The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted image data from local paths or remote URLs and processes it through external AI models. While the primary function is binary image manipulation, the skill lacks explicit sanitization or boundary markers for the data it processes, creating a theoretical surface for adversarial content to influence downstream agent tasks.
Ingestion points: The --image parameter in scripts/run.py accepts user-supplied paths and URLs.
Boundary markers: None identified; the skill processes raw binary data.
Capability inventory: The skill has file read/write access and makes network requests to the Replicate API.
Sanitization: No sanitization is performed on input URLs or the binary content before it is passed to the remote provider.
[DYNAMIC_EXECUTION]: The entry point script scripts/run.py uses os.execv() to perform a process replacement. It detects a local virtual environment (.runners-venv) and re-executes the script using the environment's Python interpreter. While this is a common pattern for managing skill dependencies, it involves runtime path resolution and the execution of a new process with user-supplied arguments (sys.argv).
[COMMAND_EXECUTION]: The skill requires Bash and is designed to be invoked via shell commands. The documentation also explicitly encourages the use of shell loops and pipes (e.g., find | xargs) for batch processing, which increases the potential impact of command injection if image filenames are not properly handled by the underlying system.

bg-remover