Meeting Processor
Intelligent meeting transcript processor that auto-detects meeting type and applies type-specific extraction with optional interactive clarification.
When to Use
- After syncing Fathom or Granola transcripts (
/fathom --today,/granola export) - When asked to process, analyze, or summarize a meeting transcript
- When a new meeting transcript appears in the vault root matching
YYYYMMDD-*.md - For coaching sessions, delegate to
coaching-session-summarizerskill instead
Prerequisites
pip install openai pyyamlRequires CEREBRAS_API_KEY environment variable (uses Cerebras API with llama-3.3-70b).
Supported Meeting Types
| Type | Description | Key Extractions |
|---|---|---|
| leadgen | Sales/business development calls | Commitments, pain points, budget, timeline, decision makers, deal stage, sentiment |
| partnership | Collaboration/partnership exploration | Opportunity overview, value proposition, strategic alignment, technical needs, fit assessment |
| coaching | Coaching/mentoring sessions | Insights, decisions, action items, themes, emotional arc, techniques, session quality |
| internal | Internal team meetings | Coming soon |
Usage
Interactive Mode (default)
Run the processor, which auto-detects meeting type and asks clarifying questions:
python3 ~/.claude/skills/meeting-processor/scripts/process.py <transcript-file> --mode interactiveInteractive flow:
- Script analyzes transcript and detects meeting type
- Extracts structured data via LLM
- Identifies missing/ambiguous fields
- Returns questions as JSON (exit code 2 signals interaction needed)
- Parse the JSON between
__INTERACTIVE_QUESTIONS__markers - Use AskUserQuestion to collect answers for each question
- Save answers to a temp JSON file and re-run with
process_with_answers.py
Handling interactive questions:
When the script exits with code 2, parse the output for questions JSON. Each question has:
question: The question textheader: Short label (used as answer key)options: Array of{label, description}for AskUserQuestion
After collecting answers, create two temp files:
questions.json— the original questions context (includespartial_data,meeting_type,transcript_file)answers.json— map of{header_lowercase: selected_label}
Then run:
python3 ~/.claude/skills/meeting-processor/scripts/process_with_answers.py questions.json answers.jsonBatch Mode
Extract only high-confidence information without user interaction:
python3 ~/.claude/skills/meeting-processor/scripts/process.py <transcript-file> --mode batchForce Meeting Type
Skip auto-detection:
python3 ~/.claude/skills/meeting-processor/scripts/process.py <transcript-file> --type leadgen
python3 ~/.claude/skills/meeting-processor/scripts/process.py <transcript-file> --type partnershipOutput
Analysis is appended to the transcript file as a ## Meeting Analysis section. Frontmatter is updated with meeting_type, processed_date, and processing_mode.
Leadgen Output Structure
- Commitments & Actions — with deadlines and owners
- Follow-up — next meeting date if scheduled
- Client Context — pain points, budget, timeline, decision makers
- Deal Assessment — stage (cold/warm/hot), probability (1-5), blocker, sentiment
Partnership Output Structure
- Opportunity — description and value proposition for both sides
- Commitments & Actions — with deadlines and owners
- Follow-up — next meeting date if scheduled
- Partnership Context — strategic alignment, technical needs, resources, challenges
- Opportunity Assessment — fit (strong/medium/weak), readiness, success factors, sentiment