alicloud-platform-multicloud-docs-api-benchmark
Pass
Audited by Gen Agent Trust Hub on Mar 11, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The script communicates with external APIs including Google Discovery API, GitHub Search API, and DuckDuckGo to identify documentation links. It also fetches content from documentation pages on official cloud domains. These interactions are limited to well-known services and trusted providers.
- [COMMAND_EXECUTION]: The workflow involves executing a local Python script with user-provided parameters. The script logic is restricted to data aggregation, scoring, and file I/O within a designated output directory, with no signs of arbitrary command injection.
- [CREDENTIALS_UNSAFE]: The documentation instructs users to configure Alibaba Cloud Access Keys. While the provided Python script does not utilize these environment variables, their presence in the environment is noted as a potential exposure point for other scripts.
- [PROMPT_INJECTION]: The skill processes untrusted content from external documentation sites to extract quality signals. Ingestion occurs via urllib in benchmark_multicloud_docs_api.py; boundary markers are absent; capabilities are limited to local file writes; and sanitization consists of basic HTML tag removal. The risk is minimized as the script only searches for specific keywords and does not pass full external text to the LLM.
- [SAFE]: The script uses domain-level allowlisting to restrict the URLs it fetches content from, effectively mitigating potential Server-Side Request Forgery (SSRF) risks.
Audit Metadata