The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several local shell commands including mkdir, cp, cat, and the runcomfy and magick (ImageMagick) binaries. These are used to create project directories, generate pet metadata, and process image assets into the required spritesheet format.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the @runcomfy/cli tool and imagemagick. It also involves downloading user-supplied images from external HTTPS URLs to be processed by the image generation model.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages authentication via the RUNCOMFY_TOKEN environment variable or a local configuration file (~/.config/runcomfy/token.json). The instructions specify setting appropriate file permissions (mode 0600) to protect the token.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of a user-provided SOURCE_URL. While this represents an attack surface for image-based prompt injection into the underlying vision model, the skill acknowledges this risk and the output is limited to static image assets used by the Codex desktop application. 1. Ingestion points: SOURCE_URL parameter in the image generation step. 2. Boundary markers: The skill passes parameters via a JSON object to the CLI to avoid shell injection. 3. Capability inventory: Local file writes to ${CODEX_HOME}, image processing via ImageMagick, and API calls to RunComfy. 4. Sanitization: The runcomfy CLI treats inputs as JSON to prevent shell expansion.

codex-pet