meeting-ingestion

name	meeting-ingestion
description	Meeting ingestion pipeline v3 with calendar triangulation, CRM enrichment, org classification, quality gates, HITL queue, and canonical S-shape generation. Unified CLI for transcript processing from ingest to archive.
compatibility	Created for Zo Computer
metadata	{"author":"<YOUR_HANDLE>.zo.computer","version":"3.2.0","created":"2026-01-26T00:00:00.000Z","rebuilt":"2026-03-17T00:00:00.000Z","last_modified":"2026-04-20T00:00:00.000Z","build_id":"meeting-system-v3"}

Meeting Ingestion Skill v3

Complete meeting transcript processing pipeline with multi-source intake: Google Drive, Fathom, Fireflies, Pocket, and manual transcript import. Unified CLI for transcript processing from ingest to archive, plus intake health checks and historical backfill.

Canonical Processing Model

The current canonical generation surface is S-shape based.

• file 'Skills/meeting-ingestion/scripts/run_shapes.py' is the primary shape generator
• file 'N5/lib/meeting_shapes.py' defines shape selection, filenames, prompts, and post-processors
• file 'Skills/meeting-ingestion/scripts/process.py' as the orchestration bridge that routes gated meetings into canonical shape generation by default

Legacy B-block scripts remain in the repo only for backward compatibility, manual override flows, and migration support. They are not the canonical generation model.

Quick Start

# Full pipeline for new transcript
python3 Skills/meeting-ingestion/scripts/meeting_cli.py ingest /path/to/transcript.md
python3 Skills/meeting-ingestion/scripts/meeting_cli.py identify <meeting-folder>
python3 Skills/meeting-ingestion/scripts/meeting_cli.py gate <meeting-folder>
python3 Skills/meeting-ingestion/scripts/meeting_cli.py process <meeting-folder>
python3 Skills/meeting-ingestion/scripts/meeting_cli.py archive --execute

# Or use orchestrator (processes entire pipeline)
python3 Skills/meeting-ingestion/scripts/meeting_cli.py tick

# Intake management
python3 Skills/meeting-ingestion/scripts/meeting_cli.py check
python3 Skills/meeting-ingestion/scripts/meeting_cli.py backfill --source all --from 2026-02-01 --to 2026-03-10 --dry-run
python3 Skills/meeting-ingestion/scripts/meeting_cli.py drive-import --scan

v3 Pipeline Overview

┌─────────────────────┐
│ Google Drive        │
│   ├─ transcript.md  │
│   └─ meeting.jsonl  │
└─────────┬───────────┘
          │ pull
          ▼
┌─────────────────────┐
│ RAW FILES           │
│ (Inbox root)        │
└─────────┬───────────┘
          │ ingest
          ▼
┌─────────────────────────────────────────┐
│ INGESTED                                │
│  YYYY-MM-DD_Participant/                │
│    ├─ transcript.md                     │
│    └─ manifest.json {status: ingested}  │
└─────────┬───────────────────────────────┘
          │ identify
          ▼
┌─────────────────────────────────────────┐
│ IDENTIFIED                              │
│  YYYY-MM-DD_Participant/                │
│    ├─ transcript.md                     │
│    └─ manifest.json {status: identified}│
│        ├─ participants: [...]           │
│        ├─ meeting_type: external        │
│        └─ calendar_match: {...}         │
└─────────┬───────────────────────────────┘
          │ gate
          ▼
┌─────────────────────────────────────────┐
│ GATED                                   │
│  YYYY-MM-DD_Participant/                │
│    ├─ transcript.md                     │
│    └─ manifest.json {status: gated}     │
│        └─ quality_gate: {passed: true}  │
└─────────┬───────────────────────────────┘
          │ process
          ▼
┌─────────────────────────────────────────┐
│ PROCESSED                               │
│  YYYY-MM-DD_Participant/                │
│    ├─ transcript.md                     │
│    ├─ S01_CONTEXT.md                    │
│    ├─ S02_ACTIONS.md                    │
│    ├─ S03_EVIDENCE.md                   │
│    ├─ S04_PEOPLE.md                     │
│    ├─ S05_WISDOM.md                     │
│    ├─ S06_RECAP.md                      │
│    ├─ S07_ZO_DIRECTIVES.md              │
│    ├─ S08_ZO_INTEL.md                   │
│    ├─ PP_GRAPH_EDGES.jsonl              │
│    └─ manifest.json {status: processed} │
└─────────┬───────────────────────────────┘
          │ closeout + archive
          ▼
┌─────────────────────────────────────────┐
│ ARCHIVED                                │
│  Week-of-YYYY-MM-DD/external/           │
│    └─ YYYY-MM-DD_Participant/           │
└─────────────────────────────────────────┘

Commands Reference

v3 Pipeline Commands (Recommended)

ingest - Raw → Ingested

Convert raw transcripts into meeting folders with v3 manifest.

python3 meeting_cli.py ingest <path> [--dry-run] [--json]

# Examples
python3 meeting_cli.py ingest transcript.md          # Single file 
python3 meeting_cli.py ingest meeting-folder/       # Folder with transcript
python3 meeting_cli.py ingest . --dry-run          # Preview all in current dir

What it does:

• Detects format (JSONL vs markdown vs folder)
• Extracts participants, date, duration from content
• Metadata quality assessment: Detects bare transcripts (no speakers, no participant context) and flags them for HITL review with specific guidance on what's missing
• Creates meeting folder with standard naming: YYYY-MM-DD_Participant-Name
• Generates v3 manifest.json with status: "ingested"
• Moves transcript to standardized transcript.md

identify - Calendar + CRM Enrichment

Match meeting to calendar and enrich participant data from CRM.

python3 meeting_cli.py identify <meeting-folder> [--dry-run] [--json]

# Example
python3 meeting_cli.py identify 2026-01-26_David-x-Acme/

What it does:

• Calendar triangulation: Queries Google Calendar by date/time/title, scores matches
• CRM enrichment: Looks up participants in local CRM database
• Type classification: Determines internal vs external meeting type
• Confidence scoring: Tracks identification confidence levels
• HITL escalation: Queues for manual review if confidence too low
• Updates manifest status: "ingested" → "identified"

gate - Quality Validation

Validate meeting readiness using 16 quality checks.

python3 meeting_cli.py gate <meeting-folder> [--dry-run] [--verbose] [--json]

# Example  
python3 meeting_cli.py gate 2026-01-26_David-x-Acme/ --verbose

Quality checks include:

• Transcript validation: Length, format, encoding, duration consistency
• Participant confidence: ≥0.7 overall, host identified, external participants verified
• Calendar matching: ≥0.6 confidence or manual override
• Type consistency: Meeting type aligns with participant roster

Pass/Fail logic:

• PASS: Overall score ≥0.8, no critical failures
• FAIL + HITL: Participant ID <0.5, encoding corruption, critical failures
• FAIL + RETRY: Transcript too short, temporary API failures

Updates manifest status: "identified" → "gated"

process - Canonical Shape Generation

Generate canonical meeting intelligence shapes for a gated meeting.

python3 meeting_cli.py process <meeting-folder> [--dry-run]

Default behavior:

• Routes into the S-shape pipeline
• Uses run_shapes.py + N5/lib/meeting_shapes.py
• Writes manifest.shapes
• Runs closeout/archive follow-up after successful processing

Compatibility stance:

• Future processing is S-only
• Older B-era meeting files remain readable as historical artifacts
• Any B→S adaptation should happen as a one-time migration, not through live generation flags

Canonical shape families:

• S01 context
• S02 actions / commitments / decisions
• S03 evidence
• S04 people
• S05 wisdom
• S06 recap
• S07 Zo directives
• S08 conditional Zo intelligence
• PP_GRAPH_EDGES post-processor output

tick - Orchestrated Pipeline

Process the full meeting pipeline automatically.

Pipeline sequence:

1. Pull / ingest new transcripts
2. Stage raw files for processing
3. Run identification pipeline on staged meetings
4. Apply quality gates with HITL escalation
5. Process gated meetings through canonical S-shape generation
6. Run closeout + archive for completed meetings

Intake Management Commands

check - All-Points Intake Scan

Scan every meeting intake source and report what's pending or actionable.

python3 meeting_cli.py check [--verbose] [--json]

Sources scanned:

• Local Inbox (Personal/Meetings/Active/): Raw files + queued meetings by status
• Pocket Inbox (Personal/Inbox/): Unprocessed meeting-type items from Pocket webhook
• Google Drive: External Transcripts folders (multiple connected accounts)
• Fathom API: Connection health check, ready for backfill
• Fireflies API: Connection health check, ready for backfill

Output: Per-source status (OK/degraded), actionable item counts, and suggested next commands.

backfill - Historical Meeting Recovery

Pull meetings from Fathom and/or Fireflies APIs for a specific date range.

python3 meeting_cli.py backfill --source <fathom|fireflies|all> --from YYYY-MM-DD --to YYYY-MM-DD [--dry-run] [--limit N]

# Examples
python3 meeting_cli.py backfill --source all --from 2026-02-01 --to 2026-03-10 --dry-run
python3 meeting_cli.py backfill --source fireflies --from 2026-02-01 --to 2026-03-01

What it does:

• Queries the API for all recordings/transcripts in the date range
• Deduplicates against existing meetings in Inbox + archive (fuzzy title + date match)
• Converts to standard transcript.md format with speaker labels
• Drops into Inbox with status: ingested manifest for pipeline processing

Requirements: FATHOM_API_KEY and/or FIREFLIES_API_KEY set in Settings > Advanced.

drive-import - Google Drive External Transcripts

Import transcript files dropped into configured Google Drive folders.

python3 meeting_cli.py drive-import --scan [--json]    # Check for new files
python3 meeting_cli.py drive-import --pull [--dry-run]  # Pull new files into Inbox

Configured folders (in N5/config/drive_locations.yaml):

• [e-mail] → External Transcripts (Business A)
• [e-mail] → External Transcripts (Business B)

What it does:

• --scan: Lists transcript files in Drive folders, highlights files not yet imported
• --pull: Downloads new files to Personal/Meetings/Active/ for pipeline processing
• Tracks imported files in N5/data/drive_import_log.json to avoid re-importing

Legacy Commands (v2 Compatibility)

status - Queue Status

Show current pipeline state across all stages.

python3 meeting_cli.py status

# Output example:
# Raw files (needs ingest): 2
# Ingested (needs identify): 1  
# Identified (needs gate): 3
# Gated (needs process): 1
# Processed (needs archive): 4
# Total archived: 847

archive - Move to Weekly Folders

Move completed meetings to Week-of-YYYY-MM-DD/internal|external/ structure.

python3 meeting_cli.py archive [--execute] [--batch-size N]

# Preview (default)
python3 meeting_cli.py archive

# Execute  
python3 meeting_cli.py archive --execute

stage - Legacy v2 Staging

Convert v2-style raw files to folder structure (compatibility mode).

python3 meeting_cli.py stage [--dry-run]

fix - Repair Malformed State

Quarantine orphaned files and repair Inbox state.

python3 meeting_cli.py fix [--dry-run]

Canonical Shape Families

Meeting Ingestion v3 now emits a stable S-shape bundle for future meetings.

• S01 - Context
• S02 - Actions / commitments / decisions
• S03 - Evidence
• S04 - People
• S05 - Wisdom
• S06 - Recap
• S07 - Zo directives
• S08 - Conditional Zo intelligence
• PP_GRAPH_EDGES - Post-processor graph output

B00 (Deferred Intents) triggered by:

• "zo take heed", "zo, remember", "zo, follow up on", "zo, note that"

HITL Queue System

When automated processing cannot complete reliably, items are escalated to the Human-In-The-Loop (HITL) queue for V's review.

Queue Location

N5/review/meetings/hitl-queue.jsonl

Escalation Triggers

• Participant identification confidence <0.5 (unknown speakers)
• Calendar match confidence <0.6 (no clear calendar match)
• Meeting topic unclear (ambiguous purpose)
• Encoding corruption (transcript quality issues)
• Duplicate detection uncertainty (potential duplicate meeting)

Priority Levels

• P0 (Immediate): System errors, encoding corruption, critical stakeholder failures
• P1 (Standard): Low participant confidence, calendar match failures
• P2 (Batch): Minor validation issues, borderline scores

SMS Notifications

V receives SMS notifications for:

• P0 items (immediately)
• P1 items (hourly digest)
• Weekly digest of all HITL activity

Quality Gate Specification

The quality gate performs 16 distinct checks across 4 pipeline stages:

Pre-processing Stage (4 checks)

• transcript_length: ≥300 characters actual content
• transcript_format: Valid UTF-8, speaker indicators
• meeting_duration_consistency: Word count within 50% of expected
• transcript_encoding: UTF-8 compliant, no artifacts

Identification Stage (5 checks)

• participant_confidence: Overall confidence ≥0.7
• external_participant_verification: External meetings have ≥1 non-V participant
• calendar_match_score: Match confidence ≥0.6 or manual override
• meeting_type_consistency: Type aligns with participant roster

Block Generation Stage (4 checks)

• block_count_reasonable: Generated blocks within expected range
• block_quality_score: LLM quality assessment ≥0.7
• missing_critical_blocks: All always-blocks generated successfully
• block_file_integrity: All block files valid markdown with frontmatter

Post-processing Stage (3 checks)

• archive_readiness: All required metadata present
• duplicate_detection: No conflicts with existing meetings
• pipeline_completion: All pipeline stages completed successfully

Pass/Fail Logic

• PASS: Overall score ≥0.8, no critical failures, no HITL escalations
• FAIL + HITL: Critical failures requiring human intervention
• FAIL + RETRY: Temporary issues (API failures, short transcripts)
• PASS with WARNING: Optional checks failed but core requirements met

Manifest v3 Schema

Every meeting folder contains a manifest.json following the v3 schema:

{
  "manifest_version": "3.0",
  "meeting_id": "2026-01-26_David-x-PartnerOrg",
  "date": "2026-01-26",
  "org": "business_a",
  "org_classification": {
    "org": "business_a",
    "confidence": 0.95,
    "signals": ["business_a", "partnership"],
    "classified_at": "2026-01-26T10:35:00Z"
  },
  "participants": [
    {"name": "David", "role": "external", "confidence": 0.95},
    {"name": "V", "role": "host", "confidence": 1.0}
  ],
  "status": "processed",
  
  "source": {
    "format": "markdown", 
    "original_path": "/path/to/transcript.md",
    "ingested_at": "2026-01-26T10:30:00Z"
  },
  
  "calendar_match": {
    "matched": true,
    "confidence": 0.85,
    "event_id": "abc123",
    "event_title": "David x PartnerOrg Partnership"
  },
  
  "crm_data": {
    "david": {
      "company": "InvestCorp",
      "title": "Partner", 
      "last_interaction": "2025-12-15",
      "relationship": "business"
    }
  },
  
  "quality_gate": {
    "passed": true,
    "score": 0.92,
    "checks": {
      "has_transcript": true,
      "participants_identified": true,
      "meeting_type_determined": true, 
      "no_hitl_pending": true
    },
    "executed_at": "2026-01-26T10:45:00Z"
  },
  
  "blocks": {
    "recipe": "external_standard",
    "requested": ["B00", "B01", "B02_B05", "B03", "B08", "B21", "B26"],
    "generated": ["B00", "B01", "B02_B05", "B03", "B08", "B21", "B26"],
    "failed": [],
    "conditional_selected": ["B06", "B28"],
    "conditional_skipped": ["B04", "B07", "B10", "B13", "B14", "B25", "B32", "B33"],
    "triggered": []
  },
  
  "hitl": {
    "queue_id": null,
    "reason": null, 
    "resolved_at": null
  },
  
  "timestamps": {
    "created_at": "2026-01-26T10:30:00Z",
    "ingested_at": "2026-01-26T10:30:00Z", 
    "identified_at": "2026-01-26T10:35:00Z",
    "gated_at": "2026-01-26T10:45:00Z",
    "processed_at": "2026-01-26T11:15:00Z",
    "archived_at": null
  }
}

State Machine

Meeting processing follows a strict state progression:

Status	Description	Next Action
raw	Unprocessed file in Inbox root	→ ingest
ingested	Folder created, transcript standardized	→ identify
routed	Arrived via Pocket webhook (treated as ingested)	→ identify
identified	Participants + calendar + CRM + org data added	→ gate
gated	Quality validation passed	→ process
processed	Intelligence blocks generated	→ archive
archived	Moved to YYYY/MM-Month/week-NN/ folder	✓ Complete
hitl_pending	Escalated to HITL queue	→ human review
failed	Unrecoverable error	→ manual intervention

State transitions are tracked in manifest.json with timestamps for audit purposes.

Directory Structure

Skills/meeting-ingestion/
├── SKILL.md                    # This file
├── scripts/                    # CLI and processing modules
│   ├── meeting_cli.py          # Unified CLI v3
│   ├── ingest.py              # Raw → ingested conversion
│   ├── calendar_match.py       # Google Calendar triangulation
│   ├── crm_enricher.py        # CRM participant enrichment  
│   ├── org_classifier.py      # Org category classification (business_a / business_b / professional / personal)
│   ├── quality_gate.py        # Quality validation (16 checks)
│   ├── process.py             # Canonical S-shape orchestration
│   ├── run_shapes.py          # Shape router + shape writers
│   ├── hitl.py                # HITL queue management
│   ├── archive.py             # Weekly folder archival
│   ├── pull.py                # Google Drive download
│   ├── stage.py               # Legacy v2 compatibility
│   ├── validate_manifest.py   # Manifest schema validation
│   └── manifest_converter.py  # v2 → v3 migration
├── references/                 # Documentation and specs  
│   ├── manifest-v3.schema.json     # JSON Schema for manifest
│   ├── meeting-id-convention.md    # Naming standards
│   ├── quality-harness-checks.md   # Quality gate specification
│   ├── hitl-queue-spec.md          # HITL queue schema
│   ├── quality-gate-docs.md        # Quality validation docs
│   └── quality-thresholds-rationale.md # Threshold rationale
└── assets/                     # (none currently)

Configuration

Drive Locations

Drive folder IDs in N5/config/drive_locations.yaml:

meetings:
  transcripts_inbox: "YOUR_FOLDER_ID"

Meeting Registry

Meeting statistics and deduplication tracking:

• Location: N5/data/meeting_registry.db (SQLite)
• Purpose: Track all processed meetings, detect duplicates, maintain statistics
• Schema: meeting_id, date, participants, type, processing metadata, archived_location

CRM Database

Participant information storage:

• Location: Personal/Knowledge/CRM/crm.db (migrated from Legacy_Inbox)
• Purpose: Participant lookup, relationship classification, contact enrichment
• Integration: Automatic lookup during identification stage

Typical Workflows

Automated Processing (Every 30 Minutes)

The Meeting Orchestrator v3 scheduled agent runs every 30 minutes using GPT 5.4:

# Agent runs:
python3 Skills/meeting-ingestion/scripts/meeting_cli.py tick --auto-process --batch-size 5
python3 Skills/meeting-ingestion/scripts/meeting_cli.py archive --execute

This processes the complete pipeline:

1. Ingest raw files into meeting folders
2. Identify participants via calendar + CRM
3. Classify org category (business_a / business_b / professional / personal)
4. Validate quality and escalate to HITL if needed
5. Generate canonical meeting shapes
6. Archive completed meetings to YYYY/MM-Month/week-NN/ folders

Processing order is LIFO (newest first) so recent meetings are prioritized over backlog. If no pending meetings exist, the agent exits immediately.

Manual Single Meeting Processing

# Process a specific raw transcript
cd /home/workspace
python3 Skills/meeting-ingestion/scripts/meeting_cli.py ingest transcript.md

# Continue pipeline manually  
MEETING=$(ls -1 Personal/Meetings/Active/2026-*/ | tail -1)
python3 Skills/meeting-ingestion/scripts/meeting_cli.py identify "$MEETING"
python3 Skills/meeting-ingestion/scripts/meeting_cli.py gate "$MEETING" --verbose
python3 Skills/meeting-ingestion/scripts/meeting_cli.py process "$MEETING" 
python3 Skills/meeting-ingestion/scripts/meeting_cli.py archive --execute

HITL Queue Management

# Check queue status
python3 Skills/meeting-ingestion/scripts/hitl.py status

# Resolve HITL item (after manual review)
python3 Skills/meeting-ingestion/scripts/hitl.py resolve HITL-20260126-001 --resolution "participant_identified" --data '{"participant": "David Chen", "company": "InvestCorp"}'

# Clear resolved items  
python3 Skills/meeting-ingestion/scripts/hitl.py cleanup --older-than 7d

Troubleshooting

Common Issues

"No transcript found in meeting folder"

• Ensure transcript file exists as transcript.md in meeting folder
• Run ingest first if you have a raw file
• Check file permissions and encoding

"Participant identification failed"

• Check if speakers are clearly labeled in transcript (e.g., "John:", "V:")
• Review CRM database for participant entries
• Low confidence triggers HITL escalation (expected behavior)

"Quality gate failed"

• Run with --verbose to see which specific checks failed
• Common failures: transcript too short (<300 chars), encoding issues, missing participants
• Check HITL queue for escalated items: hitl.py status

"Shape generation failed"

• Verify ZO_CLIENT_IDENTITY_TOKEN environment variable is set
• Check if meeting passed quality gate first
• Re-run the canonical processor: meeting_cli.py process <meeting>

"Calendar match confidence too low"

• Verify Google Calendar integration is working
• Check if meeting date/time are extracted correctly from transcript
• Low confidence triggers HITL escalation for manual calendar linking

"Meeting stuck in processing state"

• Check manifest.json for error details
• Re-run specific pipeline stage: meeting_cli.py process <meeting>
• Check logs for shape generation failures

Pipeline State Recovery

Reset meeting to earlier stage:

# Reset to identification stage (if processing failed)
python3 Skills/meeting-ingestion/scripts/validate_manifest.py <meeting> --reset-to identified

# Reset to ingested stage (if identification failed)  
python3 Skills/meeting-ingestion/scripts/validate_manifest.py <meeting> --reset-to ingested

Repair orphaned files:

# Quarantine orphaned blocks and repair structure
python3 meeting_cli.py fix --dry-run      # Preview
python3 meeting_cli.py fix                # Execute

Validate manifest integrity:

# Check manifest against v3 schema
python3 Skills/meeting-ingestion/scripts/validate_manifest.py <meeting>/manifest.json

# Batch validate all meetings in Inbox
python3 Skills/meeting-ingestion/scripts/validate_manifest.py Personal/Meetings/Active/ --recursive

Error Codes

Code	Meaning	Action
INGEST_001	Invalid transcript format	Check file encoding and content
IDENTIFY_002	Calendar API failure	Retry later or manually link
GATE_003	Quality threshold not met	Review checks, possibly escalate to HITL
PROCESS_004	Block generation API failure	Check ZO_CLIENT_IDENTITY_TOKEN
ARCHIVE_005	Naming collision detected	Resolve manually or use collision suffix
HITL_006	HITL queue item requires attention	Review HITL queue and resolve

Migration from v2

The v3 system is fully backward compatible with v2 meeting folders. Legacy folders are automatically detected and converted:

Key Changes from v2

State Management:

• v2: Folder name suffixes (_[P], _[B], etc.)
• v3: manifest.json status field

Block Selection:

• v2: Fixed block lists based on meeting type
• v3: Smart LLM-powered selector with conditional blocks

Participant Identification:

• v2: Manual or heuristic-based
• v3: Calendar triangulation + CRM enrichment with confidence scoring

Quality Assurance:

• v2: Basic validation
• v3: 16-check quality gate with HITL escalation

Pipeline Stages:

• v2: pull → stage → process → archive
• v3: ingest → identify → gate → process → archive

Migration Process

Existing v2 folders are automatically converted when processed:

1. Detection: System detects v2 folder structure (missing manifest.json)
2. Conversion: manifest_converter.py creates v3 manifest from folder state
3. Backfill: Missing data (calendar, CRM) is populated where possible
4. Validation: Converted manifest passes through quality gate
5. Processing: Continues with v3 pipeline

No manual intervention required - migration happens seamlessly during normal processing.

Performance and Limits

Batch Sizes

• Default batch size: 5 meetings per operation
• Maximum recommended: 10 meetings (prevent API rate limits)
• HITL queue: No batch limit (items processed individually)

API Rate Limits

• Google Calendar: 1000 requests/day (100 meetings)
• Zo Ask API: 50 requests/hour per meeting (10-15 blocks avg)
• Automatic retries: 3 attempts with exponential backoff

Storage Estimates

• Average meeting: 2-5 MB (transcript + 7-12 blocks)
• Annual storage: ~1-2 GB (500 meetings/year)
• Archive compression: Weekly folders reduce directory traversal

Quality Gate Performance

• Processing time: 2-3 seconds per meeting
• Pass rate: ~85% (15% escalate to HITL)
• Common failures: Short transcripts (8%), low participant confidence (5%), calendar mismatch (2%)

---

System Dependencies

• Python 3.8+ with pathlib, json, sqlite3
• Google Calendar API access (for calendar triangulation)
• Zo Ask API access (ZO_CLIENT_IDENTITY_TOKEN required)
• Local CRM database (SQLite, automatically created if missing)
• Meeting registry (SQLite, automatically maintained)

Related Systems

• Meeting Orchestrator v3: Scheduled agent (daily 6 AM ET) that runs tick command
• HITL Review System: Manual review queue at N5/review/meetings/hitl-queue.jsonl
• Block Prompt Library: Centralized prompts at Prompts/Blocks/
• Content Library: Processed blocks can be ingested for knowledge management
• Meeting Registry: Statistics and deduplication database

---

Meeting Ingestion Skill v3.0.0 | Built February 2026 | Build ID: meeting-system-v3