@kaifronsdal/transcript-viewer/src/lib/shared/utils/transcript-utils.ts

A web-based viewer for AI conversation transcripts with rollback support

import type { Transcript, TranscriptDisplay, TranscriptDisplayFull, TranscriptDisplayMeta } from '$lib/shared/types';

/**
 * Extract system prompt from transcript events
 */
export function extractSystemPrompt(transcript: Transcript): string | undefined {
  // Check if events exist and is iterable
  if (!transcript?.events || !Array.isArray(transcript.events)) {
    console.warn('extractSystemPrompt: transcript.events is not a valid array:', transcript?.events);
    return undefined;
  }

  for (const event of transcript.events) {
    if (event.type === 'transcript_event') {
      if (event.edit.operation === 'add' && event.edit.message?.role === 'system') {
        const content = event.edit.message.content;
        return typeof content === 'string' ? content : JSON.stringify(content);
      }
    }
  }
  return undefined;
}

/**
 * Extract target model from transcript metadata or system prompt content
 */
export function extractTargetModel(transcript: Transcript): string {
  // First try the metadata field
  if (transcript.metadata.target_model && transcript.metadata.target_model !== 'unknown') {
    return transcript.metadata.target_model;
  }
  
  // If metadata is unknown, try to extract from system prompt content
  const systemPrompt = extractSystemPrompt(transcript);
  if (systemPrompt) {
    // Look for pattern like "The target AI is a `anthropic:claude-3-5-sonnet-latest` model"
    const match = systemPrompt.match(/The target AI is a `([^`]+)` model/);
    if (match && match[1]) {
      return match[1];
    }
  }
  
  // Fallback to unknown
  return 'unknown';
}

/**
 * Extract model name from target model string
 */
export function extractModelName(targetModel: string): string {
  if (!targetModel || targetModel === 'unknown') {
    return 'unknown';
  }
  
  // Extract just the model name from patterns like "anthropic:claude-3-5-sonnet-latest"
  const parts = targetModel.split(':');
  if (parts.length > 1) {
    return parts[1];
  }
  
  return targetModel;
}

// Note: extractBehaviorDir and extractTranscriptNumber functions were removed
// since they required Node.js path module and weren't being used anywhere

/**
 * Extract the first sentence from a description text
 */
export function extractFirstSentence(text: string): string {
  if (!text) return '';
  
  // Split on sentence endings, but be careful with abbreviations
  // Look for periods, exclamation marks, or question marks followed by space or end of string
  const sentenceEndings = /[.!?](?:\s|$)/;
  const match = text.match(sentenceEndings);
  
  if (match) {
    const endIndex = match.index! + match[0].length - (match[0].endsWith(' ') ? 1 : 0);
    return text.substring(0, endIndex).trim();
  }
  
  // If no sentence ending found, return the whole text (up to a reasonable limit)
  return text.length > 150 ? text.substring(0, 150) + '...' : text;
}

/**
 * Collect all score descriptions from transcripts and return a map of score name to description
 * For the home page, this allows us to show tooltips even when we don't have the full transcript data
 */
export function collectScoreDescriptions(transcripts: { scoreDescriptions?: Record<string, string> }[]): Record<string, string> {
  const allDescriptions: Record<string, string> = {};
  
  for (const transcript of transcripts) {
    if (transcript.scoreDescriptions) {
      for (const [scoreName, description] of Object.entries(transcript.scoreDescriptions)) {
        // Only add if we don't already have a description for this score
        if (!allDescriptions[scoreName] && description) {
          allDescriptions[scoreName] = description;
        }
      }
    }
  }
  
  return allDescriptions;
}

/**
 * Create display data from transcript (full version with all data)
 */
export function createTranscriptDisplay(
  transcript: Transcript,
  filePath: string
): TranscriptDisplayFull {
  const targetModel = extractTargetModel(transcript);
  // Extract behavior directory (parent folder name) - client-safe version
  const pathParts = filePath.split('/');
  const behaviorDir = pathParts.length > 1 ? pathParts[pathParts.length - 2] : '';
  // Extract transcript number/name (filename without extension) - client-safe version  
  const fileName = pathParts[pathParts.length - 1];
  const transcriptNumber = fileName.endsWith('.json') ? fileName.slice(0, -5) : fileName;
  
  return {
    id: transcript.metadata.transcript_id || transcriptNumber,
    model: extractModelName(targetModel),
    split: behaviorDir,
    concerningScore: transcript.metadata.judge_output?.scores?.concerning || 0,
    summary: transcript.metadata.judge_output?.summary || transcript.metadata.description || 'No summary available',
    scores: transcript.metadata.judge_output?.scores || {},
    scoreDescriptions: transcript.metadata.judge_output?.score_descriptions,
    judgeSummary: transcript.metadata.judge_output?.summary || 'No judgment summary available',
    justification: transcript.metadata.judge_output?.justification || 'No justification available',
    tags: transcript.metadata.tags || [],
    systemPrompt: extractSystemPrompt(transcript),
    transcript: transcript,
    _filePath: filePath
  };
}

/**
 * Extract just metadata needed for display (for metadata-only operations)
 */
export function extractTranscriptMetadata(
  transcript: Transcript,
  filePath: string
): TranscriptDisplayMeta {
  const targetModel = extractTargetModel(transcript);
  // Extract behavior directory (parent folder name) - client-safe version
  const pathParts = filePath.split('/');
  const behaviorDir = pathParts.length > 1 ? pathParts[pathParts.length - 2] : '';
  // Extract transcript number/name (filename without extension) - client-safe version  
  const fileName = pathParts[pathParts.length - 1];
  const transcriptNumber = fileName.endsWith('.json') ? fileName.slice(0, -5) : fileName;
  
  return {
    id: transcript.metadata.transcript_id || transcriptNumber,
    model: extractModelName(targetModel),
    split: behaviorDir,
    concerningScore: transcript.metadata.judge_output?.scores?.concerning || 0,
    summary: transcript.metadata.judge_output?.summary || transcript.metadata.description || 'No summary available',
    scores: transcript.metadata.judge_output?.scores || {},
    scoreDescriptions: transcript.metadata.judge_output?.score_descriptions,
    judgeSummary: transcript.metadata.judge_output?.summary || 'No judgment summary available',
    justification: transcript.metadata.judge_output?.justification || 'No justification available',
    tags: transcript.metadata.tags || [],
    systemPrompt: undefined, // Don't extract system prompt for metadata-only
    _filePath: filePath
  };
}