process-transcript

name	process-transcript
description	Process a VTT transcript into a comprehensive, multi-tab HTML dashboard using specialized analysis agents. Use when the user provides a .vtt file, has a transcript to process, or mentions a meeting recording. This is the PRIMARY skill for handling transcripts.
allowed-tools	Read, Write, Bash, Glob

Process Transcript to Dashboard (Advanced)

Convert a .vtt meeting transcript into a comprehensive, deployed HTML dashboard using multiple specialized analyzers.

CRITICAL: What This Skill Does

1. ✅ Orchestrates 5 specialized analysis agents
2. ✅ Runs consolidation to ensure consistency
3. ✅ Creates a multi-tab HTML dashboard with rich insights
4. ✅ Generates a deployable dashboard (deploy via deploy-dashboard skill or node processor/deploy-local.js)
5. ✅ Returns a public URL

NEVER DO

• ❌ Create markdown (.md) files
• ❌ Skip any analysis phase
• ❌ Skip the consolidation step
• ❌ Generate a simple single-page summary
• ❌ Deploy the dashboard directly (use the deploy-dashboard skill or deploy-local.js instead)

Pipeline Steps

Step 1: Setup

• Determine project name and date
• Create folder structure
• Copy transcript to projects/{project}/transcripts/{date}.vtt

Step 2: Run Analysis Agents (IN PARALLEL if possible)

Delegate to each specialized agent in .claude/agents/:

1. timeline-analyzer → analysis/timeline.json

   • Identifies meeting segments (review, retro, planning)
   • Creates timeline with timestamps and summaries

2. people-analyzer → analysis/people.json

   • Analyzes each participant's contributions
   • Calculates speaking time vs. value contributed
   • Provides constructive feedback

3. insights-generator → analysis/insights.json

   • Finds non-obvious patterns and observations
   • Identifies risks, opportunities, team health signals
   • Generates ad-hoc creative insights

4. analytics-generator → analysis/analytics.json

   • Produces data-driven metrics and statistics
   • Creates chart-ready data
   • Calculates efficiency scores

5. longitudinal-analyzer → analysis/longitudinal.json

   • Compares with previous meetings (if available)
   • Identifies trends and patterns over time
   • Tracks improvement and recurring issues

Step 3: CONSOLIDATION (Critical Quality Step)

Run the consolidator agent to:

1. Normalize names

   • Create canonical name mapping
   • If "Alice" is identified as "Product Owner", use "Alice" everywhere
   • Replace role references with actual names where known

2. Cross-reference validation

   • Verify consistency between agent outputs
   • Resolve metric discrepancies
   • Link action items to owners

3. Deduplicate

   • Remove redundant insights
   • Merge duplicate action items
   • Consolidate repeated quotes

4. Enrich data

   • Connect decisions to timeline segments
   • Link insights to relevant participants
   • Add IDs for cross-referencing

5. Quality check

   • Flag unidentified participants
   • Note action items without owners
   • Document all changes made

Output: analysis/consolidated.json

Step 4: Generate Multi-Tab Dashboard

Create projects/{project}/dashboards/{date}/index.html using the consolidated data:

Tab 1: Overview

• Meeting summary (factual bullets)
• Key decisions (max 3 bullet points)
• Done This Sprint (outcomes, completed work — excluding decisions)
• Next Steps / action items with owners (canonical names)
• Hard truths (max 2, residual only)

Tab 2: Timeline

• Visual timeline of meeting segments
• Duration bars
• Key moments highlighted
• Participants (by canonical name)
• Energy level indicators

Tab 3: People & Roles

• Participant cards with profile photos from SSW People (with fallback initials for non-SSW)
• Canonical names with roles as subtitle
• Speaking time visualization
• Value contribution ratings
• Strengths and feedback for each person
• Team dynamics summary

Tab 4: Insights

• Ad-hoc observations
• Risk signals
• Opportunity spotting
• Team health indicators
• Notable quotes (attributed to canonical names)

Tab 5: Trends (if historical data exists)

• Week-over-week comparisons
• Recurring themes
• Improvement tracking
• Trajectory visualizations

Step 5: Deploy Dashboard

After generating the dashboard, offer deployment options:

1. Ask Claude: "deploy the dashboard" → triggers the deploy-dashboard skill
2. Run manually: node processor/deploy-local.js {project} {meeting-id}

The deployed URL will be: https://dashboards.sswtiger.com/{project}/{meeting-id}

Note: When running under processor.js (Azure pipeline), deployment is handled automatically by deployer.js. Do NOT deploy from within this skill in that context.

Step 6: Report Success

✓ Analysis complete:
  - Timeline: 5 segments identified
  - People: 6 participants analyzed  
  - Insights: 8 observations generated
  - Analytics: 12 metrics calculated
  - Trends: Compared with 3 previous meetings

✓ Consolidation:
  - Normalized 12 name references
  - Merged 2 duplicate action items
  - Identified 2 unidentified speakers
  - Quality score: 92/100

✓ Dashboard generated: projects/{project}/dashboards/{date}/index.html
✓ Deployed to: https://dashboards.sswtiger.com/{project}/{meeting-id}

File Structure

projects/{project}/
├── transcripts/{date}.vtt
├── analysis/
│   ├── timeline.json          # From timeline-analyzer
│   ├── people.json            # From people-analyzer
│   ├── insights.json          # From insights-generator
│   ├── analytics.json         # From analytics-generator
│   ├── longitudinal.json      # From longitudinal-analyzer
│   └── consolidated.json      # From consolidator (USED FOR DASHBOARD)
└── dashboards/{date}/
    └── index.html             # THE DELIVERABLE

Dashboard Quality Standards

The dashboard should be:

• Consistent - Same names/terms throughout all tabs
• Beautiful - Modern, polished design with Tailwind CSS
• Interactive - Tabs, hover effects, expandable sections
• Data-rich - Charts, metrics, visualizations
• Actionable - Clear next steps and recommendations
• Insightful - Goes beyond obvious observations
• Human - Recognizes individual contributions by name

Technology Stack

• Tailwind CSS (CDN) - Styling
• Chart.js (CDN) - Data visualizations
• Alpine.js (CDN) - Tab interactivity
• Vanilla JS - Additional interactions

Consolidation Examples

Name Normalization

BEFORE (inconsistent):
- Timeline: "Product Owner presented..."
- People: "Alice spoke for 15 minutes..."
- Action Items: "PO to follow up..."

AFTER (consolidated):
- Timeline: "Alice (Product Owner) presented..."
- People: "Alice spoke for 15 minutes..."
- Action Items: "Alice to follow up..."

Cross-Reference

BEFORE (disconnected):
- Decision: "Decided to postpone feature X"
- Timeline: Segment 3 mentions postponement
- Person: Bob raised the concern

AFTER (connected):
- Decision: "Decided to postpone feature X" 
  - Decided during: Segment 3 (Sprint Planning)
  - Raised by: Bob
  - Related insight: Technical debt concern #2