The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly recommends the use of the --madmax flag, which is documented to map to the --dangerously-bypass-approvals-and-sandbox setting. This configuration bypasses primary security guardrails and sandboxing, enabling the agent to execute actions without user approval or safety constraints.
[EXTERNAL_DOWNLOADS]: The skill instructions require the installation of the oh-my-codex package from the public npm registry. This introduces an external third-party dependency into the execution environment.
[REMOTE_CODE_EXECUTION]: The skill provides a 'Hook Extensions' feature that, when enabled via OMX_HOOK_PLUGINS=1, automatically executes JavaScript files (.mjs) from a local .omx/hooks/ directory. This creates a mechanism for arbitrary code execution based on the contents of the local project files.
[PROMPT_INJECTION]: The skill implements an automated injection of the AGENTS.md file from the current working directory into the model's system instructions. This represents a surface for indirect prompt injection.
Ingestion points: Project-level AGENTS.md file in the current working directory.
Boundary markers: None identified; the file content is directly used as a configuration parameter for model instructions.
Capability inventory: Includes shell access via Bash, file system operations via Write, and coordination of multiple autonomous agents.
Sanitization: None; the content of the local file is ingested and treated as authoritative instructions for the agent.

omx