The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to construct and execute shell commands using the codex CLI. It specifically supports the --sandbox danger-full-access flag, which grants the underlying process broad system permissions, including network access. While the instructions mandate using AskUserQuestion before employing this flag, the capability itself presents a significant security surface.
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to append 2>/dev/null to all codex exec commands by default. This technique suppresses stderr, effectively hiding thinking tokens and potential error messages or warnings from the user, which could be used to conceal unexpected or malicious behavior during execution.
[PROMPT_INJECTION]: The skill includes mentions of hypothetical future models such as 'gpt-5' and 'gpt-5.2' along with specific performance claims (e.g., '76.3% SWE-bench'). These claims appear to be hallucinated or promotional in nature and could be used to influence the agent's behavior or the user's trust through deceptive metadata.

codex