soia-dev-agent-cli-dispatch

SUSPICIOUS. The skill's core behavior is aligned with its stated purpose as an external AI dispatcher, and it includes several safety-oriented instructions. However, it materially expands agent power by installing through a transitive skill mechanism, recommending wildcard permissions, routing code/context to third-party AI CLIs, and using approval/sandbox bypass flags. This looks coherent but high-risk, not overtly malicious.

No clear embedded malware indicators are present in this fragment (no persistence, credential theft, network exfiltration, or obfuscated payloads). However, the module is a high-risk execution orchestrator: it directly executes case-provided cmd_template strings via subprocess.run(..., shell=True) without validation, which can become arbitrary command execution if cases.json (or any resumed/derived inputs) are not fully trusted. Additionally, it modifies sys.path to import local modules, increasing impact if the local filesystem/package layout is compromised. Treat this code as dangerous to run with untrusted case definitions or in an untrusted working directory.