The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The primary function of the skill is to execute arbitrary code generated by an external LLM (OpenAI Codex) directly on the host machine via the codex exec command.
[COMMAND_EXECUTION]: The skill defaults to using the --dangerously-bypass-approvals-and-sandbox flag. As noted in the documentation, this grants Codex full filesystem and network access and suppresses all approval prompts, allowing it to install packages and modify any file on the disk without user intervention.
[DATA_EXFILTRATION]: By enabling unrestricted network access and full filesystem reads by default, the skill provides a direct path for sensitive data (such as credentials or private source code) to be transmitted to external servers if the generated code is malicious.
[COMMAND_EXECUTION]: The skill explicitly configures local proxy settings (127.0.0.1:7890) for all network traffic during execution. While often used for connectivity in specific environments, hardcoding these into a skill can be used to route traffic through an intercepting proxy.
[EXTERNAL_DOWNLOADS]: The skill requires the global installation of the @openai/codex package from the npm registry, which is the core dependency for these high-risk operations.

codex-task