Researchers warn Agentjacking can abuse Sentry errors to make AI coding agents run malicious code on developer machines.