Agent polling is a communication model used by build systems and other distributed platforms in which agents regularly contact a central server to ask for jobs, instructions, or status updates. Instead of the server pushing tasks directly to each agent, the agent initiates the connection on a schedule or in a loop. This pattern is common in CI/CD tools because it works well through firewalls and NAT, and it keeps short-lived build workers easy to manage.
From a security perspective, polling creates a high-value control path. The server must trust messages from agents, so weaknesses in authentication, session handling, or data parsing can become entry points for remote code execution or command injection. In real attacks, defenders focus on the polling endpoint, the format of agent messages, and whether untrusted data is deserialized safely. Good defenses include mutual authentication, strict input validation, network restriction, patching, and running the server with least privilege so a flaw in the polling flow does not compromise the whole build environment.