The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various lark-cli commands for meeting operations. This includes lark-cli vc +search for historical data and lark-cli vc +notes for content retrieval. These commands are part of the intended vendor functionality.
[EXTERNAL_DOWNLOADS]: The skill performs downloads of meeting artifacts (e.g., transcripts and whiteboard cover images) from the Lark platform to the local file system. Files are stored in organized subdirectories under ./minutes/. All downloads target the vendor's own infrastructure.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It provides instructions to fetch external markdown content from meeting notes (lark-cli docs +fetch) and parse it for specific tokens (e.g., <whiteboard token="xxx"/>). These tokens are then interpolated directly into CLI commands (lark-cli docs +media-download --token <whiteboard_token>). This pattern could be exploited if malicious content is placed in the meeting notes to perform argument injection.
Ingestion points: Meeting notes, transcripts, and AI summaries retrieved via the Lark API (files: SKILL.md, references/lark-vc-notes.md).
Boundary markers: No explicit markers or delimiters are defined to isolate untrusted content during parsing.
Capability inventory: The agent has the ability to write to the local file system and execute CLI commands with network access.
Sanitization: The instructions do not specify any validation or escaping for the tokens extracted from external content before they are used in shell commands.

lark-vc