The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interfaces with the internal wecom-cli tool using structured JSON parameters. It explicitly prohibits the use of terminal or shell execution modes, which mitigates the risk of command injection attacks through the agent interface.
[DATA_EXPOSURE]: There are strong privacy controls regarding user identifiers. The skill mandates that internal userid strings are never displayed to the end-user, requiring the agent to always resolve and display human-readable names instead.
[PROMPT_INJECTION]: The skill uses clear, mandatory operational protocols (reference files) to constrain agent behavior. It includes a specific 'Disambiguation' process to ensure the agent confirms user intent before creating or searching for data, preventing unintended modifications.
[INDIRECT_PROMPT_INJECTION]: While the skill processes meeting transcriptions for summarization, it defines a structured workflow that limits the scope of processing to official tool outputs, reducing the likelihood of the agent following malicious instructions spoken during a meeting.

wecomcli-meeting