alibabacloud-opc-advisor
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user descriptions to drive its cloud resource sizing and SKU selection logic, creating a surface for indirect prompt injection.
- Ingestion points: The primary chat prompt where users describe their business goals and current project status (e.g., project type, concurrent user estimates).
- Boundary markers: The skill enforces a strict output protocol, requiring the diagnosis to start with the 【芁斉】 marker and forbidding any reasoning preamble, which helps isolate user-influenced logic from the final response.
- Capability inventory: The skill's active capabilities are restricted to read-only Alibaba Cloud CLI invocations (
DescribePrice,DescribeImages) via a sandboxed MCP tool and conversational dialogue. - Sanitization: While the skill uses strict internal state-machine logic to process inputs, it does not perform explicit sanitization of the user-provided text before internal processing.
- [SAFE]: The skill adheres to a strict read-only execution model. It uses the
AlibabaCloud___CallCLIMCP tool exclusively for metadata and pricing queries, with no capability for resource modification or deletion. - [SAFE]: The skill implements a robust "least privilege" security posture by defining and referencing a minimum required RAM policy, ensuring that its CLI calls are restricted to the necessary read-only permissions.
- [SAFE]: All referenced external URLs and product ecosystems are official Alibaba Cloud services or legitimate companion tools (such as QoderWork) mentioned within the context of the user's workflow, with no evidence of typosquatting or malicious redirection.
Audit Metadata