The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow files workflows/generate.lobster, workflows/batch-generate.lobster, and workflows/pip-generate.lobster contain steps that construct shell commands by directly interpolating user-controlled variables (such as ${script}, ${actor}, and ${scripts}) into the command string. This pattern is vulnerable to command injection, as a malicious user could provide a script containing shell metacharacters (e.g., ;, &&, |) to execute arbitrary commands on the host system.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the agent-media-cli package from the npm registry and provides instructions to install the skill via npx.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its data-processing instructions. Ingestion points: SKILL.md (Rule 3) instructs the agent to visit user-provided product URLs to extract image URLs from the website's content. Boundary markers: No delimiters or isolation instructions are provided to help the agent distinguish between website content and its own instructions. Capability inventory: The agent can execute shell commands via the agent-media CLI, which can result in financial costs via credit consumption and file system operations. Sanitization: No sanitization or filtering logic is specified for the content extracted from external URLs.

agent-media