opencode

SUSPICIOUS. The skill's high-level purpose is coherent, but it normalizes auto-approved delegated execution through an external CLI and may send project code to whichever providers OpenCode is configured to use, including custom base URLs. The CLI itself appears legitimate, so this is not confirmed malware, but the bypass of permission controls plus ambiguous downstream data routing makes the skill higher risk than a normal documentation/integration skill.

The fragment is a model-evaluation harness, not apparent malware. The principal security concerns are execution of an external agent with --dangerously-skip-permissions, inheritance of potentially sensitive environment variables, and unnecessary Windows shell=True combined with environment-controlled executable selection. Prompts do not provide a strong security boundary. The file also contains an unterminated string literal that prevents execution as written. No direct credential theft, exfiltration, persistence, destructive behavior, or obfuscated payload is visible.

The fragment appears to be documentation for an OpenCode-based automation workflow. It contains potentially dangerous operational guidance because it recommends --dangerously-skip-permissions and broadly permissive configuration, which can enable unrestricted agent actions when executed. No direct malware, credential theft, suspicious network exfiltration, obfuscation, or destructive behavior is visible in this portion.