vision-click

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The prompt contains explicit CLI flags (--dangerously-bypass-approvals-and-sandbox, --skip-git-repo-check) that instruct bypassing approvals and sandbox checks—security-overriding instructions unrelated to the stated purpose of locating/clicking UI elements, constituting a deceptive override of safety controls.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). High-risk: the skill explicitly instructs sending screenshots to a remote Codex CLI and uses flags that bypass approvals and the sandbox (--dangerously-bypass-approvals-and-sandbox, --skip-git-repo-check), which enables potential data exfiltration of on-screen secrets and increases the risk of remote code execution or other unsafe behavior; the local CLI server API for automated clicks further expands the attack surface for abuse.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs using Codex CLI flags like "--dangerously-bypass-approvals-and-sandbox" (and other bypass flags), which directs the agent to circumvent local security/sandbox protections and thus compromises the machine's security posture.