task-summary-extractor 9.7.0 → 9.8.0

package/README.md

@@ -1,6 +1,6 @@
 # Task Summary Extractor
 
-> **v9.7.0** — AI-powered content analysis CLI — meetings, recordings, documents, or any mix. Install globally, run anywhere.
+> **v9.8.0** — AI-powered content analysis CLI — meetings, recordings, documents, or any mix. Install globally, run anywhere.
 
 <p align="center">
   <img src="https://img.shields.io/badge/node-%3E%3D18.0.0-green" alt="Node.js" />
@@ -598,6 +598,7 @@ task-summary-extractor/
 
 | Version | Highlights |
 |---------|-----------|
+| **v9.8.0** | **Schema hardening & transcript handling** — VTT/SRT auto-excluded from deep-summary (transcripts routed to workflow, not summarizer), `normalizeAnalysis()` fills missing `summary`/`confidence`/`discussed_state` defaults before validation, batch Storage URL→File API auto-retry on `INVALID_ARGUMENT`, focused re-analysis skips sparse segments (≤2 items + low density), 367 tests |
 | **v9.7.0** | **Multi-segment batching** — groups consecutive video segments into single Gemini API calls when context window has headroom, greedy bin-packing by token budget (`planSegmentBatches`), `processSegmentBatch()` multi-video API calls, automatic fallback to single-segment on failure, `--no-batch` to disable, codebase audit fixes (unused imports, variable shadowing) |
 | **v9.6.0** | **Interactive CLI UX** — arrow-key navigation for all selectors (folder, model, run mode, formats, confidence, doc exclusion), zero-dependency prompt engine (`interactive.js`), `selectOne()` with ↑↓+Enter, `selectMany()` with Space toggle + A all/none, non-TTY fallback to number input |
 | **v9.5.0** | **Video processing flags** — `--no-compress`, `--speed`, `--segment-time` CLI flags, hardcoded 1200s for raw mode, deprecated `--skip-compression` |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "task-summary-extractor",
-  "version": "9.7.0",
+  "version": "9.8.0",
   "description": "AI-powered meeting analysis & document generation CLI — video + document processing, deep dive docs, dynamic mode, interactive CLI with model selection, confidence scoring, learning loop, git progress tracking",
   "main": "process_and_upload.js",
   "bin": {

package/src/modes/deep-summary.js

@@ -27,6 +27,20 @@ const config = require('../config');
 
 // ======================== CONSTANTS ========================
 
+/**
+ * Transcript file extensions that should NEVER be summarized.
+ * VTT/SRT files are time-sliced per segment during analysis — summarising
+ * them would destroy the timestamp-indexed structure that `sliceVttForSegment`
+ * relies on. They are automatically kept at full fidelity.
+ */
+const TRANSCRIPT_EXTENSIONS = ['.vtt', '.srt'];
+
+/** Check whether a filename is a transcript file (VTT/SRT). */
+function isTranscriptFile(fileName) {
+  const lower = (fileName || '').toLowerCase();
+  return TRANSCRIPT_EXTENSIONS.some(ext => lower.endsWith(ext));
+}
+
 /** Max tokens for a single summarization call output */
 const SUMMARY_MAX_OUTPUT = 16384;
 
@@ -262,6 +276,13 @@ async function deepSummarize(ai, contextDocs, opts = {}) {
       continue;
     }
 
+    // Auto-exclude transcript files (VTT/SRT) — they are time-sliced per
+    // segment during analysis and must retain their timestamp structure.
+    if (isTranscriptFile(doc.fileName)) {
+      keepFull.push(doc);
+      continue;
+    }
+
     // Keep excluded docs at full fidelity
     if (excludeSet.has(doc.fileName.toLowerCase())) {
       keepFull.push(doc);
@@ -294,14 +315,22 @@ async function deepSummarize(ai, contextDocs, opts = {}) {
   }
 
   // Build focus topics from excluded docs (tell summarizer what to prioritize)
+  // NOTE: transcript files (VTT/SRT) are auto-excluded but NOT used as focus
+  // topics — they are time-sliced per segment and don't represent "topics".
   const focusTopics = keepFull
     .filter(d => d.type === 'inlineText' && excludeSet.has(d.fileName.toLowerCase()))
     .map(d => d.fileName);
 
+  // Count auto-excluded transcript files for logging
+  const autoExcludedTranscripts = keepFull.filter(d => isTranscriptFile(d.fileName));
+
   // Batch documents
   const batches = buildBatches(toSummarize);
 
   console.log(`    Batched ${c.highlight(toSummarize.length)} doc(s) into ${c.highlight(batches.length)} summarization batch(es)`);
+  if (autoExcludedTranscripts.length > 0) {
+    console.log(`    Auto-excluded ${c.highlight(autoExcludedTranscripts.length)} transcript file(s) (VTT/SRT — time-sliced per segment)`);
+  }
   if (focusTopics.length > 0) {
     console.log(`    Focus topics from ${c.highlight(focusTopics.length)} excluded doc(s):`);
     focusTopics.forEach(t => console.log(`      ${c.dim('•')} ${c.cyan(t)}`));
@@ -350,6 +379,12 @@ async function deepSummarize(ai, contextDocs, opts = {}) {
       continue;
     }
 
+    // Auto-exclude transcript files (VTT/SRT)
+    if (isTranscriptFile(doc.fileName)) {
+      resultDocs.push(doc);
+      continue;
+    }
+
     // Check if we have a summary for this doc
     const summaryKey = doc.fileName.toLowerCase();
     const summary = allSummaries.get(summaryKey);
@@ -399,6 +434,8 @@ module.exports = {
   deepSummarize,
   summarizeBatch,
   buildBatches,
+  isTranscriptFile,
+  TRANSCRIPT_EXTENSIONS,
   SUMMARY_MAX_OUTPUT,
   BATCH_MAX_CHARS,
   MIN_SUMMARIZE_LENGTH,

package/src/modes/focused-reanalysis.js

@@ -136,9 +136,24 @@ function identifyWeaknesses(qualityReport, analysis) {
     );
   }
 
+  // ── Skip focused pass for simple / sparse segments ──────────────────────
+  // When the analysis has very few extracted items AND the density dimension
+  // is low, the segment is likely simple (chit-chat, small-talk, intro) or
+  // the AI legitimately had nothing to extract. A focused pass won't help.
+  const totalItems = [
+    ...(analysis.tickets || []),
+    ...(analysis.action_items || []),
+    ...(analysis.change_requests || []),
+    ...(analysis.blockers || []),
+    ...(analysis.scope_changes || []),
+  ].length;
+
+  const isSparseSegment = totalItems <= 2 && dims.density && dims.density.score < 30;
+
   const shouldReanalyze = focusInstructions.length > 0 &&
     qualityReport.score < 60 &&       // Only re-analyze if quality is truly lacking
-    weakAreas.length >= 2;            // At least 2 weak areas to justify the cost
+    weakAreas.length >= 2 &&          // At least 2 weak areas to justify the cost
+    !isSparseSegment;                 // Don't waste tokens on sparse / simple segments
 
   const focusPrompt = focusInstructions.length > 0
     ? focusInstructions.join('\n\n')

package/src/phases/process-media.js

@@ -352,18 +352,43 @@ async function phaseProcessVideo(ctx, videoPath, videoIndex) {
         }
 
         try {
-          const batchRun = await processSegmentBatch(
-            ai, batchSegs,
-            `${callName}_${baseName}_batch${bIdx}`,
-            contextDocs, previousAnalyses, userName, PKG_ROOT,
-            {
-              segmentIndices: batchIndices,
-              totalSegments: segments.length,
-              segmentTimes: batchTimes,
-              thinkingBudget: opts.thinkingBudget || 24576,
-              noStorageUrl: !!opts.noStorageUrl,
+          let batchRun;
+          try {
+            batchRun = await processSegmentBatch(
+              ai, batchSegs,
+              `${callName}_${baseName}_batch${bIdx}`,
+              contextDocs, previousAnalyses, userName, PKG_ROOT,
+              {
+                segmentIndices: batchIndices,
+                totalSegments: segments.length,
+                segmentTimes: batchTimes,
+                thinkingBudget: opts.thinkingBudget || 24576,
+                noStorageUrl: !!opts.noStorageUrl,
+              }
+            );
+          } catch (batchErr) {
+            const msg = batchErr.message || '';
+            // If Storage URL was rejected, retry batch with forced File API uploads
+            if (!opts.noStorageUrl && msg.includes('INVALID_ARGUMENT') && batchSegs.some(s => s.storageUrl)) {
+              console.log(`    ${c.warn('Storage URL rejected — retrying batch with File API uploads...')}`);
+              log.warn(`Batch ${bIdx} Storage URL rejected — retrying with noStorageUrl=true`);
+              batchRun = await processSegmentBatch(
+                ai, batchSegs,
+                `${callName}_${baseName}_batch${bIdx}`,
+                contextDocs, previousAnalyses, userName, PKG_ROOT,
+                {
+                  segmentIndices: batchIndices,
+                  totalSegments: segments.length,
+                  segmentTimes: batchTimes,
+                  thinkingBudget: opts.thinkingBudget || 24576,
+                  noStorageUrl: true,
+                }
+              );
+              console.log(`    ${c.success('File API batch retry succeeded')}`);
+            } else {
+              throw batchErr;
             }
-          );
+          }
 
           // Save batch run file
           const ts = new Date().toISOString().replace(/[:.]/g, '-').slice(0, 19);

package/src/utils/cli.js

@@ -587,9 +587,12 @@ async function selectConfidence() {
  * @returns {Promise<string[]>} Array of excluded fileName strings
  */
 async function selectDocsToExclude(contextDocs) {
-  // Only show inlineText docs with actual content
+  const { isTranscriptFile } = require('../modes/deep-summary');
+
+  // Only show inlineText docs with actual content, excluding transcript files
+  // (VTT/SRT are auto-excluded from summarization — no need to show them)
   const eligible = contextDocs
-    .filter(d => d.type === 'inlineText' && d.content && d.content.length > 0)
+    .filter(d => d.type === 'inlineText' && d.content && d.content.length > 0 && !isTranscriptFile(d.fileName))
     .map(d => ({
       fileName: d.fileName,
       chars: d.content.length,

package/src/utils/schema-validator.js

@@ -316,8 +316,12 @@ const ARRAY_DEFAULTS = [
 
 /**
  * Normalize a parsed analysis object by filling in missing array fields
- * with empty arrays. This prevents downstream code from crashing when
- * a segment legitimately has no tickets/action_items/etc.
+ * with empty arrays, ensuring required string fields exist, and patching
+ * item-level required fields (e.g. confidence) with sensible defaults.
+ *
+ * This prevents downstream code from crashing when a segment legitimately
+ * has no tickets/action_items/etc., and avoids schema-validation failures
+ * on truncated AI outputs.
  *
  * Mutates `data` in-place and returns it for convenience.
  *
@@ -326,11 +330,38 @@ const ARRAY_DEFAULTS = [
  */
 function normalizeAnalysis(data) {
   if (!data || typeof data !== 'object') return data;
+
+  // Fill missing array fields
   for (const field of ARRAY_DEFAULTS) {
     if (data[field] === undefined || data[field] === null) {
       data[field] = [];
     }
   }
+
+  // Ensure top-level required string fields
+  if (!data.summary && data.summary !== '') {
+    data.summary = data.segment_summary || data.overview || '';
+  }
+
+  // Patch ticket items — fill missing required fields with defaults
+  if (Array.isArray(data.tickets)) {
+    for (const ticket of data.tickets) {
+      if (!ticket || typeof ticket !== 'object') continue;
+      if (!ticket.confidence) ticket.confidence = 'MEDIUM';
+      if (!ticket.discussed_state && ticket.discussed_state !== null) {
+        ticket.discussed_state = { summary: '' };
+      }
+    }
+  }
+
+  // Patch action_items — fill missing confidence
+  if (Array.isArray(data.action_items)) {
+    for (const item of data.action_items) {
+      if (!item || typeof item !== 'object') continue;
+      if (!item.confidence) item.confidence = 'MEDIUM';
+    }
+  }
+
   return data;
 }