The Agent Skills Directory

[DYNAMIC_EXECUTION]: The helper script tools/pixellab.sh utilizes python3 -c to process JSON responses from the API. It constructs the Python script by interpolating shell variables such as $outdir, $prefix, and image dimensions directly into the source string. This design allows for potential arbitrary Python code execution if a user or agent provides an argument containing single quotes and embedded Python commands.- [COMMAND_EXECUTION]: The skill relies on system commands including curl, python3, sips, and identify to function. These tools handle data provided by the user, such as descriptions and file paths, which increases the attack surface for command injection if the agent processes unvalidated input from an untrusted source.- [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection as it ingests pixel art descriptions and action prompts. These strings are interpolated into JSON payloads for external API requests and used to determine local directory and file names. A malicious prompt could be used to manipulate the API call or the local file system structure.
Ingestion points: Action strings and descriptions passed as arguments to the pixellab.sh script in tools/pixellab.sh.
Boundary markers: None identified; input is directly interpolated into shell and Python contexts.
Capability inventory: Network access (curl), file system read/write, and execution of system utilities.
Sanitization: The script includes basic sanitization for default path slugs, but lacks escaping for custom user-supplied paths or description fields.- [EXTERNAL_DOWNLOADS]: The skill communicates with the PixelLab REST API (api.pixellab.ai) to generate and manipulate images. While this is the intended purpose of the skill, it involves sending project data and receiving assets from a remote third-party service.

pixellab-api