The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from calendar event descriptions and documents retrieved via lark-cli calendar +get and lark-cli docs +fetch. This content is provided by users and could contain instructions designed to influence the agent's behavior.
Ingestion points: Untrusted data enters the context through event descriptions (via +get and +agenda) and document content (via docs +fetch).
Boundary markers: The instructions do not define specific delimiters or instructions to separate untrusted data from system prompts.
Capability inventory: The agent has capabilities to modify calendar events, transfer ownership, and interact with IM/VC modules using lark-cli.
Sanitization: There is no description of content sanitization or validation before the agent processes the retrieved text.
[COMMAND_EXECUTION]: The skill uses the lark-cli binary to perform all calendar management and coordination tasks, which is the intended functional design.
[DATA_EXFILTRATION]: The skill facilitates the upload of local images to the vendor's cloud for use in event descriptions. This feature includes a security control that explicitly restricts file access to relative paths within the current working directory, preventing the reading of absolute system paths or files in parent directories.
[SAFE]: The skill implements a 'Destructive Confirmation Gate' in references/lark-calendar-recurring.md. This requires the agent to obtain explicit user confirmation before deleting or performing major updates to recurring event sequences, effectively mitigating the risk of accidental or malicious data loss.

lark-calendar