The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external meeting participants (via chat and transcripts) and shared document content. This data is used to generate summaries and can trigger downstream actions like sending messages to other users.
Ingestion points: Meeting events (transcripts/chat) in references/lark-vc-meeting-events.md, minute transcripts in references/lark-minutes-detail.md, and shared documents in scenes/query-note-and-artifacts.md.
Boundary markers: The instructions lack explicit boundary markers or 'ignore embedded instructions' warnings for processing untrusted meeting content.
Capability inventory: The skill can send live-meeting messages (+meeting-message-send), send IM messages (im +messages-send), modify permissions (drive +member-add), and write to the local file system (+download, +transcript).
Sanitization: No explicit sanitization or validation of untrusted content is mentioned before it is processed for summaries or forwarded to other channels.
[COMMAND_EXECUTION]: The skill makes extensive use of the lark-cli tool to perform administrative and interactive tasks within a Lark workspace. This includes sensitive operations like managing meeting participants and handling authentication identities (--as user vs --as bot).
[EXTERNAL_DOWNLOADS]: The skill is designed to download meeting recordings and save transcripts to the local file system.
Evidence: Instructions in references/lark-minutes-download.md and references/lark-note-transcript.md direct the agent to save files into local directory structures such as ./minutes/ and ./notes/.

lark-meeting