The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from the Lark platform, creating a surface for indirect prompt injection attacks. • Ingestion points: The lark-cli event +subscribe command in references/lark-event-subscribe.md outputs real-time event data, including user-generated chat messages. • Boundary markers: Absent. The example pipelines in references/lark-event-subscribe.md interpolate event content directly into prompts (e.g., claude -p "Reply concisely: $content") without delimiters or instructions to ignore embedded commands. • Capability inventory: The skill utilizes lark-cli which provides capabilities for network API calls and document modifications, posing a risk if triggered by malicious input. • Sanitization: Absent. The examples demonstrate parsing data with jq but do not perform any validation, filtering, or escaping of the raw message content before further processing.
[COMMAND_EXECUTION]: The documentation for the +subscribe command in references/lark-event-subscribe.md includes a --force flag that allows users to bypass a single-instance lock. This flag overrides a safety constraint intended to maintain data integrity and instance synchronization.

lark-event