provider-switch

SUSPICIOUS: the skill's core purpose is coherent and most data flows go to official vendor endpoints, but the bundled wrappers intentionally enable bypass/YOLO modes that let the agent act without normal approval prompts. There is no strong evidence of credential theft or covert exfiltration, yet the autonomous execution footprint is high enough to make the skill risky.

The fragment is documentation only and does not establish malicious behavior. The documented bypassPermissions launch mode creates a significant operational safety risk if used in an untrusted workspace; the described provider routing also sends session data to DeepSeek. Review the actual installer and wrapper before installation because they were not supplied.

This fragment is documentation rather than the package implementation, so installer and launcher behavior cannot be verified. It does not show evidence of malware. The documented launcher always enables permission bypass, creating a significant safety risk if used in an untrusted workspace; provider requests may also transmit prompts and tool output to StepFun.

No malware is evident in the supplied documentation. The documented default permission bypass creates substantial operational risk if used in an untrusted workspace. The installer and wrapper source code were not supplied, so their behavior and credential protections cannot be independently assessed.

No evidence of malware or data exfiltration appears in this script. It is a launcher whose use of --dangerously-bypass-approvals-and-sandbox materially weakens protections for the Codex invocation; use only when that behavior is intended and trusted.

No direct malware behavior is evident in this fragment. The primary security concern is that running the wrapper launches Codex with approval and sandbox protections disabled. Use only when this behavior is intentional and the invoked instructions and tools are trusted.