The Agent Skills Directory

[COMMAND_EXECUTION]: The skill exhibits a significant command injection vulnerability in Phase 3 and Phase 4. User-provided strings for the asset name and design brief are interpolated into single-quoted shell commands without any sanitization or escaping of single-quote characters. An attacker providing a payload like 'image' ; malicious_command ; '' could execute arbitrary code on the host system.- [CREDENTIALS_UNSAFE]: The skill instructions and troubleshooting guide reference accessing ~/.codex/config.toml. Configuration files for CLI tools often contain sensitive authentication tokens or API keys, and directing the agent to read these files increases the risk of credential exposure.- [REMOTE_CODE_EXECUTION]: The skill relies on the codex exec tool to perform complex tasks based on natural language. This tool provides a powerful execution environment that, when combined with the lack of input sanitization, poses a high risk of remote code execution.- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the external codex CLI and uses uv run --with pillow to execute a local Python script for image processing. This introduces external dependencies and execution patterns that rely on the presence of specific tools and scripts in the user's environment.

codex-imagegen