Indirect Prompt Injection via Repository Setup Scripts Targeting AI Agents

Ibrahim Saud@tektrix

•updated Jun 29, 2026•0•0•3

Detects suspicious command execution patterns, including reverse shell techniques (/dev/tcp, bash -i), DNS-based C2 communication (axiom-config, m100.cloud), and fileless payload retrieval (dig TXT queries). It specifically highlights commands initiated by common AI developer tools or runtimes (claude, node, python) that correlate with known adversarial setup scripts or potential prompt injection outcomes.