The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses child_process.spawn to orchestrate execution between its components, including running Node.js sub-scripts, a Python-based compliance checker, and a vendor-specific CLI tool (dlazy). While this is central to the pipeline's functionality, it involves executing external binaries based on configuration parameters.\n- [INDIRECT_PROMPT_INJECTION]: The scripts/batch.mjs script performs CSV parsing and interpolates field values into a prompt template using simple string replacement. \n
Ingestion points: The system reads product metadata from user-provided CSV files via the --input argument.\n
Boundary markers: There are no explicit delimiters or instructions provided to the model to ignore potentially malicious instructions embedded within the CSV data.\n
Capability inventory: The skill has the ability to perform network requests to multiple AI providers, write files to the local system, and execute subprocesses.\n
Sanitization: The skill does not apply any sanitization, escaping, or validation to the CSV data before it is interpolated into the prompts for image generation or quality control agents.\n- [EXTERNAL_DOWNLOADS]: The pipeline connects to several third-party AI services to perform generation and analysis, including OpenAI, Google Gemini, Fal.ai, and Replicate. It also fetches image assets from remote URLs when provided in the input manifest.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates command-line arguments and prompts at runtime. The run_loop.mjs script implements an automated feedback loop that programmatically alters prompts based on the output of a vision-language model, creating a dynamic execution path where model outputs directly influence subsequent system actions.

batch-image