Do NOT open a public issue.

Email: security via GitHub private advisory, or contact maintainer via Telegram @llllxyz with subject "AgentMesh Security".

We will acknowledge within 48h and provide timeline.

AgentMesh threat model includes:

• malicious agent, compromised agent, malicious Agent Card
• spoofed identity, replay, task hijacking, confused deputy
• SSRF, webhook abuse, authorization bypass, tenant breakout
• artifact access violations, message injection, oversized payloads
• DoS, retry storms, credential leakage, malicious extension, compromised gateway node

All controls must be implemented and tested.

• No blind trust of remote Agent Cards
• SSRF protection for webhooks and discovery
• Signed webhooks with replay protection
• Tenant isolation via RLS + middleware
• Rate limiting per org/project/agent/skill/IP
• Artifact access control + presigned URLs
• No long-lived credential storage where possible