(
  data: AggregatedData,
  insights: InsightResults,
): string {
  const markdownToHtml = (md: string): string => {
    if (!md) return ''
    return md
      .split('\n\n')
      .map(p => {
        let html = escapeHtml(p)
        html = html.replace(/\*\*(.+?)\*\*/g, '<strong>$1</strong>')
        html = html.replace(/^- /gm, '• ')
        html = html.replace(/\n/g, '<br>')
        return `<p>${html}</p>`
      })
      .join('\n')
  }

  // Build At a Glance section (new 4-part format with links to sections)
  const atAGlance = insights.at_a_glance
  const atAGlanceHtml = atAGlance
    ? `
    <div class="at-a-glance">
      <div class="glance-title">At a Glance</div>
      <div class="glance-sections">
        ${atAGlance.whats_working ? `<div class="glance-section"><strong>What's working:</strong> ${escapeHtmlWithBold(atAGlance.whats_working)} <a href="#section-wins" class="see-more">Impressive Things You Did →</a></div>` : ''}
        ${atAGlance.whats_hindering ? `<div class="glance-section"><strong>What's hindering you:</strong> ${escapeHtmlWithBold(atAGlance.whats_hindering)} <a href="#section-friction" class="see-more">Where Things Go Wrong →</a></div>` : ''}
        ${atAGlance.quick_wins ? `<div class="glance-section"><strong>Quick wins to try:</strong> ${escapeHtmlWithBold(atAGlance.quick_wins)} <a href="#section-features" class="see-more">Features to Try →</a></div>` : ''}
        ${atAGlance.ambitious_workflows ? `<div class="glance-section"><strong>Ambitious workflows:</strong> ${escapeHtmlWithBold(atAGlance.ambitious_workflows)} <a href="#section-horizon" class="see-more">On the Horizon →</a></div>` : ''}
      </div>
    </div>
    `
    : ''

  // Build project areas section
  const projectAreas = insights.project_areas?.areas || []
  const projectAreasHtml =
    projectAreas.length > 0
      ? `
    <h2 id="section-work">What You Work On</h2>
    <div class="project-areas">
      ${projectAreas
        .map(
          area => `
        <div class="project-area">
          <div class="area-header">
            <span class="area-name">${escapeHtml(area.name)}</span>
            <span class="area-count">~${area.session_count} sessions</span>
          </div>
          <div class="area-desc">${escapeHtml(area.description)}</div>
        </div>
      `,
        )
        .join('')}
    </div>
    `
      : ''

  // Build interaction style section
  const interactionStyle = insights.interaction_style
  const interactionHtml = interactionStyle?.narrative
    ? `
    <h2 id="section-usage">How You Use Claude Code</h2>
    <div class="narrative">
      ${markdownToHtml(interactionStyle.narrative)}
      ${interactionStyle.key_pattern ? `<div class="key-insight"><strong>Key pattern:</strong> ${escapeHtml(interactionStyle.key_pattern)}</div>` : ''}
    </div>
    `
    : ''

  // Build what works section
  const whatWorks = insights.what_works
  const whatWorksHtml =
    whatWorks?.impressive_workflows && whatWorks.impressive_workflows.length > 0
      ? `
    <h2 id="section-wins">Impressive Things You Did</h2>
    ${whatWorks.intro ? `<p class="section-intro">${escapeHtml(whatWorks.intro)}</p>` : ''}
    <div class="big-wins">
      ${whatWorks.impressive_workflows
        .map(
          wf => `
        <div class="big-win">
          <div class="big-win-title">${escapeHtml(wf.title || '')}</div>
          <div class="big-win-desc">${escapeHtml(wf.description || '')}</div>
        </div>
      `,
        )
        .join('')}
    </div>
    `
      : ''

  // Build friction section
  const frictionAnalysis = insights.friction_analysis
  const frictionHtml =
    frictionAnalysis?.categories && frictionAnalysis.categories.length > 0
      ? `
    <h2 id="section-friction">Where Things Go Wrong</h2>
    ${frictionAnalysis.intro ? `<p class="section-intro">${escapeHtml(frictionAnalysis.intro)}</p>` : ''}
    <div class="friction-categories">
      ${frictionAnalysis.categories
        .map(
          cat => `
        <div class="friction-category">
          <div class="friction-title">${escapeHtml(cat.category || '')}</div>
          <div class="friction-desc">${escapeHtml(cat.description || '')}</div>
          ${cat.examples ? `<ul class="friction-examples">${cat.examples.map(ex => `<li>${escapeHtml(ex)}</li>`).join('')}</ul>` : ''}
        </div>
      `,
        )
        .join('')}
    </div>
    `
      : ''

  // Build suggestions section
  const suggestions = insights.suggestions
  const suggestionsHtml = suggestions
    ? `
    ${
      suggestions.claude_md_additions &&
      suggestions.claude_md_additions.length > 0
        ? `
    <h2 id="section-features">Existing CC Features to Try</h2>
    <div class="claude-md-section">
      <h3>Suggested CLAUDE.md Additions</h3>
      <p style="font-size: 12px; color: #64748b; margin-bottom: 12px;">Just copy this into Claude Code to add it to your CLAUDE.md.</p>
      <div class="claude-md-actions">
        <button class="copy-all-btn" onclick="copyAllCheckedClaudeMd()">Copy All Checked</button>
      </div>
      ${suggestions.claude_md_additions
        .map(
          (add, i) => `
        <div class="claude-md-item">
          <input type="checkbox" id="cmd-${i}" class="cmd-checkbox" checked data-text="${escapeHtml(add.prompt_scaffold || add.where || 'Add to CLAUDE.md')}\\n\\n${escapeHtml(add.addition)}">
          <label for="cmd-${i}">
            <code class="cmd-code">${escapeHtml(add.addition)}</code>
            <button class="copy-btn" onclick="copyCmdItem(${i})">Copy</button>
          </label>
          <div class="cmd-why">${escapeHtml(add.why)}</div>
        </div>
      `,
        )
        .join('')}
    </div>
    `
        : ''
    }
    ${
      suggestions.features_to_try && suggestions.features_to_try.length > 0
        ? `
    <p style="font-size: 13px; color: #64748b; margin-bottom: 12px;">Just copy this into Claude Code and it'll set it up for you.</p>
    <div class="features-section">
      ${suggestions.features_to_try
        .map(
          feat => `
        <div class="feature-card">
          <div class="feature-title">${escapeHtml(feat.feature || '')}</div>
          <div class="feature-oneliner">${escapeHtml(feat.one_liner || '')}</div>
          <div class="feature-why"><strong>Why for you:</strong> ${escapeHtml(feat.why_for_you || '')}</div>
          ${
            feat.example_code
              ? `
          <div class="feature-examples">
            <div class="feature-example">
              <div class="example-code-row">
                <code class="example-code">${escapeHtml(feat.example_code)}</code>
                <button class="copy-btn" onclick="copyText(this)">Copy</button>
              </div>
            </div>
          </div>
          `
              : ''
          }
        </div>
      `,
        )
        .join('')}
    </div>
    `
        : ''
    }
    ${
      suggestions.usage_patterns && suggestions.usage_patterns.length > 0
        ? `
    <h2 id="section-patterns">New Ways to Use Claude Code</h2>
    <p style="font-size: 13px; color: #64748b; margin-bottom: 12px;">Just copy this into Claude Code and it'll walk you through it.</p>
    <div class="patterns-section">
      ${suggestions.usage_patterns
        .map(
          pat => `
        <div class="pattern-card">
          <div class="pattern-title">${escapeHtml(pat.title || '')}</div>
          <div class="pattern-summary">${escapeHtml(pat.suggestion || '')}</div>
          ${pat.detail ? `<div class="pattern-detail">${escapeHtml(pat.detail)}</div>` : ''}
          ${
            pat.copyable_prompt
              ? `
          <div class="copyable-prompt-section">
            <div class="prompt-label">Paste into Claude Code:</div>
            <div class="copyable-prompt-row">
              <code class="copyable-prompt">${escapeHtml(pat.copyable_prompt)}</code>
              <button class="copy-btn" onclick="copyText(this)">Copy</button>
            </div>
          </div>
          `
              : ''
          }
        </div>
      `,
        )
        .join('')}
    </div>
    `
        : ''
    }
    `
    : ''

  // Build On the Horizon section
  const horizonData = insights.on_the_horizon
  const horizonHtml =
    horizonData?.opportunities && horizonData.opportunities.length > 0
      ? `
    <h2 id="section-horizon">On the Horizon</h2>
    ${horizonData.intro ? `<p class="section-intro">${escapeHtml(horizonData.intro)}</p>` : ''}
    <div class="horizon-section">
      ${horizonData.opportunities
        .map(
          opp => `
        <div class="horizon-card">
          <div class="horizon-title">${escapeHtml(opp.title || '')}</div>
          <div class="horizon-possible">${escapeHtml(opp.whats_possible || '')}</div>
          ${opp.how_to_try ? `<div class="horizon-tip"><strong>Getting started:</strong> ${escapeHtml(opp.how_to_try)}</div>` : ''}
          ${opp.copyable_prompt ? `<div class="pattern-prompt"><div class="prompt-label">Paste into Claude Code:</div><code>${escapeHtml(opp.copyable_prompt)}</code><button class="copy-btn" onclick="copyText(this)">Copy</button></div>` : ''}
        </div>
      `,
        )
        .join('')}
    </div>
    `
      : ''

  // Build Team Feedback section (collapsible, ant-only)
  const ccImprovements =
    process.env.USER_TYPE === 'ant'
      ? insights.cc_team_improvements?.improvements || []
      : []
  const modelImprovements =
    process.env.USER_TYPE === 'ant'
      ? insights.model_behavior_improvements?.improvements || []
      : []
  const teamFeedbackHtml =
    ccImprovements.length > 0 || modelImprovements.length > 0
      ? `
    <h2 id="section-feedback" class="feedback-header">Closing the Loop: Feedback for Other Teams</h2>
    <p class="feedback-intro">Suggestions for the CC product and model teams based on your usage patterns. Click to expand.</p>
    ${
      ccImprovements.length > 0
        ? `
    <div class="collapsible-section">
      <div class="collapsible-header" onclick="toggleCollapsible(this)">
        <span class="collapsible-arrow">▶</span>
        <h3>Product Improvements for CC Team</h3>
      </div>
      <div class="collapsible-content">
        <div class="suggestions-section">
          ${ccImprovements
            .map(
              imp => `
            <div class="feedback-card team-card">
              <div class="feedback-title">${escapeHtml(imp.title || '')}</div>
              <div class="feedback-detail">${escapeHtml(imp.detail || '')}</div>
              ${imp.evidence ? `<div class="feedback-evidence"><em>Evidence:</em> ${escapeHtml(imp.evidence)}</div>` : ''}
            </div>
          `,
            )
            .join('')}
        </div>
      </div>
    </div>
    `
        : ''
    }
    ${
      modelImprovements.length > 0
        ? `
    <div class="collapsible-section">
      <div class="collapsible-header" onclick="toggleCollapsible(this)">
        <span class="collapsible-arrow">▶</span>
        <h3>Model Behavior Improvements</h3>
      </div>
      <div class="collapsible-content">
        <div class="suggestions-section">
          ${modelImprovements
            .map(
              imp => `
            <div class="feedback-card model-card">
              <div class="feedback-title">${escapeHtml(imp.title || '')}</div>
              <div class="feedback-detail">${escapeHtml(imp.detail || '')}</div>
              ${imp.evidence ? `<div class="feedback-evidence"><em>Evidence:</em> ${escapeHtml(imp.evidence)}</div>` : ''}
            </div>
          `,
            )
            .join('')}
        </div>
      </div>
    </div>
    `
        : ''
    }
    `
      : ''

  // Build Fun Ending section
  const funEnding = insights.fun_ending
  const funEndingHtml = funEnding?.headline
    ? `
    <div class="fun-ending">
      <div class="fun-headline">"${escapeHtml(funEnding.headline)}"</div>
      ${funEnding.detail ? `<div class="fun-detail">${escapeHtml(funEnding.detail)}</div>` : ''}
    </div>
    `
    : ''

  const css = `
    * { box-sizing: border-box; margin: 0; padding: 0; }
    body { font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif; background: #f8fafc; color: #334155; line-height: 1.65; padding: 48px 24px; }
    .container { max-width: 800px; margin: 0 auto; }
    h1 { font-size: 32px; font-weight: 700; color: #0f172a; margin-bottom: 8px; }
    h2 { font-size: 20px; font-weight: 600; color: #0f172a; margin-top: 48px; margin-bottom: 16px; }
    .subtitle { color: #64748b; font-size: 15px; margin-bottom: 32px; }
    .nav-toc { display: flex; flex-wrap: wrap; gap: 8px; margin: 24px 0 32px 0; padding: 16px; background: white; border-radius: 8px; border: 1px solid #e2e8f0; }
    .nav-toc a { font-size: 12px; color: #64748b; text-decoration: none; padding: 6px 12px; border-radius: 6px; background: #f1f5f9; transition: all 0.15s; }
    .nav-toc a:hover { background: #e2e8f0; color: #334155; }
    .stats-row { display: flex; gap: 24px; margin-bottom: 40px; padding: 20px 0; border-top: 1px solid #e2e8f0; border-bottom: 1px solid #e2e8f0; flex-wrap: wrap; }
    .stat { text-align: center; }
    .stat-value { font-size: 24px; font-weight: 700; color: #0f172a; }
    .stat-label { font-size: 11px; color: #64748b; text-transform: uppercase; }
    .at-a-glance { background: linear-gradient(135deg, #fef3c7 0%, #fde68a 100%); border: 1px solid #f59e0b; border-radius: 12px; padding: 20px 24px; margin-bottom: 32px; }
    .glance-title { font-size: 16px; font-weight: 700; color: #92400e; margin-bottom: 16px; }
    .glance-sections { display: flex; flex-direction: column; gap: 12px; }
    .glance-section { font-size: 14px; color: #78350f; line-height: 1.6; }
    .glance-section strong { color: #92400e; }
    .see-more { color: #b45309; text-decoration: none; font-size: 13px; white-space: nowrap; }
    .see-more:hover { text-decoration: underline; }
    .project-areas { display: flex; flex-direction: column; gap: 12px; margin-bottom: 32px; }
    .project-area { background: white; border: 1px solid #e2e8f0; border-radius: 8px; padding: 16px; }
    .area-header { display: flex; justify-content: space-between; align-items: center; margin-bottom: 8px; }
    .area-name { font-weight: 600; font-size: 15px; color: #0f172a; }
    .area-count { font-size: 12px; color: #64748b; background: #f1f5f9; padding: 2px 8px; border-radius: 4px; }
    .area-desc { font-size: 14px; color: #475569; line-height: 1.5; }
    .narrative { background: white; border: 1px solid #e2e8f0; border-radius: 8px; padding: 20px; margin-bottom: 24px; }
    .narrative p { margin-bottom: 12px; font-size: 14px; color: #475569; line-height: 1.7; }
    .key-insight { background: #f0fdf4; border: 1px solid #bbf7d0; border-radius: 8px; padding: 12px 16px; margin-top: 12px; font-size: 14px; color: #166534; }
    .section-intro { font-size: 14px; color: #64748b; margin-bottom: 16px; }
    .big-wins { display: flex; flex-direction: column; gap: 12px; margin-bottom: 24px; }
    .big-win { background: #f0fdf4; border: 1px solid #bbf7d0; border-radius: 8px; padding: 16px; }
    .big-win-title { font-weight: 600; font-size: 15px; color: #166534; margin-bottom: 8px; }
    .big-win-desc { font-size: 14px; color: #15803d; line-height: 1.5; }
    .friction-categories { display: flex; flex-direction: column; gap: 16px; margin-bottom: 24px; }
    .friction-category { background: #fef2f2; border: 1px solid #fca5a5; border-radius: 8px; padding: 16px; }
    .friction-title { font-weight: 600; font-size: 15px; color: #991b1b; margin-bottom: 6px; }
    .friction-desc { font-size: 13px; color: #7f1d1d; margin-bottom: 10px; }
    .friction-examples { margin: 0 0 0 20px; font-size: 13px; color: #334155; }
    .friction-examples li { margin-bottom: 4px; }
    .claude-md-section { background: #eff6ff; border: 1px solid #bfdbfe; border-radius: 8px; padding: 16px; margin-bottom: 20px; }
    .claude-md-section h3 { font-size: 14px; font-weight: 600; color: #1e40af; margin: 0 0 12px 0; }
    .claude-md-actions { margin-bottom: 12px; padding-bottom: 12px; border-bottom: 1px solid #dbeafe; }
    .copy-all-btn { background: #2563eb; color: white; border: none; border-radius: 4px; padding: 6px 12px; font-size: 12px; cursor: pointer; font-weight: 500; transition: all 0.2s; }
    .copy-all-btn:hover { background: #1d4ed8; }
    .copy-all-btn.copied { background: #16a34a; }
    .claude-md-item { display: flex; flex-wrap: wrap; align-items: flex-start; gap: 8px; padding: 10px 0; border-bottom: 1px solid #dbeafe; }
    .claude-md-item:last-child { border-bottom: none; }
    .cmd-checkbox { margin-top: 2px; }
    .cmd-code { background: white; padding: 8px 12px; border-radius: 4px; font-size: 12px; color: #1e40af; border: 1px solid #bfdbfe; font-family: monospace; display: block; white-space: pre-wrap; word-break: break-word; flex: 1; }
    .cmd-why { font-size: 12px; color: #64748b; width: 100%; padding-left: 24px; margin-top: 4px; }
    .features-section, .patterns-section { display: flex; flex-direction: column; gap: 12px; margin: 16px 0; }
    .feature-card { background: #f0fdf4; border: 1px solid #86efac; border-radius: 8px; padding: 16px; }
    .pattern-card { background: #f0f9ff; border: 1px solid #7dd3fc; border-radius: 8px; padding: 16px; }
    .feature-title, .pattern-title { font-weight: 600; font-size: 15px; color: #0f172a; margin-bottom: 6px; }
    .feature-oneliner { font-size: 14px; color: #475569; margin-bottom: 8px; }
    .pattern-summary { font-size: 14px; color: #475569; margin-bottom: 8px; }
    .feature-why, .pattern-detail { font-size: 13px; color: #334155; line-height: 1.5; }
    .feature-examples { margin-top: 12px; }
    .feature-example { padding: 8px 0; border-top: 1px solid #d1fae5; }
    .feature-example:first-child { border-top: none; }
    .example-desc { font-size: 13px; color: #334155; margin-bottom: 6px; }
    .example-code-row { display: flex; align-items: flex-start; gap: 8px; }
    .example-code { flex: 1; background: #f1f5f9; padding: 8px 12px; border-radius: 4px; font-family: monospace; font-size: 12px; color: #334155; overflow-x: auto; white-space: pre-wrap; }
    .copyable-prompt-section { margin-top: 12px; padding-top: 12px; border-top: 1px solid #e2e8f0; }
    .copyable-prompt-row { display: flex; align-items: flex-start; gap: 8px; }
    .copyable-prompt { flex: 1; background: #f8fafc; padding: 10px 12px; border-radius: 4px; font-family: monospace; font-size: 12px; color: #334155; border: 1px solid #e2e8f0; white-space: pre-wrap; line-height: 1.5; }
    .feature-code { background: #f8fafc; padding: 12px; border-radius: 6px; margin-top: 12px; border: 1px solid #e2e8f0; display: flex; align-items: flex-start; gap: 8px; }
    .feature-code code { flex: 1; font-family: monospace; font-size: 12px; color: #334155; white-space: pre-wrap; }
    .pattern-prompt { background: #f8fafc; padding: 12px; border-radius: 6px; margin-top: 12px; border: 1px solid #e2e8f0; }
    .pattern-prompt code { font-family: monospace; font-size: 12px; color: #334155; display: block; white-space: pre-wrap; margin-bottom: 8px; }
    .prompt-label { font-size: 11px; font-weight: 600; text-transform: uppercase; color: #64748b; margin-bottom: 6px; }
    .copy-btn { background: #e2e8f0; border: none; border-radius: 4px; padding: 4px 8px; font-size: 11px; cursor: pointer; color: #475569; flex-shrink: 0; }
    .copy-btn:hover { background: #cbd5e1; }
    .charts-row { display: grid; grid-template-columns: 1fr 1fr; gap: 24px; margin: 24px 0; }
    .chart-card { background: white; border: 1px solid #e2e8f0; border-radius: 8px; padding: 16px; }
    .chart-title { font-size: 12px; font-weight: 600; color: #64748b; text-transform: uppercase; margin-bottom: 12px; }
    .bar-row { display: flex; align-items: center; margin-bottom: 6px; }
    .bar-label { width: 100px; font-size: 11px; color: #475569; flex-shrink: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
    .bar-track { flex: 1; height: 6px; background: #f1f5f9; border-radius: 3px; margin: 0 8px; }
    .bar-fill { height: 100%; border-radius: 3px; }
    .bar-value { width: 28px; font-size: 11px; font-weight: 500; color: #64748b; text-align: right; }
    .empty { color: #94a3b8; font-size: 13px; }
    .horizon-section { display: flex; flex-direction: column; gap: 16px; }
    .horizon-card { background: linear-gradient(135deg, #faf5ff 0%, #f5f3ff 100%); border: 1px solid #c4b5fd; border-radius: 8px; padding: 16px; }
    .horizon-title { font-weight: 600; font-size: 15px; color: #5b21b6; margin-bottom: 8px; }
    .horizon-possible { font-size: 14px; color: #334155; margin-bottom: 10px; line-height: 1.5; }
    .horizon-tip { font-size: 13px; color: #6b21a8; background: rgba(255,255,255,0.6); padding: 8px 12px; border-radius: 4px; }
    .feedback-header { margin-top: 48px; color: #64748b; font-size: 16px; }
    .feedback-intro { font-size: 13px; color: #94a3b8; margin-bottom: 16px; }
    .feedback-section { margin-top: 16px; }
    .feedback-section h3 { font-size: 14px; font-weight: 600; color: #475569; margin-bottom: 12px; }
    .feedback-card { background: white; border: 1px solid #e2e8f0; border-radius: 8px; padding: 16px; margin-bottom: 12px; }
    .feedback-card.team-card { background: #eff6ff; border-color: #bfdbfe; }
    .feedback-card.model-card { background: #faf5ff; border-color: #e9d5ff; }
    .feedback-title { font-weight: 600; font-size: 14px; color: #0f172a; margin-bottom: 6px; }
    .feedback-detail { font-size: 13px; color: #475569; line-height: 1.5; }
    .feedback-evidence { font-size: 12px; color: #64748b; margin-top: 8px; }
    .fun-ending { background: linear-gradient(135deg, #fef3c7 0%, #fde68a 100%); border: 1px solid #fbbf24; border-radius: 12px; padding: 24px; margin-top: 40px; text-align: center; }
    .fun-headline { font-size: 18px; font-weight: 600; color: #78350f; margin-bottom: 8px; }
    .fun-detail { font-size: 14px; color: #92400e; }
    .collapsible-section { margin-top: 16px; }
    .collapsible-header { display: flex; align-items: center; gap: 8px; cursor: pointer; padding: 12px 0; border-bottom: 1px solid #e2e8f0; }
    .collapsible-header h3 { margin: 0; font-size: 14px; font-weight: 600; color: #475569; }
    .collapsible-arrow { font-size: 12px; color: #94a3b8; transition: transform 0.2s; }
    .collapsible-content { display: none; padding-top: 16px; }
    .collapsible-content.open { display: block; }
    .collapsible-header.open .collapsible-arrow { transform: rotate(90deg); }
    @media (max-width: 640px) { .charts-row { grid-template-columns: 1fr; } .stats-row { justify-content: center; } }
  `

  const hourCountsJson = getHourCountsJson(data.message_hours)

  const js = `
    function toggleCollapsible(header) {
      header.classList.toggle('open');
      const content = header.nextElementSibling;
      content.classList.toggle('open');
    }
    function copyText(btn) {
      const code = btn.previousElementSibling;
      navigator.clipboard.writeText(code.textContent).then(() => {
        btn.textContent = 'Copied!';
        setTimeout(() => { btn.textContent = 'Copy'; }, 2000);
      });
    }
    function copyCmdItem(idx) {
      const checkbox = document.getElementById('cmd-' + idx);
      if (checkbox) {
        const text = checkbox.dataset.text;
        navigator.clipboard.writeText(text).then(() => {
          const btn = checkbox.nextElementSibling.querySelector('.copy-btn');
          if (btn) { btn.textContent = 'Copied!'; setTimeout(() => { btn.textContent = 'Copy'; }, 2000); }
        });
      }
    }
    function copyAllCheckedClaudeMd() {
      const checkboxes = document.querySelectorAll('.cmd-checkbox:checked');
      const texts = [];
      checkboxes.forEach(cb => {
        if (cb.dataset.text) { texts.push(cb.dataset.text); }
      });
      const combined = texts.join('\\n');
      const btn = document.querySelector('.copy-all-btn');
      if (btn) {
        navigator.clipboard.writeText(combined).then(() => {
          btn.textContent = 'Copied ' + texts.length + ' items!';
          btn.classList.add('copied');
          setTimeout(() => { btn.textContent = 'Copy All Checked'; btn.classList.remove('copied'); }, 2000);
        });
      }
    }
    // Timezone selector for time of day chart (data is from our own analytics, not user input)
    const rawHourCounts = ${hourCountsJson};
    function updateHourHistogram(offsetFromPT) {
      const periods = [
        { label: "Morning (6-12)", range: [6,7,8,9,10,11] },
        { label: "Afternoon (12-18)", range: [12,13,14,15,16,17] },
        { label: "Evening (18-24)", range: [18,19,20,21,22,23] },
        { label: "Night (0-6)", range: [0,1,2,3,4,5] }
      ];
      const adjustedCounts = {};
      for (const [hour, count] of Object.entries(rawHourCounts)) {
        const newHour = (parseInt(hour) + offsetFromPT + 24) % 24;
        adjustedCounts[newHour] = (adjustedCounts[newHour] || 0) + count;
      }
      const periodCounts = periods.map(p => ({
        label: p.label,
        count: p.range.reduce((sum, h) => sum + (adjustedCounts[h] || 0), 0)
      }));
      const maxCount = Math.max(...periodCounts.map(p => p.count)) || 1;
      const container = document.getElementById('hour-histogram');
      container.textContent = '';
      periodCounts.forEach(p => {
        const row = document.createElement('div');
        row.className = 'bar-row';
        const label = document.createElement('div');
        label.className = 'bar-label';
        label.textContent = p.label;
        const track = document.createElement('div');
        track.className = 'bar-track';
        const fill = document.createElement('div');
        fill.className = 'bar-fill';
        fill.style.width = (p.count / maxCount) * 100 + '%';
        fill.style.background = '#8b5cf6';
        track.appendChild(fill);
        const value = document.createElement('div');
        value.className = 'bar-value';
        value.textContent = p.count;
        row.appendChild(label);
        row.appendChild(track);
        row.appendChild(value);
        container.appendChild(row);
      });
    }
    document.getElementById('timezone-select').addEventListener('change', function() {
      const customInput = document.getElementById('custom-offset');
      if (this.value === 'custom') {
        customInput.style.display = 'inline-block';
        customInput.focus();
      } else {
        customInput.style.display = 'none';
        updateHourHistogram(parseInt(this.value));
      }
    });
    document.getElementById('custom-offset').addEventListener('change', function() {
      const offset = parseInt(this.value) + 8;
      updateHourHistogram(offset);
    });
  `

  return `<!DOCTYPE html>
<html>
<head>
  <meta charset="utf-8">
  <title>Claude Code Insights</title>
  <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap" rel="stylesheet">
  <style>${css}</style>
</head>
<body>
  <div class="container">
    <h1>Claude Code Insights</h1>
    <p class="subtitle">${data.total_messages.toLocaleString()} messages across ${data.total_sessions} sessions${data.total_sessions_scanned && data.total_sessions_scanned > data.total_sessions ? ` (${data.total_sessions_scanned.toLocaleString()} total)` : ''} | ${data.date_range.start} to ${data.date_range.end}</p>

    ${atAGlanceHtml}

    <nav class="nav-toc">
      <a href="#section-work">What You Work On</a>
      <a href="#section-usage">How You Use CC</a>
      <a href="#section-wins">Impressive Things</a>
      <a href="#section-friction">Where Things Go Wrong</a>
      <a href="#section-features">Features to Try</a>
      <a href="#section-patterns">New Usage Patterns</a>
      <a href="#section-horizon">On the Horizon</a>
      <a href="#section-feedback">Team Feedback</a>
    </nav>

    <div class="stats-row">
      <div class="stat"><div class="stat-value">${data.total_messages.toLocaleString()}</div><div class="stat-label">Messages</div></div>
      <div class="stat"><div class="stat-value">+${data.total_lines_added.toLocaleString()}/-${data.total_lines_removed.toLocaleString()}</div><div class="stat-label">Lines</div></div>
      <div class="stat"><div class="stat-value">${data.total_files_modified}</div><div class="stat-label">Files</div></div>
      <div class="stat"><div class="stat-value">${data.days_active}</div><div class="stat-label">Days</div></div>
      <div class="stat"><div class="stat-value">${data.messages_per_day}</div><div class="stat-label">Msgs/Day</div></div>
    </div>

    ${projectAreasHtml}

    <div class="charts-row">
      <div class="chart-card">
        <div class="chart-title">What You Wanted</div>
        ${generateBarChart(data.goal_categories, '#2563eb')}
      </div>
      <div class="chart-card">
        <div class="chart-title">Top Tools Used</div>
        ${generateBarChart(data.tool_counts, '#0891b2')}
      </div>
    </div>

    <div class="charts-row">
      <div class="chart-card">
        <div class="chart-title">Languages</div>
        ${generateBarChart(data.languages, '#10b981')}
      </div>
      <div class="chart-card">
        <div class="chart-title">Session Types</div>
        ${generateBarChart(data.session_types || {}, '#8b5cf6')}
      </div>
    </div>

    ${interactionHtml}

    <!-- Response Time Distribution -->
    <div class="chart-card" style="margin: 24px 0;">
      <div class="chart-title">User Response Time Distribution</div>
      ${generateResponseTimeHistogram(data.user_response_times)}
      <div style="font-size: 12px; color: #64748b; margin-top: 8px;">
        Median: ${data.median_response_time.toFixed(1)}s &bull; Average: ${data.avg_response_time.toFixed(1)}s
      </div>
    </div>

    <!-- Multi-clauding Section (matching Python reference) -->
    <div class="chart-card" style="margin: 24px 0;">
      <div class="chart-title">Multi-Clauding (Parallel Sessions)</div>
      ${
        data.multi_clauding.overlap_events === 0
          ? `
        <p style="font-size: 14px; color: #64748b; padding: 8px 0;">
          No parallel session usage detected. You typically work with one Claude Code session at a time.
        </p>
      `
          : `
        <div style="display: flex; gap: 24px; margin: 12px 0;">
          <div style="text-align: center;">
            <div style="font-size: 24px; font-weight: 700; color: #7c3aed;">${data.multi_clauding.overlap_events}</div>
            <div style="font-size: 11px; color: #64748b; text-transform: uppercase;">Overlap Events</div>
          </div>
          <div style="text-align: center;">
            <div style="font-size: 24px; font-weight: 700; color: #7c3aed;">${data.multi_clauding.sessions_involved}</div>
            <div style="font-size: 11px; color: #64748b; text-transform: uppercase;">Sessions Involved</div>
          </div>
          <div style="text-align: center;">
            <div style="font-size: 24px; font-weight: 700; color: #7c3aed;">${data.total_messages > 0 ? Math.round((100 * data.multi_clauding.user_messages_during) / data.total_messages) : 0}%</div>
            <div style="font-size: 11px; color: #64748b; text-transform: uppercase;">Of Messages</div>
          </div>
        </div>
        <p style="font-size: 13px; color: #475569; margin-top: 12px;">
          You run multiple Claude Code sessions simultaneously. Multi-clauding is detected when sessions
          overlap in time, suggesting parallel workflows.
        </p>
      `
      }
    </div>

    <!-- Time of Day & Tool Errors -->
    <div class="charts-row">
      <div class="chart-card">
        <div class="chart-title" style="display: flex; align-items: center; gap: 12px;">
          User Messages by Time of Day
          <select id="timezone-select" style="font-size: 12px; padding: 4px 8px; border-radius: 4px; border: 1px solid #e2e8f0;">
            <option value="0">PT (UTC-8)</option>
            <option value="3">ET (UTC-5)</option>
            <option value="8">London (UTC)</option>
            <option value="9">CET (UTC+1)</option>
            <option value="17">Tokyo (UTC+9)</option>
            <option value="custom">Custom offset...</option>
          </select>
          <input type="number" id="custom-offset" placeholder="UTC offset" style="display: none; width: 80px; font-size: 12px; padding: 4px; border-radius: 4px; border: 1px solid #e2e8f0;">
        </div>
        ${generateTimeOfDayChart(data.message_hours)}
      </div>
      <div class="chart-card">
        <div class="chart-title">Tool Errors Encountered</div>
        ${Object.keys(data.tool_error_categories).length > 0 ? generateBarChart(data.tool_error_categories, '#dc2626') : '<p class="empty">No tool errors</p>'}
      </div>
    </div>

    ${whatWorksHtml}

    <div class="charts-row">
      <div class="chart-card">
        <div class="chart-title">What Helped Most (Claude's Capabilities)</div>
        ${generateBarChart(data.success, '#16a34a')}
      </div>
      <div class="chart-card">
        <div class="chart-title">Outcomes</div>
        ${generateBarChart(data.outcomes, '#8b5cf6', 6, OUTCOME_ORDER)}
      </div>
    </div>

    ${frictionHtml}

    <div class="charts-row">
      <div class="chart-card">
        <div class="chart-title">Primary Friction Types</div>
        ${generateBarChart(data.friction, '#dc2626')}
      </div>
      <div class="chart-card">
        <div class="chart-title">Inferred Satisfaction (model-estimated)</div>
        ${generateBarChart(data.satisfaction, '#eab308', 6, SATISFACTION_ORDER)}
      </div>
    </div>

    ${suggestionsHtml}

    ${horizonHtml}

    ${funEndingHtml}

    ${teamFeedbackHtml}
  </div>
  <script>${js}</script>
</body>
</html>`
}

// ============================================================================
// Export Types & Functions
// ============================================================================

/**
 * Structured export format for claudescope consumption
 */
export type InsightsExport = {
  metadata: {
    username: string
    generated_at: string
    claude_code_version: string
    date_range: { start: string; end: string }
    session_count: number
    remote_hosts_collected?: string[]
  }
  aggregated_data: AggregatedData
  insights: InsightResults
  facets_summary?: {
    total: number
    goal_categories: Record<string, number>
    outcomes: Record<string, number>
    satisfaction: Record<string, number>
    friction: Record<string, number>
  }
}

/**
 * Build export data from already-computed values.
 * Used by background upload to S3.
 */
export function buildExportData(
  data: AggregatedData,
  insights: InsightResults,
  facets: Map<string, SessionFacets>,
  remoteStats?: { hosts: RemoteHostInfo[]; totalCopied: number },
): InsightsExport {
  const version = typeof MACRO !== 'undefined' ? MACRO.VERSION : 'unknown'

  const remote_hosts_collected = remoteStats?.hosts
    .filter(h => h.sessionCount > 0)
    .map(h => h.name)

  const facets_summary = {
    total: facets.size,
    goal_categories: {} as Record<string, number>,
    outcomes: {} as Record<string, number>,
    satisfaction: {} as Record<string, number>,
    friction: {} as Record<string, number>,
  }
  for (const f of facets.values()) {
    for (const [cat, count] of safeEntries(f.goal_categories)) {
      if (count > 0) {
        facets_summary.goal_categories[cat] =
          (facets_summary.goal_categories[cat] || 0) + count
      }
    }
    facets_summary.outcomes[f.outcome] =
      (facets_summary.outcomes[f.outcome] || 0) + 1
    for (const [level, count] of safeEntries(f.user_satisfaction_counts)) {
      if (count > 0) {
        facets_summary.satisfaction[level] =
          (facets_summary.satisfaction[level] || 0) + count
      }
    }
    for (const [type, count] of safeEntries(f.friction_counts)) {
      if (count > 0) {
        facets_summary.friction[type] =
          (facets_summary.friction[type] || 0) + count
      }
    }
  }

  return {
    metadata: {
      username: process.env.SAFEUSER || process.env.USER || 'unknown',
      generated_at: new Date().toISOString(),
      claude_code_version: version,
      date_range: data.date_range,
      session_count: data.total_sessions,
      ...(remote_hosts_collected &&
        remote_hosts_collected.length > 0 && {
          remote_hosts_collected,
        }),
    },
    aggregated_data: data,
    insights,
    facets_summary,
  }
}

// ============================================================================
// Lite Session Scanning
// ============================================================================

type LiteSessionInfo = {
  sessionId: string
  path: string
  mtime: number
  size: number
}

/**
 * Scans all project directories using filesystem metadata only (no JSONL parsing).
 * Returns a list of session file info sorted by mtime descending.
 * Yields to the event loop between project directories to keep the UI responsive.
 */
async function scanAllSessions(): Promise<LiteSessionInfo[]> {
  const projectsDir = getProjectsDir()

  let dirents: Awaited<ReturnType<typeof readdir>>
  try {
    dirents = await readdir(projectsDir, { withFileTypes: true })
  } catch {
    return []
  }

  const projectDirs = dirents
    .filter(dirent => dirent.isDirectory())
    .map(dirent => join(projectsDir, dirent.name))

  const allSessions: LiteSessionInfo[] = []

  for (let i = 0; i < projectDirs.length; i++) {
    const sessionFiles = await getSessionFilesWithMtime(projectDirs[i]!)
    for (const [sessionId, fileInfo] of sessionFiles) {
      allSessions.push({
        sessionId,
        path: fileInfo.path,
        mtime: fileInfo.mtime,
        size: fileInfo.size,
      })
    }
    // Yield to event loop every 10 project directories
    if (i % 10 === 9) {
      await new Promise<void>(resolve => setImmediate(resolve))
    }
  }

  // Sort by mtime descending (most recent first)
  allSessions.sort((a, b) => b.mtime - a.mtime)
  return allSessions
}

// ============================================================================
// Main Function
// ============================================================================

export async function generateUsageReport(options?: {
  collectRemote?: boolean
}): Promise<{
  insights: InsightResults
  htmlPath: string
  data: AggregatedData
  remoteStats?: { hosts: RemoteHostInfo[]; totalCopied: number }
  facets: Map<string, SessionFacets>
}> {
  let remoteStats: { hosts: RemoteHostInfo[]; totalCopied: number } | undefined

  // Optionally collect data from remote hosts first (ant-only)
  if (process.env.USER_TYPE === 'ant' && options?.collectRemote) {
    const destDir = join(getClaudeConfigHomeDir(), 'projects')
    const { hosts, totalCopied } = await collectAllRemoteHostData(destDir)
    remoteStats = { hosts, totalCopied }
  }

  // Phase 1: Lite scan — filesystem metadata only (no JSONL parsing)
  const allScannedSessions = await scanAllSessions()
  const totalSessionsScanned = allScannedSessions.length

  // Phase 2: Load SessionMeta — use cache where available, parse only uncached
  // Read cached metas in parallel batches to avoid blocking the event loop
  const META_BATCH_SIZE = 50
  const MAX_SESSIONS_TO_LOAD = 200
  let allMetas: SessionMeta[] = []
  const uncachedSessions: LiteSessionInfo[] = []

  for (let i = 0; i < allScannedSessions.length; i += META_BATCH_SIZE) {
    const batch = allScannedSessions.slice(i, i + META_BATCH_SIZE)
    const results = await Promise.all(
      batch.map(async sessionInfo => ({
        sessionInfo,
        cached: await loadCachedSessionMeta(sessionInfo.sessionId),
      })),
    )
    for (const { sessionInfo, cached } of results) {
      if (cached) {
        allMetas.push(cached)
      } else if (uncachedSessions.length < MAX_SESSIONS_TO_LOAD) {
        uncachedSessions.push(sessionInfo)
      }
    }
  }

  // Load full message data only for uncached sessions and compute SessionMeta
  const logsForFacets = new Map<string, LogOption>()

  // Filter out /insights meta-sessions (facet extraction API calls get logged as sessions)
  const isMetaSession = (log: LogOption): boolean => {
    for (const msg of log.messages.slice(0, 5)) {
      if (msg.type === 'user' && msg.message) {
        const content = msg.message.content
        if (typeof content === 'string') {
          if (
            content.includes('RESPOND WITH ONLY A VALID JSON OBJECT') ||
            content.includes('record_facets')
          ) {
            return true
          }
        }
      }
    }
    return false
  }

  // Load uncached sessions in batches to yield to event loop between batches
  const LOAD_BATCH_SIZE = 10
  for (let i = 0; i < uncachedSessions.length; i += LOAD_BATCH_SIZE) {
    const batch = uncachedSessions.slice(i, i + LOAD_BATCH_SIZE)
    const batchResults = await Promise.all(
      batch.map(async sessionInfo => {
        try {
          return await loadAllLogsFromSessionFile(sessionInfo.path)
        } catch {
          return []
        }
      }),
    )
    // Collect metas synchronously, then save them in parallel (independent writes)
    const metasToSave: SessionMeta[] = []
    for (const logs of batchResults) {
      for (const log of logs) {
        if (isMetaSession(log) || !hasValidDates(log)) continue
        const meta = logToSessionMeta(log)
        allMetas.push(meta)
        metasToSave.push(meta)
        // Keep the log around for potential facet extraction
        logsForFacets.set(meta.session_id, log)
      }
    }
    await Promise.all(metasToSave.map(meta => saveSessionMeta(meta)))
  }

  // Deduplicate session branches (keep the one with most user messages per session_id)
  // This prevents inflated totals when a session has multiple conversation branches
  const bestBySession = new Map<string, SessionMeta>()
  for (const meta of allMetas) {
    const existing = bestBySession.get(meta.session_id)
    if (
      !existing ||
      meta.user_message_count > existing.user_message_count ||
      (meta.user_message_count === existing.user_message_count &&
        meta.duration_minutes > existing.duration_minutes)
    ) {
      bestBySession.set(meta.session_id, meta)
    }
  }
  // Replace allMetas with deduplicated list and remove unused logs from logsForFacets
  const keptSessionIds = new Set(bestBySession.keys())
  allMetas = [...bestBySession.values()]
  for (const sessionId of logsForFacets.keys()) {
    if (!keptSessionIds.has(sessionId)) {
      logsForFacets.delete(sessionId)
    }
  }

  // Sort all metas by start_time descending (most recent first)
  allMetas.sort((a, b) => b.start_time.localeCompare(a.start_time))

  // Pre-filter obviously minimal sessions to save API calls
  // (matching Python's substantive filtering concept)
  const isSubstantiveSession = (meta: SessionMeta): boolean => {
    // Skip sessions with very few user messages
    if (meta.user_message_count < 2) return false
    // Skip very short sessions (< 1 minute)
    if (meta.duration_minutes < 1) return false
    return true
  }

  const substantiveMetas = allMetas.filter(isSubstantiveSession)

  // Phase 3: Facet extraction — only for sessions without cached facets
  const facets = new Map<string, SessionFacets>()
  const toExtract: Array<{ log: LogOption; sessionId: string }> = []
  const MAX_FACET_EXTRACTIONS = 50

  // Load cached facets for all substantive sessions in parallel
  const cachedFacetResults = await Promise.all(
    substantiveMetas.map(async meta => ({
      sessionId: meta.session_id,
      cached: await loadCachedFacets(meta.session_id),
    })),
  )
  for (const { sessionId, cached } of cachedFacetResults) {
    if (cached) {
      facets.set(sessionId, cached)
    } else {
      const log = logsForFacets.get(sessionId)
      if (log && toExtract.length < MAX_FACET_EXTRACTIONS) {
        toExtract.push({ log, sessionId })
      }
    }
  }

  // Extract facets for sessions that need them (50 concurrent)
  const CONCURRENCY = 50
  for (let i = 0; i < toExtract.length; i += CONCURRENCY) {
    const batch = toExtract.slice(i, i + CONCURRENCY)
    const results = await Promise.all(
      batch.map(async ({ log, sessionId }) => {
        const newFacets = await extractFacetsFromAPI(log, sessionId)
        return { sessionId, newFacets }
      }),
    )
    // Collect facets synchronously, save in parallel (independent writes)
    const facetsToSave: SessionFacets[] = []
    for (const { sessionId, newFacets } of results) {
      if (newFacets) {
        facets.set(sessionId, newFacets)
        facetsToSave.push(newFacets)
      }
    }
    await Promise.all(facetsToSave.map(f => saveFacets(f)))
  }

  // Filter out warmup/minimal sessions (matching Python's is_minimal)
  // A session is minimal if warmup_minimal is the ONLY goal category
  const isMinimalSession = (sessionId: string): boolean => {
    const sessionFacets = facets.get(sessionId)
    if (!sessionFacets) return false
    const cats = sessionFacets.goal_categories
    const catKeys = safeKeys(cats).filter(k => (cats[k] ?? 0) > 0)
    return catKeys.length === 1 && catKeys[0] === 'warmup_minimal'
  }

  const substantiveSessions = substantiveMetas.filter(
    s => !isMinimalSession(s.session_id),
  )

  const substantiveFacets = new Map<string, SessionFacets>()
  for (const [sessionId, f] of facets) {
    if (!isMinimalSession(sessionId)) {
      substantiveFacets.set(sessionId, f)
    }
  }

  const aggregated = aggregateData(substantiveSessions, substantiveFacets)
  aggregated.total_sessions_scanned = totalSessionsScanned

  // Generate parallel insights from Claude (6 sections)
  const insights = await generateParallelInsights(aggregated, facets)

  // Generate HTML report
  const htmlReport = generateHtmlReport(aggregated, insights)

  // Save reports
  try {
    await mkdir(getDataDir(), { recursive: true })
  } catch {
    // Directory may already exist
  }

  const htmlPath = join(getDataDir(), 'report.html')
  await writeFile(htmlPath, htmlReport, {
    encoding: 'utf-8',
    mode: 0o600,
  })

  return {
    insights,
    htmlPath,
    data: aggregated,
    remoteStats,
    facets: substantiveFacets,
  }
}

function safeEntries<V>(
  obj: Record<string, V> | undefined | null,
): [string, V][] {
  return obj ? Object.entries(obj) : []
}

function safeKeys(obj: Record<string, unknown> | undefined | null): string[] {
  return obj ? Object.keys(obj) : []
}

// ============================================================================
// Command Definition
// ============================================================================

const usageReport: Command = {
  type: 'prompt',
  name: 'insights',
  description: 'Generate a report analyzing your Claude Code sessions',
  contentLength: 0, // Dynamic content
  progressMessage: 'analyzing your sessions',
  source: 'builtin',
  async getPromptForCommand(args) {
    let collectRemote = false
    let remoteHosts: string[] = []
    let hasRemoteHosts = false

    if (process.env.USER_TYPE === 'ant') {
      // Parse --homespaces flag
      collectRemote = args?.includes('--homespaces') ?? false

      // Check for available remote hosts
      remoteHosts = await getRunningRemoteHosts()
      hasRemoteHosts = remoteHosts.length > 0

      // Show collection message if collecting
      if (collectRemote && hasRemoteHosts) {
        // biome-ignore lint/suspicious/noConsole: intentional
        console.error(
          `Collecting sessions from ${remoteHosts.length} homespace(s): ${remoteHosts.join(', ')}...`,
        )
      }
    }

    const { insights, htmlPath, data, remoteStats } = await generateUsageReport(
      { collectRemote },
    )

    let reportUrl = `file://${htmlPath}`
    let uploadHint = ''

    if (process.env.USER_TYPE === 'ant') {
      // Try to upload to S3
      const timestamp = new Date()
        .toISOString()
        .replace(/[-:]/g, '')
        .replace('T', '_')
        .slice(0, 15)
      const username = process.env.SAFEUSER || process.env.USER || 'unknown'
      const filename = `${username}_insights_${timestamp}.html`
      const s3Path = `s3://anthropic-serve/atamkin/cc-user-reports/${filename}`
      const s3Url = `https://s3-frontend.infra.ant.dev/anthropic-serve/atamkin/cc-user-reports/${filename}`

      reportUrl = s3Url
      try {
        execFileSync('ff', ['cp', htmlPath, s3Path], {
          timeout: 60000,
          stdio: 'pipe', // Suppress output
        })
      } catch {
        // Upload failed - fall back to local file and show upload command
        reportUrl = `file://${htmlPath}`
        uploadHint = `\nAutomatic upload failed. Are you on the boron namespace? Try \`use-bo\` and ensure you've run \`sso\`.
To share, run: ff cp ${htmlPath} ${s3Path}
Then access at: ${s3Url}`
      }
    }

    // Build header with stats
    const sessionLabel =
      data.total_sessions_scanned &&
      data.total_sessions_scanned > data.total_sessions
        ? `${data.total_sessions_scanned.toLocaleString()} sessions total · ${data.total_sessions} analyzed`
        : `${data.total_sessions} sessions`
    const stats = [
      sessionLabel,
      `${data.total_messages.toLocaleString()} messages`,
      `${Math.round(data.total_duration_hours)}h`,
      `${data.git_commits} commits`,
    ].join(' · ')

    // Build remote host info (ant-only)
    let remoteInfo = ''
    if (process.env.USER_TYPE === 'ant') {
      if (remoteStats && remoteStats.totalCopied > 0) {
        const hsNames = remoteStats.hosts
          .filter(h => h.sessionCount > 0)
          .map(h => h.name)
          .join(', ')
        remoteInfo = `\n_Collected ${remoteStats.totalCopied} new sessions from: ${hsNames}_\n`
      } else if (!collectRemote && hasRemoteHosts) {
        // Suggest using --homespaces if they have remote hosts but didn't use the flag
        remoteInfo = `\n_Tip: Run \`/insights --homespaces\` to include sessions from your ${remoteHosts.length} running homespace(s)_\n`
      }
    }

    // Build markdown summary from insights
    const atAGlance = insights.at_a_glance
    const summaryText = atAGlance
      ? `## At a Glance

${atAGlance.whats_working ? `**What's working:** ${atAGlance.whats_working} See _Impressive Things You Did_.` : ''}

${atAGlance.whats_hindering ? `**What's hindering you:** ${atAGlance.whats_hindering} See _Where Things Go Wrong_.` : ''}

${atAGlance.quick_wins ? `**Quick wins to try:** ${atAGlance.quick_wins} See _Features to Try_.` : ''}

${atAGlance.ambitious_workflows ? `**Ambitious workflows:** ${atAGlance.ambitious_workflows} See _On the Horizon_.` : ''}`
      : '_No insights generated_'

    const header = `# Claude Code Insights

${stats}
${data.date_range.start} to ${data.date_range.end}
${remoteInfo}
`

    const userSummary = `${header}${summaryText}

Your full shareable insights report is ready: ${reportUrl}${uploadHint}`

    // Return prompt for Claude to respond to
    return [
      {
        type: 'text',
        text: `The user just ran /insights to generate a usage report analyzing their Claude Code sessions.

Here is the full insights data:
${jsonStringify(insights, null, 2)}

Report URL: ${reportUrl}
HTML file: ${htmlPath}
Facets directory: ${getFacetsDir()}

Here is what the user sees:
${userSummary}

Now output the following message exactly:

<message>
Your shareable insights report is ready:
${reportUrl}${uploadHint}

Want to dig into any section or try one of the suggestions?
</message>`,
      },
    ]
  },
}

function isValidSessionFacets(obj: unknown): obj is SessionFacets {
  if (!obj || typeof obj !== 'object') return false
  const o = obj as Record<string, unknown>
  return (
    typeof o.underlying_goal === 'string' &&
    typeof o.outcome === 'string' &&
    typeof o.brief_summary === 'string' &&
    o.goal_categories !== null &&
    typeof o.goal_categories === 'object' &&
    o.user_satisfaction_counts !== null &&
    typeof o.user_satisfaction_counts === 'object' &&
    o.friction_counts !== null &&
    typeof o.friction_counts === 'object'
  )
}

export default usageReport