The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill promotes the use of the --madmax flag, which is explicitly documented to map to the underlying --dangerously-bypass-approvals-and-sandbox flag in the Codex CLI. This instruction encourages the agent to operate without standard safety constraints and user approval workflows, which can lead to unauthorized or harmful system operations.
[DYNAMIC_EXECUTION]: The orchestration layer supports a hook system that executes local JavaScript plugins located in the .omx/hooks/*.mjs directory. These scripts are triggered by native lifecycle events (e.g., session-start, turn-complete), allowing for the execution of arbitrary code based on files present in the local environment.
[EXTERNAL_DOWNLOADS]: The installation process involves downloading and globally installing the third-party oh-my-codex package from the npm registry, alongside the well-known @openai/codex CLI. These packages execute with significant privileges on the user's system.
[INDIRECT_PROMPT_INJECTION]: The skill is configured to automatically inject a project-level AGENTS.md file into the model's instructions, creating a vulnerability surface for untrusted data.
Ingestion points: Reads and injects <cwd>/AGENTS.md from the local file system into the active instruction set.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the injected file.
Capability inventory: The skill has access to powerful tools including Bash, Write, Read, Grep, and Glob across multiple agents.
Sanitization: No evidence of content validation or sanitization is present before the file content is interpolated into the prompt.
[COMMAND_EXECUTION]: The skill requires the execution of multiple environment setup and diagnostic commands (omx setup, omx doctor, omx team) that interact with the local shell and manage persistent tmux workers and MCP servers.

omx