The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from meeting minutes and AI-generated summaries which could potentially contain malicious instructions.
Ingestion points: Content is retrieved from external documents using lark-cli docs +fetch and lark-cli minutes +get in Steps 3A and 3B.
Boundary markers: The skill explicitly mandates a human-confirmation step (Step 5: "MUST展示行动项清单，等待用户确认") which acts as a critical security boundary before any write operations.
Capability inventory: The agent has the ability to create tasks via lark-cli task +create and send IM messages via lark-cli im +messages-send.
Sanitization: The skill uses structured extraction rules (Step 4) to transform unstructured text into specific fields (who, what, when), reducing the likelihood of raw instruction execution.
[EXTERNAL_DOWNLOADS]: The skill requires the lark-cli binary and references multiple configuration files from LarkSuite's official GitHub repository. These are verified components of a well-known technology service.
[COMMAND_EXECUTION]: The skill utilizes the lark-cli tool to perform authenticated operations within the Lark/Feishu ecosystem, including searching meetings, accessing drive files, and managing tasks. This behavior is consistent with the skill's stated purpose.

lark-workflow-action-extractor