The Agent Skills Directory

[DATA_EXFILTRATION]: The vcr_image.py script enables the transmission of local file contents to a remote API (https://ai.vanscoding.com/v1/images) through the --reference-path parameter. Although it checks for image-related file extensions, the script does not validate that the requested paths are contained within the project directory. Consequently, any file reachable by the execution environment could be read and exfiltrated if its path is provided to the tool.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection attacks. Because the agent is instructed to use files from the workspace as references, an attacker could provide data that tricks the agent into referencing sensitive files outside the workspace (e.g., SSH keys or system configs). The skill lacks the necessary boundary markers or sanitization logic to prevent the agent from obeying such malicious instructions.
Ingestion points: Reference path parameters defined in SKILL.md and processed by the agent.
Boundary markers: Not present; the skill lacks delimiters or warnings to ignore instructions embedded in the data it processes.
Capability inventory: The vcr_image.py script has the capability to read files from disk and send them over the network via POST requests.
Sanitization: Insufficient; the script resolves paths but fails to enforce a root directory constraint for reference inputs.
[COMMAND_EXECUTION]: The skill executes a Python script via a PowerShell wrapper. It uses safe argument passing techniques (argument arrays) to prevent shell injection, though it does involve the execution of a secondary process on the host machine.

vcr-imagegen