subtask4llm-codex
Audited by Socket on Oct 28, 2026
2 alerts found:
SecurityAnomalySUSPICIOUS: the skill’s behavior matches its stated purpose, but it deliberately grants unsandboxed autonomous execution to child Codex workers and starts them without further approval. No clear credential theft or malicious exfiltration is shown, yet the authority and data exposure are high enough to make this a high-risk workflow skill rather than a benign helper.
This fragment is a workflow orchestrator, not evidence of malware: it reads a plan and delegates tasks to Codex, while logging progress locally. Its main security concern is launching Codex with --yolo using task content from the plan, which can permit commands and file modifications in the workspace without interactive approval; use only with trusted plans and an appropriately isolated workspace. No suspicious domains, URLs, IP addresses, or targeted credentials are present. The excerpt is incomplete or syntactically invalid, limiting confidence about the full execution path.