The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions guide the agent to execute pre-compiled binaries bundled in the bin/ directory (e.g., codex-image2-windows-amd64.exe). While source code is provided in the src/ directory, the instructions prioritize using these opaque executables, which is a vector for running unverified code.
[COMMAND_EXECUTION]: The primary function of the skill involves the agent executing the codex-image2 CLI tool with parameters to generate or edit images, which involves direct shell interaction.
[EXTERNAL_DOWNLOADS]: The CLI tool source code (src/image_gen.go) includes logic to download image data from arbitrary URLs returned by the API response (item.URL) using http.Client.Get.
[DATA_EXFILTRATION]: The tool transmits user prompts and local image files to a remote API endpoint. By default, this is https://apinebula.com, which is a third-party domain not associated with the author's stated infrastructure in the provided context.
[CREDENTIALS_UNSAFE]: The skill uses the CODEX_API_KEY environment variable for authentication. The CLI tool sends this key in the Authorization header to a configurable endpoint (CODEX_API_URL). If this URL is maliciously redirected or set to an untrusted server, the API key will be exposed.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions.
Ingestion points: The tool reads prompt text from local files via the --prompt-file flag and job configurations from JSONL files via the --input flag in batch mode.
Boundary markers: No explicit boundary markers or warnings to the agent to ignore embedded instructions are present in the processing logic for these files.
Capability inventory: The skill possesses network transmission (POST requests) and filesystem write capabilities.
Sanitization: There is no evidence of sanitization, escaping, or validation of the content read from external files before it is used in API requests or local operations.

codex-image2