The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted content by directing the agent to use the read_url tool for processing DingTalk document share links, as described in references/url-patterns.md. This content originates from external sources and could contain instructions designed to manipulate the agent's behavior during the execution of powerful CLI commands.
Ingestion points: Untrusted web content is retrieved via the read_url tool in references/url-patterns.md.
Boundary markers: While the "DWS Runtime Contract" in SKILL.md and references/runtime-contract.md requires intent verification for write operations, there are no instructions for using delimiters or explicit warnings to ignore instructions embedded within the fetched external content.
Capability inventory: The skill provides access to the dws CLI, which supports high-impact operations across multiple DingTalk products, including sending chat messages, modifying documents, deleting calendar events, and managing tasks.
Sanitization: The instructions do not specify any validation or sanitization procedures for the content retrieved from external URLs before it is processed by the agent.

dingtalk-shared