The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes natural language responses and tool results from the Alibaba Cloud DAS API. This creates a surface for indirect prompt injection if the database diagnostic data (such as SQL statements or performance metrics) were to contain malicious instructions.
Ingestion points: User input is accepted via the --question argument, and external data is received through the Server-Sent Events (SSE) stream from the DAS API in scripts/call_das_agent.py.
Boundary markers: The script implements clear boundary markers (e.g., '=== DAS AGENT RESPONSE ===') when running in --pipe mode to help the calling agent distinguish between diagnostic content and control noise.
Capability inventory: The skill is limited to performing network requests to the official das.cn-shanghai.aliyuncs.com endpoint.
Sanitization: The skill performs length and type validation on user-supplied questions and session IDs in the _validate_message and _validate_session_id methods.
[EXTERNAL_DOWNLOADS]: The skill depends on the official alibabacloud_credentials Python library and connects to the official Alibaba Cloud endpoint das.cn-shanghai.aliyuncs.com. These resources originate from the skill's authoring organization and are necessary for the intended functionality.
[CREDENTIALS_UNSAFE]: The skill correctly implements the Alibaba Cloud default credential chain via the official SDK. It avoids hardcoding secrets and instead resolves them from standard environment variables or local configuration files as per vendor best practices.

alibabacloud-das-agent