Email Integration

Email integration connects Elmar's Microsoft 365 Outlook account to the Luci system via the MS Graph API. Emails are fetched, classified by project, summarized, and stored in a searchable SQLite database.

Components

Component	Path	Purpose
`graph_api.py`	`~/workspace/scripts/graph_api.py`	MS Graph API client (auth, email, calendar, files)
`email_sync.py`	`~/workspace/scripts/email_sync.py`	Daily email sync pipeline
`email_backfill.py`	`~/workspace/scripts/email_backfill.py`	Backfill summaries for existing emails
`email_attachment_sync.py`	`~/workspace/scripts/email_attachment_sync.py`	Download attachments for synced emails
`email.db`	`~/workspace/email.db`	SQLite database of classified emails
`email-index` skill	`~/.claude/skills/email-index/`	Query interface for searching email.db

MS Graph API Client (graph_api.py)

Self-contained Python script with zero external dependencies (stdlib only). Handles its own OAuth via device code flow.

Authentication

OAuth flow: Device code (no redirect URI needed)
Token cache: ~/.graph-api-token.json
Auto-refresh: Tokens refresh automatically; manual re-login via python graph_api.py login if refresh fails
Legacy migration: Can migrate tokens from the old m365 CLI MSAL cache (~/.cli-m365-msal.json)
App ID: Registered Azure AD application with scopes for Mail, Calendar, Files, Tasks, Teams

Available Commands

python graph_api.py login                          # One-time auth
python graph_api.py list-messages [--since DATE]   # List emails
python graph_api.py read-message <id>              # Read full email
python graph_api.py send-message --to ... --subject ... --body ...
python graph_api.py create-reply <id> --body ...   # Draft reply
python graph_api.py send-draft <id>                # Send a draft
python graph_api.py search --query "keyword"       # Search Outlook
python graph_api.py search-files --query "..."     # OneDrive/SharePoint search
python graph_api.py calendar [--start ...] [--end ...]
python graph_api.py upload-to-onedrive <local> <remote>

email.db Schema

Column	Type	Notes
`internet_message_id`	TEXT UNIQUE	RFC immutable key (survives moves/archive)
`conversation_id`	TEXT	Thread grouping
`subject`	TEXT	Email subject
`sender_email`	TEXT	Sender address
`sender_name`	TEXT	Display name
`received_date`	TEXT	ISO date
`project`	TEXT	Classified project name
`summary`	TEXT	1-2 sentence summary
`confidence`	TEXT	high/medium/low
`has_attachments`	INTEGER	0 or 1

Key design decision: No email bodies are stored in email.db -- only metadata + summaries
Full body retrieval: Use graph_api.py read-message <graph-message-id> to fetch the full content on demand from Graph API

Daily Email Sync Pipeline

See 12-data-flows/email-pipeline for the full flow diagram.

Scheduler task: email-sync (runs at 07:00 UTC weekdays)
Pipeline:
Fetch emails from last 2 days via Graph API
Filter noise using inbox rules (cowork/context/inbox-rules.md)
Classify each email to a project
Generate 1-sentence summary
Insert into email.db via the email-index skill
Trigger attachment sync for each email (best-effort)

Noise Filtering

Inbox rules are parsed from inbox-rules.md: - Sender blocklist: Exact email matches and wildcard patterns (e.g., *@zapiermail.com) - Subject skip patterns: Auto-skip emails with specific subject keywords

Project Classification

Sender-to-project mapping: Built from existing email.db data; cached in .email-sender-map.json
Gemini Flash: Used for classification + summarization (~$0.01/day cost)
Known projects: ~25 canonical project names (Heron, Finance, Harith Aviation, Fleet, etc.)

Email Backfill

email_backfill.py enriches existing email.db records that lack summaries: 1. Fetches emails without summaries from email.db 2. Reads body from Graph API by internet_message_id 3. Sends to Gemini Flash for classification + summary 4. Updates email.db with results - Rate limited: 1s between Graph API calls, batched Gemini calls

Searching Emails

Via email-index skill (recommended)

python ~/.claude/skills/email-index/query.py project "Project Heron"
python ~/.claude/skills/email-index/query.py search "MOI harith"
python ~/.claude/skills/email-index/query.py sender "stephan.spamer"
python ~/.claude/skills/email-index/query.py thread "AAQkAD..."
python ~/.claude/skills/email-index/query.py stats
python ~/.claude/skills/email-index/query.py recent --limit 10

Via Graph API (live search)

python graph_api.py search --query "subject:keyword" --top 20

When to Use What

Need	Tool
Past emails by project/person	`email-index` skill (queries email.db)
Full email body	email.db lookup then `graph_api.py read-message <graph-message-id>`
Live inbox / new emails	`inbox-assist` skill
Reply / draft / send	`inbox-assist` or `graph_api.py`
Ad-hoc Outlook search	`graph_api.py search`

Key Takeaways

Email integration uses the MS Graph API with device-code OAuth -- no browser needed
email.db stores metadata and summaries only, never full email bodies
The daily sync pipeline classifies emails to projects and generates summaries via Gemini Flash
The internet_message_id (RFC standard) is the primary key -- it survives email moves and archiving
Two search paths: email.db for historical classified emails, Graph API for live/ad-hoc searches

Wiki