The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes content from untrusted external sources, creating a surface for indirect prompt injection.\n- Ingestion points: Content is ingested from IM messages, emails, documents, calendar entries, and reports through the dws aisearch enterprise and dws aisearch behavior commands as defined in SKILL.md and references/aisearch.md.\n- Boundary markers: The instructions do not define clear delimiters or use specialized prompts to instruct the model to ignore potential instructions embedded within the retrieved search results or snippet previews.\n- Capability inventory: The agent can use data from initial search results to perform follow-up actions, such as fetching full documents via nodeId or accessing detailed user contact information via userId, as described in the multi-hop workflows in references/lite-recipes.md.\n- Sanitization: No sanitization, filtering, or escaping mechanisms are defined for handling the external content before it is processed by the AI model.

dingtalk-aisearch