Overview
Use this skill to turn local Outlook mail into structured work intelligence on Windows PCs without Graph or Exchange dependencies. The skill supports live Outlook desktop access, .msg imports, and .pst archive imports, then stores normalized records in a local workspace and SQLite index for reporting and follow-up workflows.
Deployment hygiene:
- never point
--workspaceat a path inside this skill directory - keep workspaces under a user-private path such as
%USERPROFILE%\\Documents\\outlook-mail-assistant-runtime - treat exported JSONL, SQLite, CSV, XLSX, DOCX, and audit DB files as sensitive mail data
Core Workflow
- Select the ingestion surface:
- live Outlook desktop profile
.msgfile or directory.pstarchive
- Normalize messages into the canonical schema described in
references/canonical-schema.md. - Persist records to:
- workspace artifacts for transparency
- SQLite for search, reporting, and auditability
- Extract:
- weekly and monthly summaries
- tasks, deadlines, decisions, and follow-ups
- Ask one focused question at a time when:
- semantic intent is ambiguous
- an action would materially change mailbox, task, or calendar state
- Produce outputs:
- Markdown reports
- DOCX reports via the bundled converter in this repo
- XLSX or CSV outputs via the bundled exporters in this repo
- Outlook tasks or calendar items
Execution Boundaries
- Do not silently perform high-impact actions.
- Always ask before sending, deleting, bulk-moving, bulk-archiving, bulk-categorizing, or creating Outlook items from inferred intent.
- Record explicit facts, inferred facts, user-confirmed facts, and executed actions in the local audit trail.
- Prefer dry-run execution for automation scripts during testing and validation.
Reference Files
- Read
references/canonical-schema.mdbefore changing normalization rules. - Read
references/sqlite-schema.mdbefore changing database tables.
Script Surface
Primary Python modules live under scripts/outlook_mail_assistant/.
canonical_schema.py- canonical record model
- stable dedupe hashing
storage.py- workspace bootstrap
- SQLite initialization
Keep adapters isolated. The rest of the skill should consume normalized records rather than source-specific Outlook data structures.