The Agent Skills Directory

[PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection through data ingestion and processing.
Ingestion points: The creative design workflow in creative-design.md parses external CSV, JSON, XLSX, Zip files, and document formats like Word and PDF.
Boundary markers: No specific delimiters or instructions to ignore natural language commands within the data are required when interpolating content into the agent context.
Capability inventory: The agent has capabilities to execute shell commands via lark-cli and git, perform file system operations, and manage network requests through the CLI.
Sanitization: The prompts do not define specific sanitization or filtering logic for data extracted from these external sources.
[EXTERNAL_DOWNLOADS]: The skill fetches several external JavaScript libraries for its design components.
Sources: Downloads React, ReactDOM, Babel, and ECharts from sf3-scmcdn-cn.feishucdn.com.
Context: These are well-known libraries hosted on the vendor's official content delivery network and are used within the HTML applications generated by the skill.
[COMMAND_EXECUTION]: The skill executes various commands using a specialized platform tool.
Evidence: Uses lark-cli apps command family for database operations, application releases, and configuration management.
Context: These commands are the intended mechanism for interacting with the platform's API and managing the developer's cloud resources.

lark-apps