The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it processes untrusted messages from external Lark/Feishu group chats.
Ingestion points: Group messages are fetched via poll.js using lark-cli and emitted to the agent context.
Boundary markers: The skill uses an LLM-based intent recognition layer to map messages to specific keys. It also includes explicit logic to reject bot 'start/stop' commands if they originate from the group (FLAGS=bot_switch_from_group).
Capability inventory: The skill can execute arbitrary shell commands defined in project profiles, perform file operations, manage system processes, and spawn background sub-agents.
Sanitization: Includes an image size validation step (check-image-size.js) before reading images to prevent API-level errors.
[COMMAND_EXECUTION]: The skill makes extensive use of system commands for its core functionality.
Execution list: Uses lark-cli for communication, git for version control, and Node.js for various utility scripts (permission.js, intent.js, dispatch.js).
Process Management: Employs kill, taskkill, and pgrep to manage the lifecycle of the polling process and singleton locks.
[EXTERNAL_DOWNLOADS]: The skill communicates with the Lark/Feishu API to fetch message history and download user-uploaded resources.
Evidence: Uses lark-cli im +messages-resources-download to download files and images from the chat to a local temporary directory (.cc-bot/bot_temp/).

lark-bot