The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell script templates (notably in references/arkcli-gen.md) that utilize subshell execution and variable interpolation, such as VER=$(arkcli models get <name> ... | tr -d '"'). This pattern is vulnerable to command injection if the agent interpolates untrusted user input into the shell command without proper escaping, potentially allowing arbitrary code execution on the local system.
[DATA_EXFILTRATION]: The skill uses the arkcli tool with the --input @<path> argument to read local files and transmit them to a remote service for processing. While intended for providing reference material (e.g., images for Image-to-Video tasks), this capability can be abused to exfiltrate sensitive local files (such as SSH keys or credentials) if the agent is manipulated into reading arbitrary paths.
[PROMPT_INJECTION]: The skill constitutes an indirect prompt injection surface (Category 8) as it processes user-provided prompts and reference files which are then passed as arguments to a command-line tool.
Ingestion points: User-supplied text prompts and local reference files ingested via the --input flag.
Boundary markers: None. The skill does not instruct the agent to use delimiters or ignore embedded instructions within the user input.
Capability inventory: The skill executes a local binary (arkcli) with network capabilities, reads local files, and writes files to the disk (--save-to). It also uses shell utilities like tr, jq, and base66 for data processing.
Sanitization: No validation or sanitization logic is provided to ensure that user-provided strings or file contents do not contain malicious payloads intended to influence the CLI tool's behavior.
[EXTERNAL_DOWNLOADS]: The skill uses the vendor's CLI tool to download generated media assets from the 'volcengine' Ark platform to the local filesystem. These downloads are part of the skill's primary purpose and originate from the author's infrastructure.

arkcli-gen