The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it ingests and processes external data.
Ingestion points: The skill ingests untrusted data from external API responses in scripts/maidalv_check_api_flash.py and local file content in scripts/upload_image.py.
Boundary markers: The instructions in SKILL.md advise the agent to "validate returned fields and preserve the distinction between source facts, calculations, and recommendations," which acts as a partial boundary but does not fully mitigate adversarial input in data fields.
Capability inventory: The skill has the capability to perform network requests (urllib.request), read local files (open), and write to the local filesystem (open for caching and session data) across both provided scripts.
Sanitization: While the scripts use json.loads to parse inputs, there is no explicit sanitization or strict schema validation for the external API payloads before they are summarized for the user.
[COMMAND_EXECUTION]: The skill is designed to execute specific local Python scripts (scripts/maidalv_check_api_flash.py and scripts/upload_image.py) to perform its core functions. These scripts are invoked with arguments derived from user input or agent-derived parameters.
[DATA_EXFILTRATION]: The skill transmits data, including potential product images and the NEXSCOPE_API_KEY, to the configured NEXSCOPE_PROXY_BASE. This behavior is documented as the primary function of the skill for communicating with the vendor's services.

ecommerce-maidalv-product-tro-detection