The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Lark Wiki nodes (titles, descriptions, and content). This content could contain instructions designed to influence agent behavior during discovery or organization tasks.
Ingestion points: wiki +node-get, wiki +node-list, and various token/URL parsing steps across all reference files (e.g., SKILL.md).
Boundary markers: The instructions require the agent to present candidates to the user and obtain explicit confirmation for high-risk operations like deletion via the --yes flag.
Capability inventory: Extensive use of lark-cli for node management, member management, and deletions across all referenced documentation files.
Sanitization: No explicit sanitization or escaping of external content is defined in the skill instructions; it relies on the internal parameter handling of the lark-cli binary.
[COMMAND_EXECUTION]: The skill performs various operations by executing lark-cli commands. These commands utilize arguments such as tokens and URLs provided by users or retrieved from the Lark API, which could potentially lead to command injection if the underlying CLI tool does not properly sanitize inputs.

lark-wiki