Installs into .claude/skills of the current project.
Are you the author of Claude Mail Bridge Team Rules?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/codingjay-1-claude-mail-bridge-team-rules)
---
name: claude-mail-bridge-team_rules
description: Use when operations involve emailing teammates (mail_teammates MCP tool). Defines strict privacy rules for team communication.
---
# Claude Mail Bridge Team Rules
## Overview
This skill defines strict behavioral guidelines for operations involving emailing teammates. These operations involve privacy risks and require explicit user approval.
## When to Use This Skill
Use when operations involve emailing teammates.
## Core Rule: Explicit Approval Required
**CRITICAL**: Before calling `mail_teammates`, you MUST get explicit user approval.
### Approval Process
1. **Show the draft**: Explain to the user in natural language:
- The recipient (email or alias)
- The subject
- Content summary
- **Attachments** (if any): list the files being attached
2. **Get confirmation**: Wait for user to explicitly approve (e.g., "yes", "send it", "go ahead", "好的", "发送")
3. **Call with approval flag**: Set `_user_check_and_approved=True` when calling the tool
### Example
```
AI: I need to ask Alice about the deployment status.
Should I send her an email with:
- Subject: "Deployment status inquiry"
- Attachments: ["/var/log/deploy.log", "config.yml"]
Here's what I plan to say: [summary of content]
User: Yes, send it
AI: [calls mail_teammates with _user_check_and_approved=True and attachments=[...]]
```
### Attachment Rules
When including attachments in teammate emails:
- **Always mention attachments** when showing the draft to user
- **Verify file paths exist** before calling the tool (validation is automatic, but be aware)
- Max 15MB per file
- Allowed types: text/code, images/PDFs, archives, documents
## Prohibited Behaviors
**Absolutely prohibited:**
- Do NOT email teammates without getting explicit user approval first
- Do NOT assume approval based on past interactions
- Do NOT send sensitive information without explicit confirmation
- Do NOT call `mail_teammates` without first invoking this skill