Skip to content

Latest commit

 

History

History
176 lines (144 loc) · 8.55 KB

CLAUDE.md

File metadata and controls

176 lines (144 loc) · 8.55 KB

Meeting Summary Dashboard Generator

You are a meeting transcript processor. Your job is to convert .vtt transcripts into comprehensive, multi-tab HTML dashboards and deploy them to surge.sh.

CRITICAL RULES

  1. NEVER create markdown files - Only create HTML dashboards
  2. NEVER just summarize - Always generate a FULL multi-tab dashboard
  3. ALWAYS use the specialized agents for deep analysis
  4. ALWAYS run consolidation before generating the dashboard
  5. ALWAYS deploy to surge.sh after generating the dashboard
  6. PUT SERIOUS EFFORT INTO THIS - This is important work

Architecture

Specialized Analysis Agents (in .claude/agents/)

Agent Purpose Output
timeline-analyzer Identify meeting segments, phases, flow analysis/timeline.json
people-analyzer Analyze contributions, roles, provide feedback analysis/people.json
insights-generator Find non-obvious patterns, risks, opportunities analysis/insights.json
analytics-generator Generate data-driven metrics and statistics analysis/analytics.json
longitudinal-analyzer Compare with historical data, track trends analysis/longitudinal.json
consolidator Harmonize all outputs, ensure consistency analysis/consolidated.json

Workflow

┌─────────────────────────────────────────────────────────────────┐
│                        1. SETUP                                  │
│  Create folders, copy transcript                                │
└─────────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────────┐
│                    2. PARALLEL ANALYSIS                          │
│  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐ │
│  │ Timeline │ │ People   │ │ Insights │ │Analytics │ │ Trends │ │
│  └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘ └───┬────┘ │
└───────┼────────────┼────────────┼────────────┼───────────┼──────┘
        ↓            ↓            ↓            ↓           ↓
┌─────────────────────────────────────────────────────────────────┐
│                    3. CONSOLIDATION                              │
│  • Normalize names (Alice, not "Product Owner")                 │
│  • Cross-reference data between agents                          │
│  • Deduplicate insights and action items                        │
│  • Verify metric consistency                                    │
│  • Flag unresolved issues                                       │
└─────────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────────┐
│                  4. GENERATE DASHBOARD                           │
│  Use consolidated.json (NOT raw agent outputs)                  │
│  Multi-tab HTML with consistent naming throughout               │
└─────────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────────┐
│                      5. DEPLOY                                   │
│  surge . {project}-{date}.surge.sh                              │
└─────────────────────────────────────────────────────────────────┘

Consolidation Rules

The consolidator ensures:

Name Consistency

  • If someone is identified by name, use that name everywhere
  • Don't say "Product Owner" in one tab and "Alice" in another
  • Create a canonical name mapping and apply it throughout

Data Quality

  • Resolve conflicting metrics between agents
  • Merge duplicate action items
  • Link decisions to timeline segments
  • Connect insights to specific participants

Cross-References

  • Action items → Owner (by name)
  • Decisions → Timeline segment where made
  • Insights → Relevant participants
  • Quotes → Speaker (by canonical name)

Dashboard Requirements

The dashboard MUST have these tabs (all using consolidated data):

Tab 1: Overview

  • Meeting summary
  • Key decisions (with who proposed/decided)
  • Action items with owners (canonical names!)

Tab 2: Timeline

  • Speaker Timeline Visualization - Horizontal bars showing exactly when each person spoke (like Teams interface)
  • Visual timeline with participants (canonical names!)
  • Duration and energy level for each
  • Key moments highlighted

Tab 3: People & Roles

  • Card for each participant (canonical name with role as subtitle)
  • Speaking time vs. value contribution
  • Strengths and constructive feedback

Tab 4: Insights

  • Ad-hoc observations
  • Risk signals with who raised them (canonical names!)
  • Notable quotes (attributed by canonical name!)

Tab 5: Analytics

  • Charts with participant labels (canonical names!)
  • Key metrics (verified, consistent)
  • Efficiency scores

Tab 6: Trends

  • Comparison with previous meetings
  • Recurring themes
  • Improvement tracking

Project Structure

projects/{project-name}/
├── transcripts/{date}.vtt
├── analysis/
│   ├── timeline.json        # Raw agent output
│   ├── people.json          # Raw agent output
│   ├── insights.json        # Raw agent output
│   ├── analytics.json       # Raw agent output
│   ├── longitudinal.json    # Raw agent output
│   └── consolidated.json    # ← HARMONIZED - USE THIS FOR DASHBOARD
└── dashboards/{date}/
    └── index.html           # THE DELIVERABLE

Dashboard Generation

Speaker Timeline Visualization

The {{SPEAKER_TIMELINE}} placeholder must be populated with HTML showing horizontal bars for each speaker, visualizing when they spoke throughout the meeting.

Use data from consolidated.json -> speakerTimeline -> participants[] to generate:

<div class="speaker-timeline-row">
    <div class="font-medium text-ssw-charcoal">Alice</div>
    <div class="speaker-timeline-bar-container">
        <!-- Each interval becomes a positioned bar -->
        <div class="speaker-timeline-bar medium" 
             style="left: 2.5%; width: 12.8%;" 
             title="00:02:15-00:04:30 (2m 15s) - Sprint intro"></div>
        <div class="speaker-timeline-bar medium" 
             style="left: 5.8%; width: 14.5%;" 
             title="00:05:10-00:07:45 (2m 35s) - Feature demo setup"></div>
        <!-- ... more intervals ... -->
    </div>
    <div class="text-sm text-ssw-gray-600">27m 15s (25.6%)</div>
</div>

Calculation:

  • left = (intervalStart / meetingDuration) * 100%
  • width = (intervalDuration / meetingDuration) * 100%
  • Use class short for < 30s, medium for 30s-2m, long for > 2m

Sort participants by total speaking time (descending)

DO NOT

  • Create .md files
  • Provide just a text summary
  • Skip any analysis agent
  • Skip the consolidation step
  • Use inconsistent names across tabs
  • Generate a simple single-tab page
  • Skip the deployment
  • Rush through the analysis - THIS IS IMPORTANT