The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The documentation instructs the agent to bypass Codex's internal security boundaries by using the --sandbox danger-full-access flag in specific execution environments.
[DYNAMIC_EXECUTION]: The skill encourages the use of high-risk execution flags such as --yolo, which executes AI-generated code without a sandbox or human approvals, and --full-auto, which automatically applies file modifications to the workspace.
[DATA_EXFILTRATION]: The skill identifies the file paths for sensitive authentication and session tokens, specifically ~/.hermes/auth.json and ~/.codex/auth.json.
[COMMAND_EXECUTION]: The skill relies on the execution of complex shell command strings via the terminal tool to perform repository management and interact with the coding agent CLI.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by cloning external, untrusted code repositories and then applying an autonomous agent to that code without proper sanitization.
Ingestion points: External source code fetched via git clone from user-specified URLs.
Boundary markers: Absent; the instructions do not provide delimiters or guidance to ignore instructions embedded within the processed source code.
Capability inventory: Full command execution via terminal, file system write access via the coding agent, and network access through git and gh tools.
Sanitization: Absent; no filtering or validation is performed on the untrusted code before the agent processes it.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the @openai/codex CLI tool from the official NPM registry.

codex