The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on various shell-based tools, including lark-cli for interacting with the Lark platform and npx for running the @larksuite/whiteboard-cli utility package.
[REMOTE_CODE_EXECUTION]: To handle complex geometric calculations for diagrams like Fishbone, Funnel, and Flywheel charts, the skill instructs the agent to generate CommonJS (.cjs) scripts at runtime. These scripts are then executed using node to produce JSON data for the whiteboard (e.g., as described in routes/dsl.md and several files in scenes/). While the scripts are based on provided templates, the execution of AI-generated code on the local system represents a notable security surface.
[EXTERNAL_DOWNLOADS]: The skill's image preparation workflow (references/image.md) involves using curl to download images from external URLs or third-party image libraries. These images are then uploaded to Lark's media service.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection due to its read-process-write cycle.
Ingestion points: Data is ingested from external whiteboards via the lark-cli whiteboard +query command (references/lark-whiteboard-query.md).
Boundary markers: The instructions do not specify the use of delimiters or specific warnings to ignore instructions embedded within the whiteboard content during processing.
Capability inventory: The environment includes shell access (lark-cli, node, npx, curl), file system write access for output products, and network access via the Lark CLI and curl.
Sanitization: Validation steps like the --check flag for SVG rendering are used to ensure layout integrity, but no security-focused sanitization of external content is mentioned.

lark-whiteboard