If you discover a security vulnerability in this project, please report it using GitHub's private vulnerability reporting. Go to the Security tab of this repository and click Report a vulnerability.

Please include:

• A description of the vulnerability
• Steps to reproduce
• Potential impact
• Suggested fix (if any)

Note: This is a community reference architecture, not a Workday product. Security reports are handled by project maintainers on a best-effort basis.

This project handles sensitive data including:

• OAuth tokens — LINE WORKS JWT-based authentication
• API keys — LLM provider and MCP credentials
• User messages — Conversation content from enterprise messaging platforms

• Never commit .env files or API keys to version control (.gcloudignore excludes them from Cloud Build source archives)
• Use environment variables or secret managers for all credentials
• Deploy the bridge service behind HTTPS in production
• Restrict MCP server access to the bridge service (and authenticate production MCP endpoints)
• Regularly rotate API keys and tokens
• Use official Workday MCP servers (Agent Gateway) instead of the demo server in production
• Set LW_API_20_BOT_SECRET for LINE WORKS (callbacks are rejected without it); treat DingTalk's DINGTALK_ALLOWED_USERS filter as demo-only without additional ingress controls

For detailed technical hardening recommendations (rate limiting, PII redaction, retry logic, prompt injection defenses, observability, and more), see the Enterprise Hardening Guide.

For v0.2.0 channel-auth and in-memory state limits, see Changelog — Known limitations.