The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflows for knowledge organization and thematic data collection ingest untrusted document content, creating a surface for indirect prompt injection attacks.
Ingestion points: Untrusted content enters the agent context through tools like docs +fetch and sheets +cells-get during the CONTENT_READ and CONTENT_VERIFY phases defined in the reference files.
Boundary markers: The instructions explicitly direct the agent to focus on organizational problems and relevance analysis (e.g., "Detect problems from organization perspective only") rather than following commands found within the documents.
Capability inventory: The skill possesses high-privilege capabilities including drive +move, drive +delete, and wiki +node-delete.
Sanitization: All destructive or high-risk operations are mitigated by a robust "Confirm" state in the internal logic, requiring explicit user approval of a generated plan before any write command is executed.

lark-drive