The Agent Skills Directory

[COMMAND_EXECUTION]: The script evals/files/restage_product.py utilizes os.system() to invoke npx sharp-cli. The command string is constructed by interpolating file paths using f-strings without shell escaping or sanitization, which creates a risk of command injection if filenames are derived from untrusted sources.\n- [EXTERNAL_DOWNLOADS]: The evaluation script evals/files/restage_product.py calls npx, which dynamically fetches and executes the sharp-cli package from the official npm registry at runtime.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from markdown files and user-provided scene descriptions, which are interpolated into model prompts. While documentation discusses safety, the provided scripts lack explicit sanitization or strict boundary markers for these inputs.\n- [DYNAMIC_EXECUTION]: The use of os.system() to execute shell-constructed strings represents a dynamic execution pattern that increases the skill's attack surface.

generative-media