The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes product-level data (titles, descriptions, brand names) from the Ozon marketplace, which is an external and attacker-controllable source. Maliciously crafted product information could attempt to influence the agent's behavior.
Ingestion points: Data is fetched via API calls in scripts/seerfar_ozon_product_report_search.py.
Boundary markers: The SKILL.md file includes instructions to "Present data only" and "show product metrics in a clear table without subjective advice," which acts as a basic instruction boundary.
Capability inventory: The skill environment has network access via the urllib and requests modules and filesystem write capabilities within the project directory (linkfox/ subfolder).
Sanitization: There is no evidence of specific sanitization or escaping of the external data before it is interpolated into the agent's display context.
[PERSISTENCE]: The onboarding documentation encourages the user to manually modify their system shell configuration files to persist API keys.
Evidence: references/onboarding.md provides commands for users to append LINKFOX_AGENT_API_KEY to ~/.bashrc and ~/.zshrc. While these are manual instructions for the user rather than automated script actions, they fall under persistence management.
[EXTERNAL_DOWNLOADS]: The skill's onboarding script relies on third-party Python packages that may need to be installed by the user at runtime.
Evidence: scripts/onboarding.py references and attempts to import requests, qrcode, and pillow, providing installation instructions if the imports fail.

linkfox-seerfar-ozon-product-report-search