The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local helper script scripts/dt_helper.sh to manage configurations and fetch tokens. It also generates and executes temporary shell scripts in /tmp to perform complex API operations.
[DYNAMIC_EXECUTION]: The instructions direct the agent to write multi-line logic to /tmp/<task>.sh and then execute it, allowing for dynamic assembly of API requests and error handling at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to be sent as message content through DingTalk APIs. There is a potential risk of parameter injection if the agent does not properly escape the content when building the JSON payload for the generated shell scripts.
Ingestion points: User-supplied message content and configuration parameters managed in SKILL.md.
Boundary markers: The instructions explicitly warn the agent not to invent content and to ask the user if details are missing, providing a level of human-in-the-loop validation.
Capability inventory: The skill has the capability to write to the local filesystem (/tmp and ~/.dingtalk-skills/config) and make network requests via curl to DingTalk's official API domains.
Sanitization: The skill lacks explicit sanitization routines for message content, relying on the agent's internal logic to correctly escape characters for JSON payloads.

dingtalk-message