The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent and user to download and install a Claude plugin and clone a repository from an external, non-trusted GitHub account (github.com/lemomo-ai/lemo-opuscar).
[REMOTE_CODE_EXECUTION]: The provided instructions use shell commands (claude plugin install lemo-opuscar@lemolab) to fetch and execute third-party code within the agent's environment, which constitutes a remote code execution risk when the source is not a verified or trusted vendor.
[COMMAND_EXECUTION]: The documentation includes shell commands for plugin management (marketplace add, plugin install) and suggests cloning repositories, which can lead to arbitrary code execution if the upstream content is compromised.
[DYNAMIC_EXECUTION]: The skill's primary workflow involves generating and executing Canvas and WebGL code at runtime to render video frames. This dynamic code generation and execution is a powerful capability that could be abused if malicious logic is injected.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input (themes, stories, and styles) to drive the automated video 'Director' and code generation logic.
Ingestion points: User-provided themes and story descriptions in the SKILL.md workflow.
Boundary markers: None. There are no delimiters or warnings to the agent to ignore instructions embedded within the user's story content.
Capability inventory: The agent has the ability to write and execute code, interact with the file system to save MP4 files, and invoke plugin tools.
Sanitization: No sanitization or validation of the user-provided narrative content is specified before it is used to influence the code generation process.

lemo-opuscar