The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the codex CLI binary using a bash wrapper script (scripts/codex-exec.sh). The script follows security best practices by using bash arrays for command execution to prevent shell injection and validating that output file paths reside within the /tmp directory.
[EXTERNAL_DOWNLOADS]: Documentation within the skill suggests installing the @openai/codex package from the NPM registry. This package originates from a well-known and trusted service provider.
[PROMPT_INJECTION]: The skill is designed to take user-provided text and pass it as an instruction to an external AI model via the Codex CLI. This architecture creates a surface for indirect prompt injection.
Ingestion points: User requests are captured via the SKILL.md routing logic and passed to scripts/codex-exec.sh as the PROMPT variable.
Boundary markers: The prompt construction does not employ delimiters or explicit instructions to ignore embedded commands within the user input.
Capability inventory: The underlying codex tool has capabilities to read files and, if explicitly requested by the user, perform write operations in the workspace.
Sanitization: The skill does not perform sanitization or filtering of the user input before passing it to the external tool.

Codex Bridge