The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses well-known Content Delivery Networks (CDNs) like jsdelivr.net to load industry-standard visualization libraries (e.g., ECharts, Chart.js). These are used for rendering charts within the agent's output environment and originate from established, trusted sources.
[COMMAND_EXECUTION]: The skill includes a developer-oriented Python script, scripts/validate_skill.py, which is designed for local validation of the skill's internal structure, file consistency, and adherence to specific formatting rules. The script performs static analysis of the skill's own files and does not engage in dangerous operations such as network access or sensitive file modification.
[DATA_EXFILTRATION]: The instructions contain explicit prohibitions against including sensitive information like tokens, cookies, or private keys in the output. It mandates the use of verified data sources and distinguishes between user-provided, verified, and example data.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted user input (images and structured data). It incorporates security best practices by instructing the agent to treat user-provided images as immutable evidence and provides clear boundary markers to separate user context from the agent's logic, effectively mitigating the risk of indirect prompt injection.

doubao-visualization