The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Automated reputation scans identified references/lark-vc-recording.md as malicious (FileRepMalware). While the file appears to be markdown documentation for the VC recording command, the automated detection represents a confirmed security alert that requires investigation.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates a significant attack surface for indirect prompt injection by processing external data with associated write capabilities.
Ingestion points: The skill ingests untrusted data from meeting transcripts, chat messages, and shared document content via vc +meeting-events (referenced in references/lark-vc-meeting-events.md), minutes +detail (referenced in references/lark-minutes-detail.md), and docs +fetch.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' warnings for the processed meeting content.
Capability inventory: The agent has write access to the workspace, including the ability to send meeting messages (vc +meeting-message-send), update meeting titles (minutes +update), and modify drive resource permissions (drive +member-add).
Sanitization: No sanitization or validation of external text is defined before it is analyzed or used to generate outputs.
[DATA_EXFILTRATION]: The skill accesses sensitive organizational data, including meeting transcripts, audio/video media, and AI-generated summaries. While it performs local downloads, the processing of this content makes it a target for data exposure via prompt manipulation.
[PRIVILEGE_ESCALATION]: The skill provides automated routines for managing permissions, such as adding members to drive resources (drive +member-add) and applying for resource access (minutes +apply-permission). These capabilities could be exploited to escalate privileges or grant unauthorized access if the agent's logic is subverted by malicious input.

lark-meeting