talk-aggregation

name	talk-aggregation
description	Analyzes multiple talks to extract themes, expertise areas, and CNCF project focus
version	1.0.0

Talk Aggregation Skill

Purpose

Analyze multiple presenter talks to identify expertise patterns, recurring themes, CNCF project focus, and speaking statistics. This skill synthesizes data across all talks to create a comprehensive view of the presenter's technical focus and community contributions.

Input Format

{
  "talks": [
    {
      "video_id": "abc123",
      "title": "Kubernetes Community Management Best Practices",
      "date": "2025-10-15",
      "duration": 2700,
      "transcript": "Full transcript text...",
      "description": "Jeffrey discusses community management...",
      "event": "KubeCon North America 2025"
    },
    {
      "video_id": "def456",
      "title": "GitOps with Argo CD",
      "date": "2024-05-20",
      "duration": 1320,
      "transcript": "Full transcript text...",
      "description": "Introduction to GitOps practices...",
      "event": "CloudNativeCon Europe 2024"
    }
  ]
}

Field Descriptions:

• talks (array): List of all talks by the presenter
  • video_id (string): YouTube video identifier
  • title (string): Talk title
  • date (string): Presentation date (YYYY-MM-DD format)
  • duration (number): Talk length in seconds
  • transcript (string): Full corrected transcript
  • description (string): YouTube video description
  • event (string, optional): Conference or event name

Output Format

{
  "expertise_areas": [
    {
      "area": "Kubernetes",
      "context": "Deep community involvement, contributor experience, scalability discussions",
      "talk_count": 5,
      "evidence": ["Container orchestration patterns", "Kubernetes governance", "SIG management"]
    },
    {
      "area": "Community Management",
      "context": "Best practices for open source communities, contributor engagement, sustainability",
      "talk_count": 3,
      "evidence": ["Building inclusive communities", "Maintainer burnout", "Governance models"]
    }
  ],
  "cncf_projects" 
    
       
       
       
       
       
    
    
       
       
       
       
       
    
  
   
    
    
    
    
  
   
    
       
       
       
        
        
        
      
         
    
  
   
     
     
       
       
       
    
     
     
     
  

Field Descriptions:

• expertise_areas (array): Technical or domain expertise identified
  • area (string): Expertise domain name
  • context (string): Description of expertise manifestation
  • talk_count (number): Number of talks covering this area
  • evidence (array of strings): Specific topics demonstrating expertise
• cncf_projects (array): CNCF projects discussed across talks
  • name (string): Official CNCF project name
  • talk_count (number): Number of talks mentioning project
  • usage_context (string): How project is discussed/used
  • first_mention (string): Earliest talk date (YYYY-MM format)
  • latest_mention (string): Most recent talk date (YYYY-MM format)
• recurring_themes (array of strings): Cross-talk themes and patterns
• talk_summaries (array): Concise summary for each talk
  • video_id (string): Video identifier
  • summary (string): 50-150 word talk summary
  • key_points (array of strings): 3-5 main takeaways
  • topics (array of strings): Primary topics covered
• stats (object): Speaking activity statistics
  • total_talks (number): Total presentation count
  • years_active (object): Speaking timeframe
  • total_speaking_minutes (number): Sum of all talk durations
  • most_active_year (number): Year with most presentations
  • average_talk_length_minutes (number): Mean talk duration

Execution Instructions

Step 1: Analyze Individual Talks

For each talk, extract:

Topics Covered:

• Main technical subjects
• CNCF projects mentioned
• Problem domains addressed
• Solutions discussed

Key Points:

• Main arguments or lessons
• Novel insights
• Practical recommendations
• Case studies or examples

CNCF Projects:

• Project names (exact capitalization)
• How projects are used/discussed
• Context: implementation, comparison, case study
• Relationship between projects

Technical Depth:

• Implementation details
• Architecture patterns
• Specific features discussed
• Production experiences

Step 2: Identify Expertise Areas

Look across all talks to identify recurring technical domains:

Criteria for Expertise Area:

• Appears in 2+ talks OR
• Central to 1 major talk (>30 min) with deep technical content
• Demonstrable knowledge depth (not just mentions)

Common Expertise Areas:

• Specific technologies (Kubernetes, Prometheus, Istio)
• Technical domains (Observability, Security, Networking)
• Practices (GitOps, SRE, Platform Engineering)
• Organizational topics (Community Management, Developer Experience)

For each area, determine:

• Context: How expertise manifests (architecture, operations, community, etc.)
• Talk count: Number of talks covering this area
• Evidence: Specific subtopics that demonstrate expertise

Example:

{
  "area": "Observability",
  "context": "Distributed tracing, metrics collection, monitoring best practices for microservices",
  "talk_count": 4,
  "evidence": [
    "Prometheus query optimization",
    "Jaeger deployment patterns",
    "OpenTelemetry instrumentation",
    "SLO/SLI definition"
  ]
}

Step 3: Track CNCF Projects

Common CNCF Projects:

• Kubernetes, Prometheus, Envoy, CoreDNS, containerd
• Fluentd, Jaeger, Helm, Argo, Flux, Vitess
• Cilium, Linkerd, Istio, etcd, Harbor, Falco
• Dragonfly, Rook, TiKV, gRPC, CNI, Knative
• OpenTelemetry, SPIFFE, SPIRE, Cortex, Thanos

For each CNCF project mentioned:

1. Count talks mentioning project
2. Extract usage context - How is it used/discussed?
   • Implementation ("deployed Istio for service mesh")
   • Comparison ("evaluated Linkerd vs Istio")
   • Case study ("Prometheus handles 10M metrics/sec")
   • Tutorial ("how to configure Helm charts")
3. Track timeline
   • First mention date (YYYY-MM)
   • Latest mention date (YYYY-MM)

Prioritization:

• Order by talk_count (descending)
• Include all projects mentioned in 2+ talks
• Include projects central to 1 major talk

Step 4: Identify Recurring Themes

Look for conceptual patterns across talks:

Theme Types:

Technical Patterns:

• "Scalable architecture design"
• "Progressive delivery strategies"
• "Multi-cluster management"

Organizational:

• "Developer experience optimization"
• "Platform engineering approaches"
• "Inner source adoption"

Community/Cultural:

• "Open source sustainability"
• "Building inclusive communities"
• "Contributor engagement"

Operational:

• "Production reliability practices"
• "Cost optimization strategies"
• "Security-first architecture"

Criteria for Theme:

• Appears in 3+ talks OR
• Central message of 2+ talks
• Represents higher-level concept (not specific tool)

Format: Short phrase capturing the theme (5-8 words)

Step 5: Generate Talk Summaries

For each talk, write a 50-150 word summary:

Structure:

1. Opening sentence: Main topic/focus (what is this talk about?)
2. Body (2-3 sentences): Key insights, approaches, or findings
3. Closing: Outcome, recommendation, or takeaway

Extract 3-5 key points:

• Concrete takeaways
• Actionable recommendations
• Notable insights or findings
• Measurable results if applicable

Identify 3-5 primary topics:

• Specific technologies
• Domain areas
• Problem/solution categories

Example:

{
  "video_id": "xyz789",
  "summary": "Explores the challenges of scaling Kubernetes clusters beyond 1000 nodes, focusing on etcd performance, scheduler optimization, and network plugin selection. Presents real-world case studies from managing a 3000-node cluster, including lessons learned about control plane architecture and monitoring strategies. Provides actionable recommendations for organizations planning large-scale Kubernetes deployments.",
  "key_points": [
    "etcd performance becomes critical above 1000 nodes",
    "Custom scheduler plugins reduce pod scheduling latency",
    "CNI choice significantly impacts network performance",
    "Control plane HA requires 5+ etcd members",
    "Monitoring overhead grows non-linearly with scale"
  ],
  "topics": ["Kubernetes", "Scalability", "etcd", "Networking", "Performance"]
}

Step 6: Calculate Statistics

Total Talks:

• Count of talks in input array

Years Active:

• Extract year from each talk date
• first: Earliest year
• latest: Most recent year
• span: latest - first (years active)

Total Speaking Minutes:

• Sum all talk durations (in seconds)
• Convert to minutes: sum(durations) / 60
• Round to nearest integer

Most Active Year:

• Count talks per year
• Return year with highest count
• If tie, use most recent year

Average Talk Length:

• total_speaking_minutes / total_talks
• Round to nearest integer

Examples

Example 1: Multi-Topic Speaker

Input:

{
  "talks": [
    {
      "video_id": "k8s001",
      "title": "Scaling Kubernetes to 5000 Nodes",
      "date": "2025-06-15",
      "duration": 2400,
      "transcript": "...discusses etcd performance, scheduler optimization...",
      "description": "Deep dive into large-scale Kubernetes"
    },
    {
      "video_id": "gitops002",
      "title": "GitOps with Flux and Argo CD",
      "date": "2025-03-20",
      "duration": 1800,
      "transcript": "...compares Flux and Argo CD for continuous delivery...",
      "description": "GitOps patterns and tooling comparison"
    
    
       
       
       
       
       
       
    
  

Output:

{
  "expertise_areas": [
    {
      "area": "Kubernetes",
      "context": "Large-scale cluster management, networking, and performance optimization",
      "talk_count": 2,
      "evidence": [
        "Scaling beyond 1000 nodes",
        "etcd performance tuning",
        "CNI selection and optimization",
        "eBPF-based networking"
      ]
    },
    {
      "area": "GitOps",
      "context": "Continuous delivery patterns, tooling evaluation, and best practices",
      "talk_count": 1,
      "evidence": [
        "Flux vs Argo CD comparison",
        "Declarative deployment workflows",
        "Git-based infrastructure management"
      ]
    
  
   
    
       
       
       
       
       
    
    
       
       
       
       
       
    
    
       
       
       
       
       
    
    
       
       
       
       
       
    
  
   
    
    
    
  
   
    
       
       
       
        
        
        
        
        
      
           
    
    
       
       
       
        
        
        
        
        
      
           
    
    
       
       
       
        
        
        
        
        
      
           
    
  
   
     
     
       
       
       
    
     
     
     
  

Example 2: Specialized Speaker

Input:

{
  "talks": [
    {
      "video_id": "obs001",
      "title": "OpenTelemetry in Production",
      "date": "2025-09-12",
      "duration": 1920,
      "transcript": "...implementing distributed tracing at scale...",
      "description": "OpenTelemetry adoption journey"
    },
    {
      "video_id": "obs002",
      "title": "Prometheus Query Optimization",
      "date": "2025-05-18",
      "duration": 1680,
      "transcript": "...optimizing PromQL queries for large datasets...",
      "description": "Performance tuning for Prometheus"
    
    
       
       
       
       
       
       
    
  

Output:

{
  "expertise_areas": [
    {
      "area": "Observability",
      "context": "Distributed tracing, metrics collection, monitoring best practices, and organizational culture",
      "talk_count": 3,
      "evidence": [
        "OpenTelemetry production deployment",
        "Prometheus query optimization",
        "SLO/SLI definition",
        "Observability-driven development",
        "Cross-team instrumentation standards"
      ]
    }
  ],
  "cncf_projects": [
    {
      "name": "OpenTelemetry",
      "talk_count": 1,
      "usage_context": "Distributed tracing and telemetry collection at scale",
      "first_mention": "2025-09"
       
    
    
       
       
       
       
       
    
  
   
    
    
    
  
   
    
       
       
       
        
        
        
        
        
      
          
    
    
       
       
       
        
        
        
        
        
      
          
    
    
       
       
       
        
        
        
        
        
      
           
    
  
   
     
     
       
       
       
    
     
     
     
  

Quality Guidelines

Expertise Area Identification

• Minimum 2 talks for expertise claim (or 1 deep technical talk)
• Evidence-based: List specific subtopics demonstrating depth
• Avoid over-generalization: "Kubernetes" not "Cloud Computing"
• Context matters: Explain how expertise manifests

CNCF Project Tracking

• Use official names: "Argo CD" not "ArgoCD"
• Accurate context: Describe actual usage, not generic descriptions
• Timeline precision: Use YYYY-MM format from talk dates
• Prioritize by frequency: Order by talk_count descending

Recurring Theme Detection

• Higher-level concepts: Not specific tools, but patterns/practices
• Consistent across talks: Appears in multiple presentations
• Concise phrasing: 5-8 words max
• Actionable or descriptive: Clear meaning

Talk Summary Quality

• Length: 50-150 words (strict)
• Factual: Based on transcript/description content
• Specific: Concrete topics, not vague overviews
• Key points: 3-5 actionable takeaways
• Topics: 3-5 specific subject areas

Statistics Accuracy

• Count carefully: Verify total_talks matches input length
• Date parsing: Handle various date formats
• Duration conversion: Seconds to minutes, rounded
• Most active year: Handle ties (use most recent)

Common Pitfalls to Avoid

❌ Claiming Expertise Without Evidence

Bad:

{
  "area": "Cloud Native Architecture",
  "context": "General cloud-native knowledge",
  "talk_count": 1,
  "evidence": []
}

Why: Too broad, insufficient evidence

Good:

{
  "area": "Service Mesh Architecture",
  "context": "Istio implementation, sidecar patterns, traffic management",
  "talk_count": 2,
  "evidence": ["Istio deployment strategies", "mTLS configuration", "Traffic routing patterns"]
}

❌ Incorrect Project Names

Bad: "ArgoCD", "K8s", "OTel"

Good: "Argo CD", "Kubernetes", "OpenTelemetry"

❌ Vague Usage Context

Bad:

{
  "name": "Prometheus",
  "usage_context": "Used for monitoring"
}

Good:

{
  "name": "Prometheus",
  "usage_context": "Metrics collection, query optimization, large-scale time-series data management"
}

❌ Too Many Themes

Bad: 15 themes for 5 talks (over-segmented)

Good: 4-6 major themes that genuinely recur

❌ Summary Too Long or Too Short

Bad (too short): "This talk is about Kubernetes networking."

Bad (too long): 300-word detailed description

Good: 75-100 word focused summary with key insights

❌ Incorrect Statistics

Bad: Claiming 10 talks when input has 8

Good: Count matches input array length exactly

❌ Missing Evidence for Expertise

Bad:

{
  "area": "Security",
  "evidence": []
}

Good:

{
  "area": "Security",
  "evidence": ["Zero-trust architecture", "mTLS implementation", "RBAC policies", "Secrets management"]
}

Important Notes

• This skill feeds into presenter-profile-generation skill
• Statistics drive the stats table in presenter profiles
• Expertise areas and themes become profile narrative content
• Talk summaries populate the "Talk Highlights" section
• CNCF project data identifies presenter's technical focus
• Quality here determines profile depth and accuracy

Validation Checklist

Before returning output, verify:

• All expertise areas have 2+ talks (or 1 deep talk)
• Evidence array populated for each expertise area
• CNCF project names use official capitalization
• Project timelines use YYYY-MM format
• Projects ordered by talk_count descending
• Recurring themes are higher-level concepts
• Each theme appears in 3+ talks (or 2+ major talks)
• Talk summaries are 50-150 words each
• Key points are 3-5 per talk
• Topics are 3-5 per talk
• total_talks matches input array length
• years_active span is calculated correctly
• total_speaking_minutes is sum of durations / 60
• most_active_year is year with most talks
• average_talk_length is total_minutes / total_talks
• All statistics are integers (rounded)