The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill's behavior is entirely consistent with its stated purpose of managing Lark/Feishu IM communications. External resources and API calls are directed to the vendor's own infrastructure.
[COMMAND_EXECUTION]: The skill utilizes the lark-cli binary to execute tasks. In lark-im-messages-resources-download.md, the download functionality includes a specific restriction prohibiting directory traversal (..) in the output path, which is a key security control.
[PROMPT_INJECTION]: The skill has an inherent surface for indirect prompt injection since it reads and processes chat messages from external users.
Ingestion points: +chat-messages-list, +messages-search, and +threads-messages-list fetch untrusted content from chat conversations.
Boundary markers: Not present in the instruction text for data processing.
Capability inventory: The agent can send messages (+messages-send, +messages-reply), modify chat settings (+chat-update), and download files (+messages-resources-download) via subprocess calls to lark-cli.
Sanitization: The skill relies on internal normalization of content by the CLI tool.
Mitigation: The instructions explicitly mandate that the agent must obtain user approval before any write operations (sending messages or modifying groups), providing a robust human-in-the-loop defense against automated injection attacks.

lark-im