Privacy & Safety
Understanding how the Gmail Labeler skill handles your data and what safeguards are in place.
This Showcase Website
The Original Skill
What the Skill Reads
- Sender address and display name - To classify by provider
- Subject line - For content type classification (receipt vs newsletter)
- Message snippet and body text - For determining keep vs archive action
- Existing label IDs - To skip threads already correctly labeled
- Attachment filenames - For context only, contents never read
What the Skill Can Modify
- Add labels to threads - Applies provider labels based on classification
- Remove INBOX label (archive) - For routine mail that should be filed away
- Create new labels - Creates category and provider labels as needed
What the Skill Never Does
- Delete or trash mail - Only archives (removes from inbox view)
- Send, forward, or reply to mail - Read-only plus labeling
- Mark mail as read - Unread state is preserved
- Read attachment contents - Only sees filenames for context
- Change star state or importance - Your flags stay intact
Where Your Data Goes
Your Gmail Account
The skill accesses Gmail through your local MCP server using your OAuth credentials. No credentials are ever sent anywhere else.
Local MCP Server
Runs on your machine, handles Gmail API calls. Email metadata and content stay local until sent to the AI provider for classification.
Your AI Provider (Anthropic, OpenAI, etc.)
Email content is sent to your chosen AI provider for classification. Check their data retention policies. Anthropic does not train on Claude API data; OpenAI may train unless opted out.
Local Files (MEMORY.md, LOG.md)
The skill writes session state and action logs to local markdown files. These never leave your machine unless you commit them to version control.
Safety Features
Dry-Run Mode
First-time runs default to dry-run. See exactly what would happen before making any changes.
Complete Audit Trail
Every action is logged to LOG.md with timestamps, thread IDs, and labels applied.
Rule-Satisfied Skip
Threads already correctly labeled are skipped entirely. No redundant processing or API calls.
Skip-by-Design Senders
OTP codes, security alerts, and personal mail are automatically skipped. No automated filing of sensitive messages.
Important Considerations
Email Content Goes to Your AI Provider
The skill sends email sender, subject, and body text to your AI provider (Anthropic, OpenAI, etc.) for classification. If you handle sensitive or confidential information, understand your provider's data retention policies before using this skill.
Use a Test Account First
Set up the skill on a throwaway Gmail account or a test inbox before running it on your main email. Verify the behavior matches your expectations.
No Warranty, Use at Your Own Risk
This is a personal project released under GPL-3.0 with no warranty. The author is not liable for data loss, mislabeled mail, or any other issues. Always review dry-run results before applying changes.