The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface where the agent processes external, untrusted data sourced from Yandex Market, such as buyer reviews and seller-authored product titles.
Ingestion points: Untrusted data enters the agent context through the outputs of the yandex_search and yandex_card tools.
Boundary markers: The skill includes an explicit 'Trust boundary' section that instructs the agent to treat seller/buyer-authored text as data and not as policy, effectively warning it to ignore instructions embedded in reviews.
Capability inventory: The skill facilitates price analysis, rating distribution checks, and review summarization based on these external inputs.
Sanitization: The skill relies on natural language instructions for the agent to maintain a trust boundary rather than programmatic sanitization or delimiters.

yandex-connector