battle

The file itself is a coordinating red-team agent that delegates high-risk operations to external scripts (HACK_SKILL/run.sh) and locally imported modules (hack.cascade_integration). The source code contains no explicit hardcoded credentials or obvious malicious payloads, but it creates clear execution and supply-chain sinks: arbitrary script execution via subprocess.run and dynamic local imports. If the external script/module are compromised, this agent will execute arbitrary code and persist any outputs. In untrusted environments treat HACK_SKILL and local 'hack' modules as untrusted and audit them before use. Disable 'chaos' exploit mode in sensitive environments and add output redaction, integrity checks (signatures/hashes) for external artifacts, and stricter access controls for memory persistence.

SUSPICIOUS. The skill is internally consistent as a security competition orchestrator, but it gives an AI agent high-risk offensive security and autonomous execution capabilities over code, containers, and firmware. No clear credential theft, exfiltration, or deceptive third-party routing is shown in this snippet, so it is not confirmed malicious; the main concern is dangerous capability and broad action scope inherent to the skill.