The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation recommends installing the videoclaw package using the uvx tool. This results in the agent fetching and installing code from a remote public registry without built-in verification of the package's integrity or source.
[REMOTE_CODE_EXECUTION]: By directing the agent to run uvx videoclaw, the skill enables the execution of remote code. If the videoclaw package were to be replaced by a malicious version in the public registry, it could lead to arbitrary code execution on the host system.
[COMMAND_EXECUTION]: The skill is designed to interact with the system shell by executing the videoclaw utility. It passes several arguments directly to the command line, including project names, image paths, and user-supplied prompts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text strings and interpolates them into the --prompt argument for the CLI tool. This creates a vulnerability where instructions hidden in the user input could potentially influence the tool's behavior.
Ingestion points: The --prompt (-t) parameter in SKILL.md.
Boundary markers: None; the input is passed directly as a string argument.
Capability inventory: Execution of external CLI tools (videoclaw) via the shell.
Sanitization: No sanitization or input validation is specified in the skill instructions.

video-i2v