The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses node:child_process spawn within scripts/lib/providers.mjs to execute the vendor's dlazy CLI tool. This is a legitimate primary function of the skill. The implementation avoids shell injection risks by passing arguments as an array rather than using a shell environment.
[EXTERNAL_DOWNLOADS]: The scripts/gen.mjs script uses fetch to download generated assets from official AI provider domains, such as dlazy.com and openai.com, to save them to the local output/ directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of image files and URLs provided by the user or agent.
Ingestion points: The --images parameter in scripts/gen.mjs and the req.images array in scripts/lib/providers.mjs serve as points where external paths or URLs enter the skill context.
Boundary markers: Prompts and image references are passed as distinct, separate arguments to API endpoints and CLI tools, maintaining a clear boundary between instructions and data.
Capability inventory: The skill possesses capabilities for file system access (readFile, writeFile), network operations (fetch), and subprocess management (spawn) across scripts/gen.mjs and scripts/lib/providers.mjs.
Sanitization: Subprocess execution is hardened against injection via the use of spawn with argument arrays. Image inputs are encoded to base64 or passed as URLs to established providers, limiting their execution potential within the host environment.
[DYNAMIC_EXECUTION]: The skill dynamically constructs and executes provider-specific commands in scripts/lib/providers.mjs based on environment variables and model selection. This logic is well-structured and follows expected vendor-provided SDK patterns.

image-fusion