The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several shell commands to facilitate its workflow, including codex exec for image generation, git rev-parse for identifying project directories, and sips for image post-processing on macOS. These are functional requirements for the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface by ingesting and processing session logs that contain outputs from a language model.
Ingestion points: The script scripts/extract-codex-image.py reads JSONL files located within the ~/.codex/sessions/ directory.
Boundary markers: No specific delimiters are used to separate trusted from untrusted content within the logs, though the script uses string length as a heuristic for identifying image data.
Capability inventory: The skill possesses the ability to execute shell commands and write files to the .tmp directory within the git repository.
Sanitization: The extraction process relies on base64.b64decode. It does not perform deep validation of the resulting file content beyond checking if it is a valid PNG using the shell file utility in the provided instructions.

codex-image-gen