huawei-cloud-skill-tester

SUSPICIOUS. The testing purpose justifies cloud credentials and live Huawei API calls, and official Huawei CLI/SDK prerequisites appear consistent. But the bundled quality-report SDK materially expands scope: it reads host credential/session sources and automatically reports execution data, with a guest fallback to a non-Huawei domain and optional TLS verification disablement. This is not fundamentally incompatible with telemetry, but it is disproportionate to a skill tester and creates meaningful credential/data-flow risk.

The fragment is a telemetry SDK with legitimate quality-reporting functionality, but it performs broad automatic collection of credentials, local agent usage data, user/session content, and diagnostic information. It transmits that data remotely, including by default to the suspicious skillsop.topxtopx.com guest endpoint when IAM credentials are unavailable. The guest endpoint is not restricted to an allowlist, TLS verification can be disabled, and a Node.js helper is used to obtain cloud tokens. These behaviors create a substantial supply-chain and privacy risk. The fragment does not show destructive malware or a reverse shell, but its automatic credential discovery and data exfiltration behavior warrants treating the package as high risk unless independently verified and its endpoints, collection scope, and helper script are trusted.

This is an automated cloud test executor with legitimate command-execution functionality, not clear malware. However, it executes test-case-controlled shell and Python code with inherited credentials and can invoke destructive cloud operations. The claimed per-write user confirmation is not implemented; user_confirmed is set true automatically. It should only process trusted, integrity-verified test definitions in a sandbox with least-privilege credentials, and should enforce real confirmation plus path containment before execution.

This is an end-to-end testing/orchestration script rather than evident malware. It contains high-impact dynamic execution: metadata-controlled shell and Python commands, dynamic exec() of a utility file, and credential inheritance by child processes. These behaviors may be legitimate for a test framework, but untrusted Phase 1 metadata or skill content could achieve arbitrary code execution and credential access. No direct exfiltration or sabotage is visible in this fragment. Multi-skill result reporting is also misleading because steps are marked pass without execution.

该脚本的主要用途是安装生命周期测试和完整性报告，未显示明确恶意行为。需要修复或限制未加引号的 SKILL_INSTALL_CMD 执行方式，并对 install_dir、目标路径和删除操作实施更严格的规范化及白名单校验。由于依赖的两个 source 文件未提供，无法对其行为作完整结论。