randroid

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This skill explicitly implements an unattended, autonomous loop that invokes external LLM CLIs (Codex/Claude), can include local state/directions in prompts sent to those services, and automates git/PR operations (push/open PR/auto-merge) — enabling data exfiltration and autonomous remote code changes that are high-risk for abuse or backdoor-like behavior.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The randroid:loop workflow ingests and iterates over outsider-controlled free text provided via user “directions” (injected into the loop prompt and persisted in state/loop.local.md) without first selecting a specific item, so the LLM can be exposed to indirect prompt injection through that text.