Context-Level Secret Isolation for AI Coding Agents with Agentmask

Once a secret enters the agent's context window, the attack surface is wide open: Prompt injection via external tools - a compromised MCP server, a malicious package README, or a crafted API response